Search Results

Search found 14213 results on 569 pages for 'distributed programming'.

Page 236/569 | < Previous Page | 232 233 234 235 236 237 238 239 240 241 242 243  | Next Page >

  • Would you consider using training/mentoring from LearnersParadise.com?

    - by HK1
    My initial question deserves some explanation. I signed up for an account at learnersparadise.com. After signing up I couldn't login so I opted to use their "send password" feature. Upon receiving my password in my email I confirmed two things A) They trimmed off 2 of the last digits of my 10-digit password without informing me and saved it that way in their database B) my password is not saved in their database using a one-way hash since they were able to email me my password. I'm quite certain that both of these are perfectly awful programming practices. I suspect that the mentors/trainers at learnersparadise are not necessarily affiliated with the website and it's design since they are basically people like you and me (hopefully more skill than me) who have signed up to become mentors. However, I'm still uncertain about signing up for training/mentoring at a site that uses such poor programming practices themselves? Would you let learnersparadise poor programming practices affect your opinion of their trainers/mentors?

    Read the article

  • What is it going here in my solution?

    - by bbb
    I am a asp.net mvc programmer and if I want to start a project I do this: I make a class library named Model for my models. I make a class library named Infrastructure.Repository for database processes I make a class library named Application for business logic layer And finally I make a MVC project for the UI. But now some things are confusing me. Am I using 3-tier programming? If yes so what is n-tier programming and which one is better? If no so what is 3-tier programming? Some where I see that the tiers namings are DAL and BIZ. Which one is correct according to the naming convention?

    Read the article

  • Cost of maintenance depending on paradigms

    - by Anto
    Is there any data on which paradigms allow for code which is easier/cheaper to maintain? Certainly, independantly of the chosen paradigm, good design is cheaper to maintain than bad, but there should probably be major differences coming only from the paradigm choice. Unstructured programming, for instance, generates very messy code (spaghetti code) which is expensive to maintain. In object oriented programming, implementation details are hidden and thus it should be pretty cheap to change those. In functional programming, there are no side effects, thus there is lesser risk of introducing bugs during maintainance, which should be cheaper. Is there any data on which paradigms are the most cost-efficient when coming down to maintenance? If no such data exists, what is your take on the question?

    Read the article

  • What are the pro/cons of Unity3D as a choice to make games?

    - by jokoon
    We are doing our school project with Unity3d, since they were using Shiva the previous year (which seems horrible to me), and I wanted to know your point of view for this tool. Pros: multi platform, I even heard Google is going to implement it in Chrome everything you need is here scripting languages makes it a good choice for people who are not programming gurus Cons: multiplayer ? proprietary, you are totally dependent of unity and its limit and can't extend it it's less "making a game from scratch" C++ would have been a cool thing I really think this kind of tool is interesting, but is it worth it to use at school for a project that involves more than 3 programming persons ? What do we really learn in term of programming from using this kind of tool (I'm ok with python and js, but I hate C#) ? We could have use Ogre instead, even if we were learning direct x starting january...

    Read the article

  • What is the worst programmer habit?

    - by 0x4a6f4672
    Many people get into programming because programming is fun. At least in the beginning. After some time doing it professionally, programming is no longer fun, often just hard work. Sometimes we develop bad habits along the way to make it fun again. Some bad habits of programmers are well known, for example the "I fix that in a second" habit, the "reinvent the wheel" practice or the "all code except mine is crap" attitude (which often leads to "I will re-write the entire program from scratch" syndrome). There are things which a programmer should never do. What is the worst programmer habit?

    Read the article

  • Career Shifters: How to compete with IT/ComSci graduates

    - by CareerShifter
    I am wondering what are the chances of a career shifter (mid 20's), who have maybe 3-6 months programming experience vs. younger fresh IT/Com Sci graduates. You see, even though I really love programming (Java/J2EE), but nobody gives me a feedback when I apply online. maybe because they preferred IT/ComSci graduates vs a career shifter like me.. So can you advice on how to improve my chance on being hired. How can i get a real-job programming experince if nobody is hiring me. I can make my own projects (working e-commerce site blah blah) but it is still different from the real job. And my codes are working but it still needs a lot of improvement and no one can tell me how to improve it because no one sees it (because I'm doing it alone?). Do you know any open source websites (java/j2ee/jee) / online home-based jobs who accepts java/j2ee/jee trainees.. Thank you very much

    Read the article

  • job offer in dead technology

    - by bold
    I have a job offer in a dead technology (specific programming language) that I don't want to work with nor do I believe it will offer many jobs in the future. It requires twice a year travels abroad, which not a plus in my eyes. On the other hand the money on the table is high. What would you do? edit: as its not clear I got a job in a programming language that is different from the academic programming language I worked with. Now I see it as a mistake to head to that direction.

    Read the article

  • Climbing the hacker ladder

    - by cobie
    This is not a question in which I am asking for opinions rather I am asking for first hand experience. I have been programming in python for quite a while and I feel solid enough in python programming. I can come up with algorithms for problems and implement them but I somehow feel I am stuck with remaining an apprentice. What are some first hand experiences on how to climb up the ladder and become better at programming as in learning about browsers security, compilers etc. Personal experiences would be valued in responses.

    Read the article

  • Becoming an expert vs boredom [closed]

    - by QAH
    I am a college student, and I love to program, period. I code all kinds of things in different kinds of languages. Although I enjoy programming, I have an extremely hard time sticking to one project for a long time. I attribute this shortcoming to my high level of curiosity, exploring different technologies, languages, libraries, etc. What would be best? Should I settle down more and spend time on becoming an expert in one or two programming fields, or should I be more of a jack of all trades, trying out all kinds of new technologies, languages, programming methods, etc.? I'm guessing that somewhere in the middle would be best. I'm always amazed at how many developers are able to create one or two projects, and develop on them for years. What techniques do you guys employ to help you stay focused on a project?

    Read the article

  • I'm a beginner Java programmer but I want to be useful

    - by user105418
    Programming has always interested me, but after learning some of the basics of Java(I'm talking high school level), I don't really know what to do from there. I want to be able to apply what I learned in some way, whether it be a volunteer project or something, but I probably don't know enough programming. Is it possible for a novice Java programmer to be useful in some way whatsoever. I want to do this because I feel like I could learn more about programming by helping people in theirs, but I'm not sure if I'm even able to this though. Does anyone have any suggestions on how I can contribute to other people's project in some way or how to apply it in some way?

    Read the article

  • Should I continue to learn and program using ansi c or non standard? [closed]

    - by Erik
    I am a Cs Student, enthusiastic about programming. C is my first programming language that I learned. (Never been exposed to programming, data structures, algorithms before) I failed the exam because I studied on my own and didn't know how to use windows non standard libraries like conio.h. (I had to draw circles, etc using get x and get y). I told them but they just don't care. What do I do because for a beginner this is very confusing? Do I continue to study ansi c on my own, or should I study non standard c? Should I do them in parallel? *I did use the search bar but found nothing useful that could help me.

    Read the article

  • Usefull skills from a computer science degree

    - by Tom Squires
    I did my degree in physics and moved later into programming. I have two and a half years experience under my belt and like to think I write good code. I am, however, concerned that not doing a compsci degree has left holes in my knowledge. I would like to fill those up now since I know I want to be doing programming for the rest of my career. What skills/techniques did you learn in your compsci degree that one wouldn't pick up from on-the-job programming?

    Read the article

  • Why do we need private variables?

    - by rak
    Why do we need private variables in classes in the context of programming? Every book on programming I've read says this is a private variable, this is how you define it but stops there. The wording of these explanations always seemed to me like we really have a crisis of trust in our profession. The explanations always sounded like other programmers are out to mess up our code. Yet, there are many programming languages that do not have private variables. What do private variables help prevent? How do you decide if a particular of properties should be private or not? If by default every field SHOULD be private then why are there public data members in a class? Under what circumstances should a variable be made public?

    Read the article

  • 27 vidéos techniques des Qt DevDays 2005, 2006 et 2008 sont désormais rendues publiques par Qt eLear

    L'équipe eLearning de Qt a depuis quelques temps cherché à récupérer des vidéos techniques issues des conférences des anciens QtDevDays dans l'optique de les faire partager à tout le monde. C'est aujourd'hui chose faite avec la publication en ligne de 27 présentations techniques ce qui correspond à 22h30min de vidéos. Les sujets traités sont toujours valides aujourd'hui, même si le framework a évolué au fil des années. 2005 :All About Qt Widgets Effective Graphics Programming Practical Model/View Programming Threaded Programming with Qt - Good Practise Writing Custom Styles with QStyle Writing plugin applications with Qt 2006 :Advanced Item Views...

    Read the article

  • Windows Phone App with 4SQ

    - by Nuttanon Pornpipak
    I'm want to create a my own Coffee shop app for semester's project. It's Windows Phone App. The App can i.e. view who is check-in here now , view menu , view photo by using 4SQ Endpoint APIs. And my problem is I don't know how to start it...which book i should read about C# and I don't know which knowledge (keyword) should i google it i.e. GET POST METHOD , JSON I ever used 4SQ Endpoint APIs once with javascript (jquery) $.ajax{(.....)} to get data from 4SQ Endpoint APIs So I googled and found JSON.NET Class but I don't know how to use it because i never programming in C# I'm just begin programming. I can programming in C only. Thank you Sorry for my bad grammar

    Read the article

  • For asp.net mvc is this a three tiered solution?

