Search Results

Search found 9062 results on 363 pages for 'big o'.

Page 45/363 | < Previous Page | 41 42 43 44 45 46 47 48 49 50 51 52  | Next Page >

  • Social HCM: Is Your Team Listening?

    - by Mike Stiles
    Does integrating Social HCM into your enterprise make sense? Consider Sam and Christina. Sam is a new hire at a big company. On the job 3 weeks, a question has come up on how to properly file an expense report to get reimbursed. It was covered in the onboarding session, but shockingly enough, Sam didn’t memorize or write down every word of the session. The answer is probably in a handout, in a stack of handouts 2 inches thick. It also might be on the employee web site…somewhere. Christina is a new hire at a different big company. She has the same question. She logs into her company’s social network, goes to the “new hires” group, asks her question and gets an answer in seconds. Christina says, “Cool!” Sam says, “Grrrr.” It’s safe to say the qualified talent your company wants is accustomed to using social platforms to communicate and get quick answers. As such, Christina is comfortable at her new company, whereas Sam is wondering what he’s gotten himself into. Companies that cling to talent communication and management systems that don’t speak to talent’s needs or expectations put themselves at risk. Right from the recruiting stage, prospects can determine if a company has embraced the communications tools of the 21st century. If they don’t see it, alarm bells go off. With great talent more in demand than ever, enterprises should reconsider making “this is the way we do it, you adapt to us” their mantra. Other blogs have clearly outlined that apart from meeting top recruits’ expectations, Social HCM benefits the organization itself in terms of efficiency, talent performance & measurement. Recruiting: Jobvite shows 64% of companies hired using social. 89% of job seekers are using social in their search. Social can give employers access to relevant communities of prospects and advance the brand. Nucleus Research found general hiring software can provide over 1,000% ROI by reducing churn and improving screening. Social talent acquisition should perform at least as well. Learning & Development:Employees, learning from the company or from peers, can be kept on top of the latest needed skillsets and engage in self-paced training so as to advance within the company. Performance Management:Just as gamers are egged on by levels and achievements, talent can reach for workplace kudos, be they shout-outs from peers & managers or formally established milestones. Plus employee reviews become consistent and fair as managers have access to the cumulative feedback social offers. Workflow and Collaboration:With workforces dispersing in terms of physical location, social provides a platform that helps eliminate drawbacks that would have brought just 10 years ago. Finding and connecting with just the right colleague to get the most relevant info at any given time has never been more possible…or expected. While yes, marketing has taken the social lead inside the enterprise, HCM (with the word “human” right there in its name) is the obvious locale for the next big integration of social in business. The technology is there. At Oracle, Fusion HCM apps are deeply embedded with Social HCM…just one example of systems taking social across the enterprise. Christina’s company is communicating with her in ways she’s used to. Sam’s company may as well be trying to talk to him using signal flags. @mikestilesPhoto via stock.xchng

    Read the article

  • My .NET Technology picks for 2011

    - by shiju
    My Technology predictions for 2011 Cloud computing and Mobile application development will be the hottest trends for 2011. I hope that Windows Azure will be very hot in year 2011 and lot of cloud computing adoption will be happen with Windows Azure on 2011. Web application scalability will be the big challenge for Architects in the next year and architecture approaches like CQRS will get some attention on next year. Architects will look on different options for web application scalability and adoption of NoSQL and Document databases will be more in the year 2011. The following are the my technology picks for .Net stack Windows Azure Windows Azure will be one of the hottest technologies of 2011. Adoption of Cloud and Windows Azure will get big attention on next year. The Windows Azure platform is a flexible cloud–computing platform that lets you focus on solving business problems and addressing customer needs. No need to invest upfront on expensive infrastructure. Pay only for what you use, scale up when you need capacity and pull it back when you don’t. We handle all the patches and maintenance — all in a secure environment with over 99.9% uptime. Silverlight 5 Silverlight is becoming a common technology for variety of development platforms. You can develop Silverlight applications for web, desktop and windows phone. The new Silverlight 5 beta will be available during the starting quarter of the next year with new capabilities and lot of new features. Silverlight 5 will be powerful development platform for both web-based business apps and rich media solutions. We can expect final version of Silverlight 5 on end of 2011. Windows Phone 7 Development Tools Mobile application development will be very hot in year 2011 and Windows Phone 7 will be one of the hottest technologies of next year. You can get introduction on Windows Phone 7 Development Tools from somasegar’s blog post and MSDN documentation available from here. EF Code First I am a big fan of Entity Framework’s Code First approach and hope that Code First approach will attract more people onto Entity Framework 4. EF Code First lets you focus on domain model which will enable Domain-Driven Development for applications. I hope that DDD fans will love the EF Code First approach. The Entity Framework 4 now supports three types of approaches and these will attract different types of developer audience. ASP.NET MVC 3 The ASP.NET MVC 3 will be the hottest technology of Microsoft web stack on the next year. ASP.NET developers will widely move to the ASP.NET MVC Framework from their WebForms development. The new Razor view engine is great and it will increase the adoption of ASP.NET MVC 3. Razor the will improve the productivity when working with ASP.NET MVC 3 Views. You can build great web applications using ASP.NET MVC 3 and jQuery with better maintainability, generation of clean HTML and even better performance. In my opinion, the best technology stack for web development is ASP.NET MVC 3 and Entity Framework 4 Code First as ORM. On the next year, you can expect more articles from my blog on ASP.NET MVC 3 and Entity Framework 4 Code First. RavenDB NoSQL and Document databases will get more attention on the coming year and RavenDB will be the most notable document database in the .NET stack. RavenDB is an Open Source (with a commercial option) document database for the .NET/Windows platform developed by Ayende Rahien. RavenDB is .NET focused document database which comes with a fully functional .NET client API and supports LINQ. I have written few articles on RavenDB and you can read it from here. Managed Extensibility Framework (MEF) Many people didn't realized the power of MEF. The MEF lets you create extensible applications and provides a great solution for the runtime extensibility problem. I hope that .NET developers will more adopt the MEF on the next year for their .NET applications. You can get an excellent introduction on MEF from Anoop Madhusudanan’s blog post MEF or Managed Extensibility Framework – Creating a Zoo and Animals