    - by bbb
    I am a asp.net mvc programmer and if I want to start a project I do this: I make a class library named Model for my models. I make a class library named Infrastructure.Repository for database processes I make a class library named Application for business logic layer And finally I make a MVC project for the UI. But now some things are confusing me. Am I using 3-tier programming? If yes so what is n-tier programming and which one is better? If no so what is 3-tier programming? Some where I see that the tiers namings are DAL and BIZ. Which one is correct according to the naming convention?

    Read the article

  • Which C# Book to take?

    - by Fischkopf
    I was searching for a book to learn C#, but now i'm kinda stuck. I found many people asking the same question, and many people gave answers, but there are so many books about C# that it is really hard to decide which one to take. Now i reduced my choice on two books, but I just can't decide between them. Namely, there are: Programming C# 4.0 and C# 4.0 In A Nutshell The first thing I want to know, are these good choices? I'm not completely new to programming, but I just didn't find the right language until know, but i think C# is the one I was searching for. I know all the bassic stuff from Delphi/Java/Python so I think i'm not a complete beginner in programming. Is there anyone out there that read both books and can cleary explain whats the difference between them? I haven't found many reviews and sort of, so I just don't know which one to chose. Or is there any book that is better suiting me?

    Read the article

  • Orchestrating the Virtual Enterprise

    - by John Murphy
    During the American Industrial Revolution, the Ford Motor Company did it all. It turned raw materials into a showroom full of Model Ts. It owned a steel mill, a glass factory, and an automobile assembly line. The company was both self-sufficient and innovative and went on to become one of the largest and most profitable companies in the world. Nowadays, it's unusual for any business to follow this vertical integration model because its much harder to be best in class across such a wide a range of capabilities and services. Instead, businesses focus on their core competencies and outsource other business functions to specialized suppliers. They exchange vertical integration for collaboration. When done well, all parties benefit from this arrangement and the collaboration leads to the creation of an agile, lean and successful "virtual enterprise." Case in point: For Sun hardware, Oracle outsources most of its manufacturing and all of its logistics to third parties. These are vital activities, but ones where Oracle doesn't have a core competency, so we shift them to business partners who do. Within our enterprise, we always retain the core functions of product development, support, and most of the sales function, because that's what constitutes our core value to our customers. This is a perfect example of a virtual enterprise.  What are the implications of this? It means that we must exchange direct internal control for indirect external collaboration. This fundamentally changes the relative importance of different business processes, the boundaries of security and information sharing, and the relationship of the supply chain systems to the ERP. The challenge is that the systems required to support this virtual paradigm are still mired in "island enterprise" thinking. But help is at hand. Developments such as the Web, social networks, collaboration, and rules-based orchestration offer great potential to fundamentally re-architect supply chain systems to better support the virtual enterprise.  Supply Chain Management Systems in a Virtual Enterprise Historically enterprise software was constructed to automate the ERP - and then the supply chain systems extended the ERP. They were joined at the hip. In virtual enterprises, the supply chain system needs to be ERP agnostic, sitting above each of the ERPs that are distributed across the virtual enterprise - most of which are operating in other businesses. This is vital so that the supply chain system can manage the flow of material and the related information through the multiple enterprises. It has to have strong collaboration tools. It needs to be highly flexible. Users need to be able to see information that's coming from multiple sources and be able to react and respond to events across those sources.  Oracle Fusion Distributed Order Orchestration (DOO) is a perfect example of a supply chain system designed to operate in this virtual way. DOO embraces the idea that a company's fulfillment challenge is a distributed, multi-enterprise problem. It enables users to manage the process and the trading partners in a uniform way and deliver a consistent user experience while operating over a heterogeneous, virtual enterprise. This is a fundamental shift at the core of managing supply chains. It forces virtual enterprises to think architecturally about how best to construct their supply chain systems.  Case in point, almost everyone has ordered from Amazon.com at one time or another. Our orders are as likely to be fulfilled by third parties as they are by Amazon itself. To deliver the order promptly and efficiently, Amazon has to send it to the right fulfillment location and know the availability in that location. It needs to be able to track status of the fulfillment and deal with exceptions. As a virtual enterprise, Amazon's operations, using thousands of trading partners, requires a very different approach to fulfillment than the traditional 'take an order and ship it from your own warehouse' model. Amazon had no choice but to develop a complex, expensive and custom solution to tackle this problem as there used to be no product solution available. Now, other companies who want to follow similar models have a better off-the-shelf choice -- Oracle Distributed Order Orchestration (DOO).  Consider how another of our customers is using our distributed orchestration solution. This major airplane manufacturer has a highly complex business and interacts regularly with the U.S. Government and major airlines. It sits in the middle of an intricate supply chain and needed to improve visibility across its many different entities. Oracle Fusion DOO gives the company an orchestration mechanism so it could improve quality, speed, flexibility, and consistency without requiring an organ transplant of these highly complex legacy systems. Many retailers face the challenge of dealing with brick and mortar, Web, and reseller channels. They all need to be knitted together into a virtual enterprise experience that is consistent for their customers. When a large U.K. grocer with a strong brick and mortar retail operation added an online business, they turned to Oracle Fusion DOO to bring these entities together. Disturbing the Peace with Acquisitions Quite often a company's ERP system is disrupted when it acquires a new company. An acquisition can inject a new set of processes and systems -- or even introduce an entirely new business like Sun's hardware did at Oracle. This challenge has been a driver for some of our DOO customers. A large power management company is using Oracle Fusion DOO to provide the flexibility to rapidly integrate additional products and services into its central fulfillment operation. The Flip Side of Fulfillment Meanwhile, we haven't ignored similar challenges on the supply side of the equation. Specifically, how to manage complex supply in a flexible way when there are multiple trading parties involved? How to manage the supply to suppliers? How to manage critical components that need to merge in a tier two or tier three supply chain? By investing in supply orchestration solutions for the virtual enterprise, we plan to give users better visibility into their network of suppliers to help them drive down costs. We also think this technology and full orchestration process can be applied to the financial side of organizations. An example is transactions that flow through complex internal structures to minimize tax exposure. We can help companies manage those transactions effectively by thinking about the internal organization as a virtual enterprise and bringing the same solution set to this internal challenge.  The Clear Front Runner No other company is investing in solving the virtual enterprise supply chain issues like Oracle is. Oracle is in a unique position to become the gold standard in this market space. We have the infrastructure of Oracle technology. We already have an Oracle Fusion DOO application which embraces the best of what's required in this area. And we're absolutely committed to extending our Fusion solution to other use cases and delivering even more business value.

    Read the article

  • Is your team is a high-performing team?

    As a child I can remember looking out of the car window as my father drove along the Interstate in Florida while seeing prisoners wearing bright orange jump suits and prison guards keeping a watchful eye on them. The prisoners were taking part in a prison road gang. These road gangs were formed to help the state maintain the state highway infrastructure. The prisoner’s primary responsibilities are to pick up trash and debris from the roadway. This is a prime example of a work group or working group used by most prison systems in the United States. Work groups or working groups can be defined as a collection of individuals or entities working together to achieve a specific goal or accomplish a specific set of tasks. Typically these groups are only established for a short period of time and are dissolved once the desired outcome has been achieved. More often than not group members usually feel as though they are expendable to the group and some even dread that they are even in the group. "A team is a small number of people with complementary skills who are committed to a common purpose, performance goals, and approach for which they are mutually accountable." (Katzenbach and Smith, 1993) So how do you determine that a team is a high-performing team?  This can be determined by three base line criteria that include: consistently high quality output, the promotion of personal growth and well being of all team members, and most importantly the ability to learn and grow as a unit. Initially, a team can successfully create high-performing output without meeting all three criteria, however this will erode over time because team members will feel detached from the group or that they are not growing then the quality of the output will decline. High performing teams are similar to work groups because they both utilize a collection of individuals or entities to accomplish tasks. What distinguish a high-performing team from a work group are its characteristics. High-performing teams contain five core characteristics. These characteristics are what separate a group from a team. The five characteristics of a high-performing team include: Purpose, Performance Measures, People with Tasks and Relationship Skills, Process, and Preparation and Practice. A high-performing team is much more than a work group, and typically has a life cycle that can vary from team to team. The standard team lifecycle consists of five states and is comparable to a human life cycle. The five states of a high-performing team lifecycle include: Formulating, Storming, Normalizing, Performing, and Adjourning. The Formulating State of a team is first realized when the team members are first defined and roles are assigned to all members. This initial stage is very important because it can set the tone for the team and can ultimately determine its success or failure. In addition, this stage requires the team to have a strong leader because team members are normally unclear about specific roles, specific obstacles and goals that my lay ahead of them.  Finally, this stage is where most team members initially meet one another prior to working as a team unless the team members already know each other. The Storming State normally arrives directly after the formulation of a new team because there are still a lot of unknowns amongst the newly formed assembly. As a general rule most of the parties involved in the team are still getting used to the workload, pace of work, deadlines and the validity of various tasks that need to be performed by the group.  In this state everything is questioned because there are so many unknowns. Items commonly questioned include the credentials of others on the team, the actual validity of a project, and the leadership abilities of the team leader.  This can be exemplified by looking at the interactions between animals when they first meet.  If we look at a scenario where two people are walking directly toward each other with their dogs. The dogs will automatically enter the Storming State because they do not know the other dog. Typically in this situation, they attempt to define which is more dominating via play or fighting depending on how the dogs interact with each other. Once dominance has been defined and accepted by both dogs then they will either want to play or leave depending on how the dogs interacted and other environmental variables. Once the Storming State has been realized then the Normalizing State takes over. This state is entered by a team once all the questions of the Storming State have been answered and the team has been tested by a few tasks or projects.  Typically, participants in the team are filled with energy, and comradery, and a strong alliance with team goals and objectives.  A high school football team is a perfect example of the Normalizing State when they start their season.  The player positions have been assigned, the depth chart has been filled and everyone is focused on winning each game. All of the players encourage and expect each other to perform at the best of their abilities and are united by competition from other teams. The Performing State is achieved by a team when its history, working habits, and culture solidify the team as one working unit. In this state team members can anticipate specific behaviors, attitudes, reactions, and challenges are seen as opportunities and not problems. Additionally, each team member knows their role in the team’s success, and the roles of others. This is the most productive state of a group and is where all the time invested working together really pays off. If you look at an Olympic figure skating team skate you can easily see how the time spent working together benefits their performance. They skate as one unit even though it is comprised of two skaters. Each skater has their routine completely memorized as well as their partners. This allows them to anticipate each other’s moves on the ice makes their skating look effortless. The final state of a team is the Adjourning State. This state is where accomplishments by the team and each individual team member are recognized. Additionally, this state also allows for reflection of the interactions between team members, work accomplished and challenges that were faced. Finally, the team celebrates the challenges they have faced and overcome as a unit. Currently in the workplace teams are divided into two different types: Co-located and Distributed Teams. Co-located teams defined as the traditional group of people working together in an office, according to Andy Singleton of Assembla. This traditional type of a team has dominated business in the past due to inadequate technology, which forced workers to primarily interact with one another via face to face meetings.  Team meetings are primarily lead by the person with the highest status in the company. Having personally, participated in meetings of this type, usually a select few of the team members dominate the flow of communication which reduces the input of others in group discussions. Since discussions are dominated by a select few individuals the discussions and group discussion are skewed in favor of the individuals who communicate the most in meetings. In addition, Team members might not give their full opinions on a topic of discussion in part not to offend or create controversy amongst the team and can alter decision made in meetings towards those of the opinions of the dominating team members. Distributed teams are by definition spread across an area or subdivided into separate sections. That is exactly what distributed teams when compared to a more traditional team. It is common place for distributed teams to have team members across town, in the next state, across the country and even with the advances in technology over the last 20 year across the world. These teams allow for more diversity compared to the other type of teams because they allow for more flexibility regarding location. A team could consist of a 30 year old male Italian project manager from New York, a 50 year old female Hispanic from California and a collection of programmers from India because technology allows them to communicate as if they were standing next to one another.  In addition, distributed team members consult with more team members prior to making decisions compared to traditional teams, and take longer to come to decisions due to the changes in time zones and cultural events. However, team members feel more empowered to speak out when they do not agree with the team and to notify others of potential issues regarding the work that the team is doing. Virtual teams which are a subset of the distributed team type is changing organizational strategies due to the fact that a team can now in essence be working 24 hrs a day because of utilizing employees in various time zones and locations.  A primary example of this is with customer services departments, a company can have multiple call centers spread across multiple time zones allowing them to appear to be open 24 hours a day while all a employees work from 9AM to 5 PM every day. Virtual teams also allow human resources departments to go after the best talent for the company regardless of where the potential employee works because they will be a part of a virtual team all that is need is the proper technology to be setup to allow everyone to communicate. In addition to allowing employees to work from home, the company can save space and resources by not having to provide a desk for every team member. In fact, those team members that randomly come into the office can actually share one desk amongst multiple people. This is definitely a cost cutting plus given the current state of the economy. One thing that can turn a team into a high-performing team is leadership. High-performing team leaders need to focus on investing in ongoing personal development, provide team members with direction, structure, and resources needed to accomplish their work, make the right interventions at the right time, and help the team manage boundaries between the team and various external parties involved in the teams work. A team leader needs to invest in ongoing personal development in order to effectively manage their team. People have said that attitude is everything; this is very true about leaders and leadership. A team takes on the attitudes and behaviors of its leaders. This can potentially harm the team and the team’s output. Leaders must concentrate on self-awareness, and understanding their team’s group dynamics to fully understand how to lead them. In addition, always learning new leadership techniques from other effective leaders is also very beneficial. Providing team members with direction, structure, and resources that they need to accomplish their work collectively sounds easy, but it is not.  Leaders need to be able to effectively communicate with their team on how their work helps the company reach for its organizational vision. Conversely, the leader needs to allow his team to work autonomously within specific guidelines to turn the company’s vision into a reality.  This being said the team must be appropriately staffed according to the size of the team’s tasks and their complexity. These tasks should be clear, and be meaningful to the company’s objectives and allow for feedback to be exchanged with the leader and the team member and the leader and upper management. Now if the team is properly staffed, and has a clear and full understanding of what is to be done; the company also must supply the workers with the proper tools to achieve the tasks that they are asked to do. No one should be asked to dig a hole without being given a shovel.  Finally, leaders must reward their team members for accomplishments that they achieve. Awards could range from just a simple congratulatory email, a party to close the completion of a large project, or other monetary rewards. Managing boundaries is very important for team leaders because it can alter attitudes of team members and can add undue stress to the team which will force them to loose focus on the tasks at hand for the group. Team leaders should promote communication between team members so that burdens are shared amongst the team and solutions can be derived from hearing the opinions of multiple sources. This also reinforces team camaraderie and working as a unit. Team leaders must manage the type and timing of interventions as to not create an even bigger mess within the team. Poorly timed interventions can really deflate team members and make them question themselves. This could really increase further and undue interventions by the team leader. Typically, the best time for interventions is when the team is just starting to form so that all unproductive behaviors are removed from the team and that it can retain focus on its agenda. If an intervention is effectively executed the team will feel energized about the work that they are doing, promote communication and interaction amongst the group and improve moral overall. High-performing teams are very import to organizations because they consistently produce high quality output and develop a collective purpose for their work. This drive to succeed allows team members to utilize specific talents allowing for growth in these areas.  In addition, these team members usually take on a sense of ownership with their projects and feel that the other team members are irreplaceable. References: http://blog.assembla.com/assemblablog/tabid/12618/bid/3127/Three-ways-to-organize-your-team-co-located-outsourced-or-global.aspx Katzenbach, J.R. & Smith, D.K. (1993). The Wisdom of Teams: Creating the High-performance Organization. Boston: Harvard Business School.

    Read the article

  • What's up with OCFS2?