    Read the article

  • SQL Saturday 27 (Portland, Oregon)

    - by BuckWoody
    I’m sitting in the Seattle airport, waiting for my flight to Silicon Valley California for the SQL Server 2008 R2 Launch Event. By some quirk of nature, they are asking me to Emcee the event – but that’s another post entirely.   I’m reflecting on the SQL Saturday 27 event that was just held in Portland, Oregon this last Saturday. These are not Microsoft-sponsored events – it’s truly the community at work. Think of a big user-group meeting – I mean REALLY big – held in a central location, like at a college (as ours was) or some larger, inexpensive venue like that. Everyone there is volunteering – it’s my own money and time to drive several hours to a hotel for the night, feed myself and present. It’s their own time and money for the folks that organize the event – unless a vendor or two steps in to help. It’s their own time and money for the attendees to drive a long way, spend the night and their Saturday to listen to the speakers. Why do all this?   Because everybody benefits. Every speaker learns something new, meets new people, and reaches a new audience. Every volunteer does the same. And the attendees? Well, it’s pretty obvious what they get. A 7Am to 10PM extravaganza of knowledge from every corner of the product. In fact, this year the Portland group hooked up with the CodeCamp folks and held a combined event. We had over 850 people, and I had everyone from data professionals to developers in my sessions.   So I’ll take this opportunity to do two things: to say “thank you” to all of the folks who attended, from those who spoke to those who worked and those who came to listen, and to challenge you to attend the next SQL Saturday anywhere near you. You can find the list here: http://www.sqlsaturday.com/. Don’t see anything in your area? Start one! The PASS folks have a package that will show you how. Sure, it’s a big job, but the key is to get as many people helping you as possible. Even if you have only a few dozen folks show up the first time, no worries. The first events I presented at had about 20 in the room. But not this week.   See you at the Launch Event if you’re near the San Francisco area tomorrow, and see you at the Redmond SQL Saturday and TechEd if not.   Share this post: email it! | bookmark it! | digg it! | reddit! | kick it! | live it!