    - by wcoekaer
    On Linux there are many filesystem choices and even from Oracle we provide a number of filesystems, all with their own advantages and use cases. Customers often confuse ACFS with OCFS or OCFS2 which then causes assumptions to be made such as one replacing the other etc... I thought it would be good to write up a summary of how OCFS2 got to where it is, what we're up to still, how it is different from other options and how this really is a cool native Linux cluster filesystem that we worked on for many years and is still widely used. Work on a cluster filesystem at Oracle started many years ago, in the early 2000's when the Oracle Database Cluster development team wrote a cluster filesystem for Windows that was primarily focused on providing an alternative to raw disk devices and help customers with the deployment of Oracle Real Application Cluster (RAC). Oracle RAC is a cluster technology that lets us make a cluster of Oracle Database servers look like one big database. The RDBMS runs on many nodes and they all work on the same data. It's a Shared Disk database design. There are many advantages doing this but I will not go into detail as that is not the purpose of my write up. Suffice it to say that Oracle RAC expects all the database data to be visible in a consistent, coherent way, across all the nodes in the cluster. To do that, there were/are a few options : 1) use raw disk devices that are shared, through SCSI, FC, or iSCSI 2) use a network filesystem (NFS) 3) use a cluster filesystem(CFS) which basically gives you a filesystem that's coherent across all nodes using shared disks. It is sort of (but not quite) combining option 1 and 2 except that you don't do network access to the files, the files are effectively locally visible as if it was a local filesystem. So OCFS (Oracle Cluster FileSystem) on Windows was born. Since Linux was becoming a very important and popular platform, we decided that we would also make this available on Linux and thus the porting of OCFS/Windows started. The first version of OCFS was really primarily focused on replacing the use of Raw devices with a simple filesystem that lets you create files and provide direct IO to these files to get basically native raw disk performance. The filesystem was not designed to be fully POSIX compliant and it did not have any where near good/decent performance for regular file create/delete/access operations. Cache coherency was easy since it was basically always direct IO down to the disk device and this ensured that any time one issues a write() command it would go directly down to the disk, and not return until the write() was completed. Same for read() any sort of read from a datafile would be a read() operation that went all the way to disk and return. We did not cache any data when it came down to Oracle data files. So while OCFS worked well for that, since it did not have much of a normal filesystem feel, it was not something that could be submitted to the kernel mail list for inclusion into Linux as another native linux filesystem (setting aside the Windows porting code ...) it did its job well, it was very easy to configure, node membership was simple, locking was disk based (so very slow but it existed), you could create regular files and do regular filesystem operations to a certain extend but anything that was not database data file related was just not very useful in general. Logfiles ok, standard filesystem use, not so much. Up to this point, all the work was done, at Oracle, by Oracle developers. Once OCFS (1) was out for a while and there was a lot of use in the database RAC world, many customers wanted to do more and were asking for features that you'd expect in a normal native filesystem, a real "general purposes cluster filesystem". So the team sat down and basically started from scratch to implement what's now known as OCFS2 (Oracle Cluster FileSystem release 2). Some basic criteria were : Design it with a real Distributed Lock Manager and use the network for lock negotiation instead of the disk Make it a Linux native filesystem instead of a native shim layer and a portable core Support standard Posix compliancy and be fully cache coherent with all operations Support all the filesystem features Linux offers (ACL, extended Attributes, quotas, sparse files,...) Be modern, support large files, 32/64bit, journaling, data ordered journaling, endian neutral, we can mount on both endian /cross architecture,.. Needless to say, this was a huge development effort that took many years to complete. A few big milestones happened along the way... OCFS2 was development in the open, we did not have a private tree that we worked on without external code review from the Linux Filesystem maintainers, great folks like Christopher Hellwig reviewed the code regularly to make sure we were not doing anything out of line, we submitted the code for review on lkml a number of times to see if we were getting close for it to be included into the mainline kernel. Using this development model is standard practice for anyone that wants to write code that goes into the kernel and having any chance of doing so without a complete rewrite or.. shall I say flamefest when submitted. It saved us a tremendous amount of time by not having to re-fit code for it to be in a Linus acceptable state. Some other filesystems that were trying to get into the kernel that didn't follow an open development model had a lot harder time and a lot harsher criticism. March 2006, when Linus released 2.6.16, OCFS2 officially became part of the mainline kernel, it was accepted a little earlier in the release candidates but in 2.6.16. OCFS2 became officially part of the mainline Linux kernel tree as one of the many filesystems. It was the first cluster filesystem to make it into the kernel tree. Our hope was that it would then end up getting picked up by the distribution vendors to make it easy for everyone to have access to a CFS. Today the source code for OCFS2 is approximately 85000 lines of code. We made OCFS2 production with full support for customers that ran Oracle database on Linux, no extra or separate support contract needed. OCFS2 1.0.0 started being built for RHEL4 for x86, x86-64, ppc, s390x and ia64. For RHEL5 starting with OCFS2 1.2. SuSE was very interested in high availability and clustering and decided to build and include OCFS2 with SLES9 for their customers and was, next to Oracle, the main contributor to the filesystem for both new features and bug fixes. Source code was always available even prior to inclusion into mainline and as of 2.6.16, source code was just part of a Linux kernel download from kernel.org, which it still is, today. So the latest OCFS2 code is always the upstream mainline Linux kernel. OCFS2 is the cluster filesystem used in Oracle VM 2 and Oracle VM 3 as the virtual disk repository filesystem. Since the filesystem is in the Linux kernel it's released under the GPL v2 The release model has always been that new feature development happened in the mainline kernel and we then built consistent, well tested, snapshots that had versions, 1.2, 1.4, 1.6, 1.8. But these releases were effectively just snapshots in time that were tested for stability and release quality. OCFS2 is very easy to use, there's a simple text file that contains the node information (hostname, node number, cluster name) and a file that contains the cluster heartbeat timeouts. It is very small, and very efficient. As Sunil Mushran wrote in the manual : OCFS2 is an efficient, easily configured, quickly installed, fully integrated and compatible, feature-rich, architecture and endian neutral, cache coherent, ordered data journaling, POSIX-compliant, shared disk cluster file system. Here is a list of some of the important features that are included : Variable Block and Cluster sizes Supports block sizes ranging from 512 bytes to 4 KB and cluster sizes ranging from 4 KB to 1 MB (increments in power of 2). Extent-based Allocations Tracks the allocated space in ranges of clusters making it especially efficient for storing very large files. Optimized Allocations Supports sparse files, inline-data, unwritten extents, hole punching and allocation reservation for higher performance and efficient storage. File Cloning/snapshots REFLINK is a feature which introduces copy-on-write clones of files in a cluster coherent way. Indexed Directories Allows efficient access to millions of objects in a directory. Metadata Checksums Detects silent corruption in inodes and directories. Extended Attributes Supports attaching an unlimited number of name:value pairs to the file system objects like regular files, directories, symbolic links, etc. Advanced Security Supports POSIX ACLs and SELinux in addition to the traditional file access permission model. Quotas Supports user and group quotas. Journaling Supports both ordered and writeback data journaling modes to provide file system consistency in the event of power failure or system crash. Endian and Architecture neutral Supports a cluster of nodes with mixed architectures. Allows concurrent mounts on nodes running 32-bit and 64-bit, little-endian (x86, x86_64, ia64) and big-endian (ppc64) architectures. In-built Cluster-stack with DLM Includes an easy to configure, in-kernel cluster-stack with a distributed lock manager. Buffered, Direct, Asynchronous, Splice and Memory Mapped I/Os Supports all modes of I/Os for maximum flexibility and performance. Comprehensive Tools Support Provides a familiar EXT3-style tool-set that uses similar parameters for ease-of-use. The filesystem was distributed for Linux distributions in separate RPM form and this had to be built for every single kernel errata release or every updated kernel provided by the vendor. We provided builds from Oracle for Oracle Linux and all kernels released by Oracle and for Red Hat Enterprise Linux. SuSE provided the modules directly for every kernel they shipped. With the introduction of the Unbreakable Enterprise Kernel for Oracle Linux and our interest in reducing the overhead of building filesystem modules for every minor release, we decide to make OCFS2 available as part of UEK. There was no more need for separate kernel modules, everything was built-in and a kernel upgrade automatically updated the filesystem, as it should. UEK allowed us to not having to backport new upstream filesystem code into an older kernel version, backporting features into older versions introduces risk and requires extra testing because the code is basically partially rewritten. The UEK model works really well for continuing to provide OCFS2 without that extra overhead. Because the RHEL kernel did not contain OCFS2 as a kernel module (it is in the source tree but it is not built by the vendor in kernel module form) we stopped adding the extra packages to Oracle Linux and its RHEL compatible kernel and for RHEL. Oracle Linux customers/users obviously get OCFS2 included as part of the Unbreakable Enterprise Kernel, SuSE customers get it by SuSE distributed with SLES and Red Hat can decide to distribute OCFS2 to their customers if they chose to as it's just a matter of compiling the module and making it available. OCFS2 today, in the mainline kernel is pretty much feature complete in terms of integration with every filesystem feature Linux offers and it is still actively maintained with Joel Becker being the primary maintainer. Since we use OCFS2 as part of Oracle VM, we continue to look at interesting new functionality to add, REFLINK was a good example, and as such we continue to enhance the filesystem where it makes sense. Bugfixes and any sort of code that goes into the mainline Linux kernel that affects filesystems, automatically also modifies OCFS2 so it's in kernel, actively maintained but not a lot of new development happening at this time. We continue to fully support OCFS2 as part of Oracle Linux and the Unbreakable Enterprise Kernel and other vendors make their own decisions on support as it's really a Linux cluster filesystem now more than something that we provide to customers. It really just is part of Linux like EXT3 or BTRFS etc, the OS distribution vendors decide. Do not confuse OCFS2 with ACFS (ASM cluster Filesystem) also known as Oracle Cloud Filesystem. ACFS is a filesystem that's provided by Oracle on various OS platforms and really integrates into Oracle ASM (Automatic Storage Management). It's a very powerful Cluster Filesystem but it's not distributed as part of the Operating System, it's distributed with the Oracle Database product and installs with and lives inside Oracle ASM. ACFS obviously is fully supported on Linux (Oracle Linux, Red Hat Enterprise Linux) but OCFS2 independently as a native Linux filesystem is also, and continues to also be supported. ACFS is very much tied into the Oracle RDBMS, OCFS2 is just a standard native Linux filesystem with no ties into Oracle products. Customers running the Oracle database and ASM really should consider using ACFS as it also provides storage/clustered volume management. Customers wanting to use a simple, easy to use generic Linux cluster filesystem should consider using OCFS2. To learn more about OCFS2 in detail, you can find good documentation on http://oss.oracle.com/projects/ocfs2 in the Documentation area, or get the latest mainline kernel from http://kernel.org and read the source. One final, unrelated note - since I am not always able to publicly answer or respond to comments, I do not want to selectively publish comments from readers. Sometimes I forget to publish comments, sometime I publish them and sometimes I would publish them but if for some reason I cannot publicly comment on them, it becomes a very one-sided stream. So for now I am going to not publish comments from anyone, to be fair to all sides. You are always welcome to email me and I will do my best to respond to technical questions, questions about strategy or direction are sometimes not possible to answer for obvious reasons.

    Read the article

  • How John Got 15x Improvement Without Really Trying

    - by rchrd
    The following article was published on a Sun Microsystems website a number of years ago by John Feo. It is still useful and worth preserving. So I'm republishing it here.  How I Got 15x Improvement Without Really Trying John Feo, Sun Microsystems Taking ten "personal" program codes used in scientific and engineering research, the author was able to get from 2 to 15 times performance improvement easily by applying some simple general optimization techniques. Introduction Scientific research based on computer simulation depends on the simulation for advancement. The research can advance only as fast as the computational codes can execute. The codes' efficiency determines both the rate and quality of results. In the same amount of time, a faster program can generate more results and can carry out a more detailed simulation of physical phenomena than a slower program. Highly optimized programs help science advance quickly and insure that monies supporting scientific research are used as effectively as possible. Scientific computer codes divide into three broad categories: ISV, community, and personal. ISV codes are large, mature production codes developed and sold commercially. The codes improve slowly over time both in methods and capabilities, and they are well tuned for most vendor platforms. Since the codes are mature and complex, there are few opportunities to improve their performance solely through code optimization. Improvements of 10% to 15% are typical. Examples of ISV codes are DYNA3D, Gaussian, and Nastran. Community codes are non-commercial production codes used by a particular research field. Generally, they are developed and distributed by a single academic or research institution with assistance from the community. Most users just run the codes, but some develop new methods and extensions that feed back into the general release. The codes are available on most vendor platforms. Since these codes are younger than ISV codes, there are more opportunities to optimize the source code. Improvements of 50% are not unusual. Examples of community codes are AMBER, CHARM, BLAST, and FASTA. Personal codes are those written by single users or small research groups for their own use. These codes are not distributed, but may be passed from professor-to-student or student-to-student over several years. They form the primordial ocean of applications from which community and ISV codes emerge. Government research grants pay for the development of most personal codes. This paper reports on the nature and performance of this class of codes. Over the last year, I have looked at over two dozen personal codes from more than a dozen research institutions. The codes cover a variety of scientific fields, including astronomy, atmospheric sciences, bioinformatics, biology, chemistry, geology, and physics. The sources range from a few hundred lines to more than ten thousand lines, and are written in Fortran, Fortran 90, C, and C++. For the most part, the codes are modular, documented, and written in a clear, straightforward manner. They do not use complex language features, advanced data structures, programming tricks, or libraries. I had little trouble understanding what the codes did or how data structures were used. Most came with a makefile. Surprisingly, only one of the applications is parallel. All developers have access to parallel machines, so availability is not an issue. Several tried to parallelize their applications, but stopped after encountering difficulties. Lack of education and a perception that parallelism is difficult prevented most from trying. I parallelized several of the codes using OpenMP, and did not judge any of the codes as difficult to parallelize. Even more surprising than the lack of parallelism is the inefficiency of the codes. I was able to get large improvements in performance in a matter of a few days applying simple optimization techniques. Table 1 lists ten representative codes [names and affiliation are omitted to preserve anonymity]. Improvements on one processor range from 2x to 15.5x with a simple average of 4.75x. I did not use sophisticated performance tools or drill deep into the program's execution character as one would do when tuning ISV or community codes. Using only a profiler and source line timers, I identified inefficient sections of code and improved their performance by inspection. The changes were at a high level. I am sure there is another factor of 2 or 3 in each code, and more if the codes are parallelized. The study’s results show that personal scientific codes are running many times slower than they should and that the problem is pervasive. Computational scientists are not sloppy programmers; however, few are trained in the art of computer programming or code optimization. I found that most have a working knowledge of some programming language and standard software engineering practices; but they do not know, or think about, how to make their programs run faster. They simply do not know the standard techniques used to make codes run faster. In fact, they do not even perceive that such techniques exist. The case studies described in this paper show that applying simple, well known techniques can significantly increase the performance of personal codes. It is important that the scientific community and the Government agencies that support scientific research find ways to better educate academic scientific programmers. The inefficiency of their codes is so bad that it is retarding both the quality and progress of scientific research. # cacheperformance redundantoperations loopstructures performanceimprovement 1 x x 15.5 2 x 2.8 3 x x 2.5 4 x 2.1 5 x x 2.0 6 x 5.0 7 x 5.8 8 x 6.3 9 2.2 10 x x 3.3 Table 1 — Area of improvement and performance gains of 10 codes The remainder of the paper is organized as follows: sections 2, 3, and 4 discuss the three most common sources of inefficiencies in the codes studied. These are cache performance, redundant operations, and loop structures. Each section includes several examples. The last section summaries the work and suggests a possible solution to the issues raised. Optimizing cache performance Commodity microprocessor systems use caches to increase memory bandwidth and reduce memory latencies. Typical latencies from processor to L1, L2, local, and remote memory are 3, 10, 50, and 200 cycles, respectively. Moreover, bandwidth falls off dramatically as memory distances increase. Programs that do not use cache effectively run many times slower than programs that do. When optimizing for cache, the biggest performance gains are achieved by accessing data in cache order and reusing data to amortize the overhead of cache misses. Secondary considerations are prefetching, associativity, and replacement; however, the understanding and analysis required to optimize for the latter are probably beyond the capabilities of the non-expert. Much can be gained simply by accessing data in the correct order and maximizing data reuse. 6 out of the 10 codes studied here benefited from such high level optimizations. Array Accesses The most important cache optimization is the most basic: accessing Fortran array elements in column order and C array elements in row order. Four of the ten codes—1, 2, 4, and 10—got it wrong. Compilers will restructure nested loops to optimize cache performance, but may not do so if the loop structure is too complex, or the loop body includes conditionals, complex addressing, or function calls. In code 1, the compiler failed to invert a key loop because of complex addressing do I = 0, 1010, delta_x IM = I - delta_x IP = I + delta_x do J = 5, 995, delta_x JM = J - delta_x JP = J + delta_x T1 = CA1(IP, J) + CA1(I, JP) T2 = CA1(IM, J) + CA1(I, JM) S1 = T1 + T2 - 4 * CA1(I, J) CA(I, J) = CA1(I, J) + D * S1 end do end do In code 2, the culprit is conditionals do I = 1, N do J = 1, N If (IFLAG(I,J) .EQ. 0) then T1 = Value(I, J-1) T2 = Value(I-1, J) T3 = Value(I, J) T4 = Value(I+1, J) T5 = Value(I, J+1) Value(I,J) = 0.25 * (T1 + T2 + T5 + T4) Delta = ABS(T3 - Value(I,J)) If (Delta .GT. MaxDelta) MaxDelta = Delta endif enddo enddo I fixed both programs by inverting the loops by hand. Code 10 has three-dimensional arrays and triply nested loops. The structure of the most computationally intensive loops is too complex to invert automatically or by hand. The only practical solution is to transpose the arrays so that the dimension accessed by the innermost loop is in cache order. The arrays can be transposed at construction or prior to entering a computationally intensive section of code. The former requires all array references to be modified, while the latter is cost effective only if the cost of the transpose is amortized over many accesses. I used the second approach to optimize code 10. Code 5 has four-dimensional arrays and loops are nested four deep. For all of the reasons cited above the compiler is not able to restructure three key loops. Assume C arrays and let the four dimensions of the arrays be i, j, k, and l. In the original code, the index structure of the three loops is L1: for i L2: for i L3: for i for l for l for j for k for j for k for j for k for l So only L3 accesses array elements in cache order. L1 is a very complex loop—much too complex to invert. I brought the loop into cache alignment by transposing the second and fourth dimensions of the arrays. Since the code uses a macro to compute all array indexes, I effected the transpose at construction and changed the macro appropriately. The dimensions of the new arrays are now: i, l, k, and j. L3 is a simple loop and easily inverted. L2 has a loop-carried scalar dependence in k. By promoting the scalar name that carries the dependence to an array, I was able to invert the third and fourth subloops aligning the loop with cache. Code 5 is by far the most difficult of the four codes to optimize for array accesses; but the knowledge required to fix the problems is no more than that required for the other codes. I would judge this code at the limits of, but not beyond, the capabilities of appropriately trained computational scientists. Array Strides When a cache miss occurs, a line (64 bytes) rather than just one word is loaded into the cache. If data is accessed stride 1, than the cost of the miss is amortized over 8 words. Any stride other than one reduces the cost savings. Two of the ten codes studied suffered from non-unit strides. The codes represent two important classes of "strided" codes. Code 1 employs a multi-grid algorithm to reduce time to convergence. The grids are every tenth, fifth, second, and unit element. Since time to convergence is inversely proportional to the distance between elements, coarse grids converge quickly providing good starting values for finer grids. The better starting values further reduce the time to convergence. The downside is that grids of every nth element, n > 1, introduce non-unit strides into the computation. In the original code, much of the savings of the multi-grid algorithm were lost due to this problem. I eliminated the problem by compressing (copying) coarse grids into continuous memory, and rewriting the computation as a function of the compressed grid. On convergence, I copied the final values of the compressed grid back to the original grid. The savings gained from unit stride access of the compressed grid more than paid for the cost of copying. Using compressed grids, the loop from code 1 included in the previous section becomes do j = 1, GZ do i = 1, GZ T1 = CA(i+0, j-1) + CA(i-1, j+0) T4 = CA1(i+1, j+0) + CA1(i+0, j+1) S1 = T1 + T4 - 4 * CA1(i+0, j+0) CA(i+0, j+0) = CA1(i+0, j+0) + DD * S1 enddo enddo where CA and CA1 are compressed arrays of size GZ. Code 7 traverses a list of objects selecting objects for later processing. The labels of the selected objects are stored in an array. The selection step has unit stride, but the processing steps have irregular stride. A fix is to save the parameters of the selected objects in temporary arrays as they are selected, and pass the temporary arrays to the processing functions. The fix is practical if the same parameters are used in selection as in processing, or if processing comprises a series of distinct steps which use overlapping subsets of the parameters. Both conditions are true for code 7, so I achieved significant improvement by copying parameters to temporary arrays during selection. Data reuse In the previous sections, we optimized for spatial locality. It is also important to optimize for temporal locality. Once read, a datum should be used as much as possible before it is forced from cache. Loop fusion and loop unrolling are two techniques that increase temporal locality. Unfortunately, both techniques increase register pressure—as loop bodies become larger, the number of registers required to hold temporary values grows. Once register spilling occurs, any gains evaporate quickly. For multiprocessors with small register sets or small caches, the sweet spot can be very small. In the ten codes presented here, I found no opportunities for loop fusion and only two opportunities for loop unrolling (codes 1 and 3). In code 1, unrolling the outer and inner loop one iteration increases the number of result values computed by the loop body from 1 to 4, do J = 1, GZ-2, 2 do I = 1, GZ-2, 2 T1 = CA1(i+0, j-1) + CA1(i-1, j+0) T2 = CA1(i+1, j-1) + CA1(i+0, j+0) T3 = CA1(i+0, j+0) + CA1(i-1, j+1) T4 = CA1(i+1, j+0) + CA1(i+0, j+1) T5 = CA1(i+2, j+0) + CA1(i+1, j+1) T6 = CA1(i+1, j+1) + CA1(i+0, j+2) T7 = CA1(i+2, j+1) + CA1(i+1, j+2) S1 = T1 + T4 - 4 * CA1(i+0, j+0) S2 = T2 + T5 - 4 * CA1(i+1, j+0) S3 = T3 + T6 - 4 * CA1(i+0, j+1) S4 = T4 + T7 - 4 * CA1(i+1, j+1) CA(i+0, j+0) = CA1(i+0, j+0) + DD * S1 CA(i+1, j+0) = CA1(i+1, j+0) + DD * S2 CA(i+0, j+1) = CA1(i+0, j+1) + DD * S3 CA(i+1, j+1) = CA1(i+1, j+1) + DD * S4 enddo enddo The loop body executes 12 reads, whereas as the rolled loop shown in the previous section executes 20 reads to compute the same four values. In code 3, two loops are unrolled 8 times and one loop is unrolled 4 times. Here is the before for (k = 0; k < NK[u]; k++) { sum = 0.0; for (y = 0; y < NY; y++) { sum += W[y][u][k] * delta[y]; } backprop[i++]=sum; } and after code for (k = 0; k < KK - 8; k+=8) { sum0 = 0.0; sum1 = 0.0; sum2 = 0.0; sum3 = 0.0; sum4 = 0.0; sum5 = 0.0; sum6 = 0.0; sum7 = 0.0; for (y = 0; y < NY; y++) { sum0 += W[y][0][k+0] * delta[y]; sum1 += W[y][0][k+1] * delta[y]; sum2 += W[y][0][k+2] * delta[y]; sum3 += W[y][0][k+3] * delta[y]; sum4 += W[y][0][k+4] * delta[y]; sum5 += W[y][0][k+5] * delta[y]; sum6 += W[y][0][k+6] * delta[y]; sum7 += W[y][0][k+7] * delta[y]; } backprop[k+0] = sum0; backprop[k+1] = sum1; backprop[k+2] = sum2; backprop[k+3] = sum3; backprop[k+4] = sum4; backprop[k+5] = sum5; backprop[k+6] = sum6; backprop[k+7] = sum7; } for one of the loops unrolled 8 times. Optimizing for temporal locality is the most difficult optimization considered in this paper. The concepts are not difficult, but the sweet spot is small. Identifying where the program can benefit from loop unrolling or loop fusion is not trivial. Moreover, it takes some effort to get it right. Still, educating scientific programmers about temporal locality and teaching them how to optimize for it will pay dividends. Reducing instruction count Execution time is a function of instruction count. Reduce the count and you usually reduce the time. The best solution is to use a more efficient algorithm; that is, an algorithm whose order of complexity is smaller, that converges quicker, or is more accurate. Optimizing source code without changing the algorithm yields smaller, but still significant, gains. This paper considers only the latter because the intent is to study how much better codes can run if written by programmers schooled in basic code optimization techniques. The ten codes studied benefited from three types of "instruction reducing" optimizations. The two most prevalent were hoisting invariant memory and data operations out of inner loops. The third was eliminating unnecessary data copying. The nature of these inefficiencies is language dependent. Memory operations The semantics of C make it difficult for the compiler to determine all the invariant memory operations in a loop. The problem is particularly acute for loops in functions since the compiler may not know the values of the function's parameters at every call site when compiling the function. Most compilers support pragmas to help resolve ambiguities; however, these pragmas are not comprehensive and there is no standard syntax. To guarantee that invariant memory operations are not executed repetitively, the user has little choice but to hoist the operations by hand. The problem is not as severe in Fortran programs because in the absence of equivalence statements, it is a violation of the language's semantics for two names to share memory. Codes 3 and 5 are C programs. In both cases, the compiler did not hoist all invariant memory operations from inner loops. Consider the following loop from code 3 for (y = 0; y < NY; y++) { i = 0; for (u = 0; u < NU; u++) { for (k = 0; k < NK[u]; k++) { dW[y][u][k] += delta[y] * I1[i++]; } } } Since dW[y][u] can point to the same memory space as delta for one or more values of y and u, assignment to dW[y][u][k] may change the value of delta[y]. In reality, dW and delta do not overlap in memory, so I rewrote the loop as for (y = 0; y < NY; y++) { i = 0; Dy = delta[y]; for (u = 0; u < NU; u++) { for (k = 0; k < NK[u]; k++) { dW[y][u][k] += Dy * I1[i++]; } } } Failure to hoist invariant memory operations may be due to complex address calculations. If the compiler can not determine that the address calculation is invariant, then it can hoist neither the calculation nor the associated memory operations. As noted above, code 5 uses a macro to address four-dimensional arrays #define MAT4D(a,q,i,j,k) (double *)((a)->data + (q)*(a)->strides[0] + (i)*(a)->strides[3] + (j)*(a)->strides[2] + (k)*(a)->strides[1]) The macro is too complex for the compiler to understand and so, it does not identify any subexpressions as loop invariant. The simplest way to eliminate the address calculation from the innermost loop (over i) is to define a0 = MAT4D(a,q,0,j,k) before the loop and then replace all instances of *MAT4D(a,q,i,j,k) in the loop with a0[i] A similar problem appears in code 6, a Fortran program. The key loop in this program is do n1 = 1, nh nx1 = (n1 - 1) / nz + 1 nz1 = n1 - nz * (nx1 - 1) do n2 = 1, nh nx2 = (n2 - 1) / nz + 1 nz2 = n2 - nz * (nx2 - 1) ndx = nx2 - nx1 ndy = nz2 - nz1 gxx = grn(1,ndx,ndy) gyy = grn(2,ndx,ndy) gxy = grn(3,ndx,ndy) balance(n1,1) = balance(n1,1) + (force(n2,1) * gxx + force(n2,2) * gxy) * h1 balance(n1,2) = balance(n1,2) + (force(n2,1) * gxy + force(n2,2) * gyy)*h1 end do end do The programmer has written this loop well—there are no loop invariant operations with respect to n1 and n2. However, the loop resides within an iterative loop over time and the index calculations are independent with respect to time. Trading space for time, I precomputed the index values prior to the entering the time loop and stored the values in two arrays. I then replaced the index calculations with reads of the arrays. Data operations Ways to reduce data operations can appear in many forms. Implementing a more efficient algorithm produces the biggest gains. The closest I came to an algorithm change was in code 4. This code computes the inner product of K-vectors A(i) and B(j), 0 = i < N, 0 = j < M, for most values of i and j. Since the program computes most of the NM possible inner products, it is more efficient to compute all the inner products in one triply-nested loop rather than one at a time when needed. The savings accrue from reading A(i) once for all B(j) vectors and from loop unrolling. for (i = 0; i < N; i+=8) { for (j = 0; j < M; j++) { sum0 = 0.0; sum1 = 0.0; sum2 = 0.0; sum3 = 0.0; sum4 = 0.0; sum5 = 0.0; sum6 = 0.0; sum7 = 0.0; for (k = 0; k < K; k++) { sum0 += A[i+0][k] * B[j][k]; sum1 += A[i+1][k] * B[j][k]; sum2 += A[i+2][k] * B[j][k]; sum3 += A[i+3][k] * B[j][k]; sum4 += A[i+4][k] * B[j][k]; sum5 += A[i+5][k] * B[j][k]; sum6 += A[i+6][k] * B[j][k]; sum7 += A[i+7][k] * B[j][k]; } C[i+0][j] = sum0; C[i+1][j] = sum1; C[i+2][j] = sum2; C[i+3][j] = sum3; C[i+4][j] = sum4; C[i+5][j] = sum5; C[i+6][j] = sum6; C[i+7][j] = sum7; }} This change requires knowledge of a typical run; i.e., that most inner products are computed. The reasons for the change, however, derive from basic optimization concepts. It is the type of change easily made at development time by a knowledgeable programmer. In code 5, we have the data version of the index optimization in code 6. Here a very expensive computation is a function of the loop indices and so cannot be hoisted out of the loop; however, the computation is invariant with respect to an outer iterative loop over time. We can compute its value for each iteration of the computation loop prior to entering the time loop and save the values in an array. The increase in memory required to store the values is small in comparison to the large savings in time. The main loop in Code 8 is doubly nested. The inner loop includes a series of guarded computations; some are a function of the inner loop index but not the outer loop index while others are a function of the outer loop index but not the inner loop index for (j = 0; j < N; j++) { for (i = 0; i < M; i++) { r = i * hrmax; R = A[j]; temp = (PRM[3] == 0.0) ? 1.0 : pow(r, PRM[3]); high = temp * kcoeff * B[j] * PRM[2] * PRM[4]; low = high * PRM[6] * PRM[6] / (1.0 + pow(PRM[4] * PRM[6], 2.0)); kap = (R > PRM[6]) ? high * R * R / (1.0 + pow(PRM[4]*r, 2.0) : low * pow(R/PRM[6], PRM[5]); < rest of loop omitted > }} Note that the value of temp is invariant to j. Thus, we can hoist the computation for temp out of the loop and save its values in an array. for (i = 0; i < M; i++) { r = i * hrmax; TEMP[i] = pow(r, PRM[3]); } [N.B. – the case for PRM[3] = 0 is omitted and will be reintroduced later.] We now hoist out of the inner loop the computations invariant to i. Since the conditional guarding the value of kap is invariant to i, it behooves us to hoist the computation out of the inner loop, thereby executing the guard once rather than M times. The final version of the code is for (j = 0; j < N; j++) { R = rig[j] / 1000.; tmp1 = kcoeff * par[2] * beta[j] * par[4]; tmp2 = 1.0 + (par[4] * par[4] * par[6] * par[6]); tmp3 = 1.0 + (par[4] * par[4] * R * R); tmp4 = par[6] * par[6] / tmp2; tmp5 = R * R / tmp3; tmp6 = pow(R / par[6], par[5]); if ((par[3] == 0.0) && (R > par[6])) { for (i = 1; i <= imax1; i++) KAP[i] = tmp1 * tmp5; } else if ((par[3] == 0.0) && (R <= par[6])) { for (i = 1; i <= imax1; i++) KAP[i] = tmp1 * tmp4 * tmp6; } else if ((par[3] != 0.0) && (R > par[6])) { for (i = 1; i <= imax1; i++) KAP[i] = tmp1 * TEMP[i] * tmp5; } else if ((par[3] != 0.0) && (R <= par[6])) { for (i = 1; i <= imax1; i++) KAP[i] = tmp1 * TEMP[i] * tmp4 * tmp6; } for (i = 0; i < M; i++) { kap = KAP[i]; r = i * hrmax; < rest of loop omitted > } } Maybe not the prettiest piece of code, but certainly much more efficient than the original loop, Copy operations Several programs unnecessarily copy data from one data structure to another. This problem occurs in both Fortran and C programs, although it manifests itself differently in the two languages. Code 1 declares two arrays—one for old values and one for new values. At the end of each iteration, the array of new values is copied to the array of old values to reset the data structures for the next iteration. This problem occurs in Fortran programs not included in this study and in both Fortran 77 and Fortran 90 code. Introducing pointers to the arrays and swapping pointer values is an obvious way to eliminate the copying; but pointers is not a feature that many Fortran programmers know well or are comfortable using. An easy solution not involving pointers is to extend the dimension of the value array by 1 and use the last dimension to differentiate between arrays at different times. For example, if the data space is N x N, declare the array (N, N, 2). Then store the problem’s initial values in (_, _, 2) and define the scalar names new = 2 and old = 1. At the start of each iteration, swap old and new to reset the arrays. The old–new copy problem did not appear in any C program. In programs that had new and old values, the code swapped pointers to reset data structures. Where unnecessary coping did occur is in structure assignment and parameter passing. Structures in C are handled much like scalars. Assignment causes the data space of the right-hand name to be copied to the data space of the left-hand name. Similarly, when a structure is passed to a function, the data space of the actual parameter is copied to the data space of the formal parameter. If the structure is large and the assignment or function call is in an inner loop, then copying costs can grow quite large. While none of the ten programs considered here manifested this problem, it did occur in programs not included in the study. A simple fix is always to refer to structures via pointers. Optimizing loop structures Since scientific programs spend almost all their time in loops, efficient loops are the key to good performance. Conditionals, function calls, little instruction level parallelism, and large numbers of temporary values make it difficult for the compiler to generate tightly packed, highly efficient code. Conditionals and function calls introduce jumps that disrupt code flow. Users should eliminate or isolate conditionls to their own loops as much as possible. Often logical expressions can be substituted for if-then-else statements. For example, code 2 includes the following snippet MaxDelta = 0.0 do J = 1, N do I = 1, M < code omitted > Delta = abs(OldValue ? NewValue) if (Delta > MaxDelta) MaxDelta = Delta enddo enddo if (MaxDelta .gt. 0.001) goto 200 Since the only use of MaxDelta is to control the jump to 200 and all that matters is whether or not it is greater than 0.001, I made MaxDelta a boolean and rewrote the snippet as MaxDelta = .false. do J = 1, N do I = 1, M < code omitted > Delta = abs(OldValue ? NewValue) MaxDelta = MaxDelta .or. (Delta .gt. 0.001) enddo enddo if (MaxDelta) goto 200 thereby, eliminating the conditional expression from the inner loop. A microprocessor can execute many instructions per instruction cycle. Typically, it can execute one or more memory, floating point, integer, and jump operations. To be executed simultaneously, the operations must be independent. Thick loops tend to have more instruction level parallelism than thin loops. Moreover, they reduce memory traffice by maximizing data reuse. Loop unrolling and loop fusion are two techniques to increase the size of loop bodies. Several of the codes studied benefitted from loop unrolling, but none benefitted from loop fusion. This observation is not too surpising since it is the general tendency of programmers to write thick loops. As loops become thicker, the number of temporary values grows, increasing register pressure. If registers spill, then memory traffic increases and code flow is disrupted. A thick loop with many temporary values may execute slower than an equivalent series of thin loops. The biggest gain will be achieved if the thick loop can be split into a series of independent loops eliminating the need to write and read temporary arrays. I found such an occasion in code 10 where I split the loop do i = 1, n do j = 1, m A24(j,i)= S24(j,i) * T24(j,i) + S25(j,i) * U25(j,i) B24(j,i)= S24(j,i) * T25(j,i) + S25(j,i) * U24(j,i) A25(j,i)= S24(j,i) * C24(j,i) + S25(j,i) * V24(j,i) B25(j,i)= S24(j,i) * U25(j,i) + S25(j,i) * V25(j,i) C24(j,i)= S26(j,i) * T26(j,i) + S27(j,i) * U26(j,i) D24(j,i)= S26(j,i) * T27(j,i) + S27(j,i) * V26(j,i) C25(j,i)= S27(j,i) * S28(j,i) + S26(j,i) * U28(j,i) D25(j,i)= S27(j,i) * T28(j,i) + S26(j,i) * V28(j,i) end do end do into two disjoint loops do i = 1, n do j = 1, m A24(j,i)= S24(j,i) * T24(j,i) + S25(j,i) * U25(j,i) B24(j,i)= S24(j,i) * T25(j,i) + S25(j,i) * U24(j,i) A25(j,i)= S24(j,i) * C24(j,i) + S25(j,i) * V24(j,i) B25(j,i)= S24(j,i) * U25(j,i) + S25(j,i) * V25(j,i) end do end do do i = 1, n do j = 1, m C24(j,i)= S26(j,i) * T26(j,i) + S27(j,i) * U26(j,i) D24(j,i)= S26(j,i) * T27(j,i) + S27(j,i) * V26(j,i) C25(j,i)= S27(j,i) * S28(j,i) + S26(j,i) * U28(j,i) D25(j,i)= S27(j,i) * T28(j,i) + S26(j,i) * V28(j,i) end do end do Conclusions Over the course of the last year, I have had the opportunity to work with over two dozen academic scientific programmers at leading research universities. Their research interests span a broad range of scientific fields. Except for two programs that relied almost exclusively on library routines (matrix multiply and fast Fourier transform), I was able to improve significantly the single processor performance of all codes. Improvements range from 2x to 15.5x with a simple average of 4.75x. Changes to the source code were at a very high level. I did not use sophisticated techniques or programming tools to discover inefficiencies or effect the changes. Only one code was parallel despite the availability of parallel systems to all developers. Clearly, we have a problem—personal scientific research codes are highly inefficient and not running parallel. The developers are unaware of simple optimization techniques to make programs run faster. They lack education in the art of code optimization and parallel programming. I do not believe we can fix the problem by publishing additional books or training manuals. To date, the developers in questions have not studied the books or manual available, and are unlikely to do so in the future. Short courses are a possible solution, but I believe they are too concentrated to be much use. The general concepts can be taught in a three or four day course, but that is not enough time for students to practice what they learn and acquire the experience to apply and extend the concepts to their codes. Practice is the key to becoming proficient at optimization. I recommend that graduate students be required to take a semester length course in optimization and parallel programming. We would never give someone access to state-of-the-art scientific equipment costing hundreds of thousands of dollars without first requiring them to demonstrate that they know how to use the equipment. Yet the criterion for time on state-of-the-art supercomputers is at most an interesting project. Requestors are never asked to demonstrate that they know how to use the system, or can use the system effectively. A semester course would teach them the required skills. Government agencies that fund academic scientific research pay for most of the computer systems supporting scientific research as well as the development of most personal scientific codes. These agencies should require graduate schools to offer a course in optimization and parallel programming as a requirement for funding. About the Author John Feo received his Ph.D. in Computer Science from The University of Texas at Austin in 1986. After graduate school, Dr. Feo worked at Lawrence Livermore National Laboratory where he was the Group Leader of the Computer Research Group and principal investigator of the Sisal Language Project. In 1997, Dr. Feo joined Tera Computer Company where he was project manager for the MTA, and oversaw the programming and evaluation of the MTA at the San Diego Supercomputer Center. In 2000, Dr. Feo joined Sun Microsystems as an HPC application specialist. He works with university research groups to optimize and parallelize scientific codes. Dr. Feo has published over two dozen research articles in the areas of parallel parallel programming, parallel programming languages, and application performance.

    Read the article

  • Is a university education really worth it for a good programmer?

    - by Jon Purdy
    The title says it all, but here's the personal side of it: I've been doing design and programming for about as long as I can remember. If there's a programming problem, I can figure it out. (Though admittedly StackOverflow has allowed me to skip the figuring out and get straight to the doing in many instances.) I've made games, esoteric programming languages, and widgets and gizmos galore. I'm currently working on a general-purpose programming language. There's nothing I do better than programming. However, I'm just as passionate about design. Thus when I felt leaving high school that my design skills were lacking, I decided to attend university for New Media Design and Imaging, a digital design-related major. For a year, I diligently studied art and programmed in my free time. As the next year progressed, however, I was obligated to take fewer art and design classes and more technical classes. The trouble was of course that these classes were geared toward non-technical students, and were far beneath my skill level at the time. No amount of petitioning could overcome the institution's reluctance to allow me to test out of such classes, and the major offered no promise for any greater challenge in the future, so I took the extreme route: I switched into the technical equivalent of the major, New Media Interactive Development. A lot of my credits moved over into the new major, but many didn't. It would have been infeasible to switch to a more rigorous technical major such as Computer Science, and having tutored Computer Science students at every level here, I doubt I would be exposed to anything that I haven't already or won't eventually find out on my own, since I'm so involved in the field. I'm now on track to graduate perhaps a year later than I had planned, which puts a significant financial strain on my family and my future self. My schedule continues to be bogged down with classes that are wholly unnecessary for me to take. I'm being re-introduced to subjects that I've covered a thousand times over, simply because I've always been interested in it all. And though I succeed in avoiding the cynical and immature tactic of failing to complete work out of some undeserved sense of superiority, I'm becoming increasingly disillusioned by the lack of intellectual stimulation. Further, my school requires students to complete a number of quarters of co-op work experience proportional to their major. My original major required two quarters, but my current requires three, delaying my graduation even more. To top it all off, college is putting a severe strain on my relationship with my very close partner of a few years, so I've searched diligently for co-op jobs in my area, alas to no avail. I'm now in my third year, and approaching that point past which I can no longer handle this. Either I keep my head down, get a degree no matter what it takes, and try to get a job with a company that will pay me enough to do what I love that I can eventually pay off my loans; or I cut my losses now, move wherever there is work, and in six months start paying off what debt I've accumulated thus far. So the real question is: is a university education really more than just a formality? It's a big decision, and one I can't make lightly. I think this is the appropriate venue for this kind of question, and I hope it sticks around for the sake of others who might someday find themselves in similar situations. My heartfelt thanks for reading, and in advance for your help.

    Read the article

  • Why should I use Amazon Route 53 over my registrar's DNS servers?

    - by Abtin Forouzandeh
    I am building a site that I anticipate will have high usage. Currently, my registrar (GoDaddy) is handling DNS. However, Amazon's Route 53 looks interesting. They promise high speed and offer globally distributed DNS servers and a programmable interface. While GoDaddy doesn't offer a programmable interface, I assume their servers are geographically distributed as well. What are the main reasons I should opt to use Amazon Route 53 over free registrar-based DNS?

    Read the article

  • Why is there no service-oriented language?

    - by Wolfgang
    Edit: To avoid further confusion: I am not talking about web services and such. I am talking about structuring applications internally, it's not about how computers communicate. It's about programming languages, compilers and how the imperative programming paradigm is extended. Original: In the imperative programming field, we saw two paradigms in the past 20 years (or more): object-oriented (OO), and service-oriented (SO) aka. component-based (CB). Both paradigms extend the imperative programming paradigm by introducing their own notion of modules. OO calls them objects (and classes) and lets them encapsulates both data (fields) and procedures (methods) together. SO, in contrast, separates data (records, beans, ...) from code (components, services). However, only OO has programming languages which natively support its paradigm: Smalltalk, C++, Java and all other JVM-compatibles, C# and all other .NET-compatibles, Python etc. SO has no such native language. It only comes into existence on top of procedural languages or OO languages: COM/DCOM (binary, C, C++), CORBA, EJB, Spring, Guice (all Java), ... These SO frameworks clearly suffer from the missing native language support of their concepts. They start using OO classes to represent services and records. This leads to designs where there is a clear distinction between classes that have methods only (services) and those that have fields only (records). Inheritance between services or records is then simulated by inheritance of classes. Technically, its not kept so strictly but in general programmers are adviced to make classes to play only one of the two roles. They use additional, external languages to represent the missing parts: IDL's, XML configurations, Annotations in Java code, or even embedded DSL like in Guice. This is especially needed, but not limited to, since the composition of services is not part of the service code itself. In OO, objects create other objects so there is no need for such facilities but for SO there is because services don't instantiate or configure other services. They establish an inner-platform effect on top of OO (early EJB, CORBA) where the programmer has to write all the code that is needed to "drive" SO. Classes represent only a part of the nature of a service and lots of classes have to be written to form a service together. All that boiler plate is necessary because there is no SO compiler which would do it for the programmer. This is just like some people did it in C for OO when there was no C++. You just pass the record which holds the data of the object as a first parameter to the procedure which is the method. In a OO language this parameter is implicit and the compiler produces all the code that we need for virtual functions etc. For SO, this is clearly missing. Especially the newer frameworks extensively use AOP or introspection to add the missing parts to a OO language. This doesn't bring the necessary language expressiveness but avoids the boiler platform code described in the previous point. Some frameworks use code generation to produce the boiler plate code. Configuration files in XML or annotations in OO code is the source of information for this. Not all of the phenomena that I mentioned above can be attributed to SO but I hope it clearly shows that there is a need for a SO language. Since this paradigm is so popular: why isn't there one? Or maybe there are some academic ones but at least the industry doesn't use one.

    Read the article

  • C# Winforms vs WPF

    - by m0s
    Hi pros, I am a student and I do freelance here and there when I have opportunity. I believe my strongest language is C#. I don't really know what is going on in real programming world, so I was wondering if WPF did take over WinForms? I know the differences between two and how two can be used simultaneously but, I just don't want to invest my time in learning dying technologies, I hope you understand. So, for windows desktop programming what would you recommend to master WinForms, WPF or maybe both? I also get a lot that desktop programming is dead already and one should only care about learning web programming. Thanks for attention, any comments are greatly appreciated.

    Read the article

< Previous Page | 232 233 234 235 236 237 238 239 240 241 242 243  | Next Page >