    Read the article

  • Consumer Oriented Search In Oracle Endeca Information Discovery – Part 1

    - by Bob Zurek
    Information Discovery, a core capability of Oracle Endeca Information Discovery, enables business users to rapidly search, discover and navigate through a wide variety of big data including structured, unstructured and semi-structured data. One of the key capabilities, among many, that differentiate our solution from others in the Information Discovery market is our deep support for search across this growing amount of varied big data. Our method and approach is very different than classic simple keyword search that is found in may information discovery solutions. In this first part of a series on the topic of search, I will walk you through many of the key capabilities that go beyond the simple search box that you might experience in products where search was clearly an afterthought or attempt to catch up to our core capabilities in this area. Lets explore. The core data management solution of Oracle Endeca Information Discovery is the Endeca Server, a hybrid search-analytical database that his highly scalable and column-oriented in nature. We will talk in more technical detail about the capabilities of the Endeca Server in future blog posts as this post is intended to give you a feel for the deep search capabilities that are an integral part of the Endeca Server. The Endeca Server provides best-of-breed search features aw well as a new class of features that are the first to be designed around the requirement to bridge structured, semi-structured and unstructured big data. Some of the key features of search include type a heads, automatic alphanumeric spell corrections, positional search, Booleans, wildcarding, natural language, and category search and query classification dialogs. This is just a subset of the advanced search capabilities found in Oracle Endeca Information Discovery. Search is an important feature that makes it possible for business users to explore on the diverse data sets the Endeca Server can hold at any one time. The search capabilities in the Endeca server differ from other Information Discovery products with simple “search boxes” in the following ways: The Endeca Server Supports Exploratory Search.  Enterprise data frequently requires the user to explore content through an ad hoc dialog, with guidance that helps them succeed. This has implications for how to design search features. Traditional search doesn’t assume a dialog, and so it uses relevance ranking to get its best guess to the top of the results list. It calculates many relevance factors for each query, like word frequency, distance, and meaning, and then reduces those many factors to a single score based on a proprietary “black box” formula. But how can a business users, searching, act on the information that the document is say only 38.1% relevant? In contrast, exploratory search gives users the opportunity to clarify what is relevant to them through refinements and summaries. This approach has received consumer endorsement through popular ecommerce sites where guided navigation across a broad range of products has helped consumers better discover choices that meet their, sometimes undetermined requirements. This same model exists in Oracle Endeca Information Discovery. In fact, the Endeca Server powers many of the most popular e-commerce sites in the world. The Endeca Server Supports Cascading Relevance. Traditional approaches of search reduce many relevance weights to a single score. This means that if a result with a good title match gets a similar score to one with an exact phrase match, they’ll appear next to each other in a list. But a user can’t deduce from their score why each got it’s ranking, even though that information could be valuable. Oracle Endeca Information Discovery takes a different approach. The Endeca Server stratifies results by a primary relevance strategy, and then breaks ties within a strata by ordering them with a secondary strategy, and so on. Application managers get the explicit means to compose these strategies based on their knowledge of their own domain. This approach gives both business users and managers a deterministic way to set and understand relevance. Now that you have an understanding of two of the core search capabilities in Oracle Endeca Information Discovery, our next blog post on this topic will discuss more advanced features including set search, second-order relevance as well as an understanding of faceted search mechanisms that include queries and filters.  

    Read the article

  • BI&EPM in Focus November 2012

    - by Mike.Hallett(at)Oracle-BI&EPM
    Customers ·       San Diego Unified School District Harnesses Attendance, Procurement, and Operational Data with Oracle Exalytics, Generating $4.4 Million in Savings ·       NilsonGroup chooses Oracle Exalytics In-Memory Machine as their solution to access critical data to keep its stores competitive with real-time Mobile BI:  Video ·       Nykredit, in the Danish Financial Sector, describes their experiences from testing the Exalytics Business Intelligence Machine: Video  ·       Sodexo chose Oracle Exalytics as their business analytics platform:  Video ·       AstraZeneca (US, Canada, MedImmune) Improves Insight, Analytics, and Reporting, Enterprisewide with Unified Planning on a Single Platform ·       Experian Consolidates Reporting Systems for One, Global View of Financial Data and Improves Planning for Continued Growth ·       Munchkin Gives its Line of Children’s Products Plenty of Room to Grow in an Upgraded Enterprise Application Environment ·       Top 20 EPM Customer Snapshots, in One Handy Booklet (link) ·       Customer and Partner Successes: Link to Complete Archive Enterprise Performance Management ·       Nov 15: Is Hope and Email the Core of your Reconciliation Process? (link) ·       Replay: Integrated Business Planning, Featuring Leggett & Platt (link) ·       Whitepaper: The New Competitive Advantage - Strategic CIO's Embrace the Cloud (link) ·       Press: Oracle Enterprise Performance Management Driving Significant Improvements in Budget Management and Reporting for Public Sector Organizations (link) ·       Enterprise Performance Management Video Feature Overviews, Now Available on YouTube (link) ·       NEW Solution Brief - Oracle Hyperion Planning on the Oracle Exalytics In-Memory Machine (link) ·       For Insurance sector: Datasheet for new release V2.0 - Oracle Quantitative Management and Reporting for Solvency II (link) ·       Whitepaper FSN 2012: Managing Risk and Uncertainty, an Executive's Guide to Integrated Business Planning (link) ·       NEW Datasheet for Oracle Planning and Budgeting Cloud Service (link) ·       Blog: Planning in the Cloud - For Real Business Intelligence ·       Press: Latest Release of Oracle Exalytics In-Memory Machine Software Enables Customers to View and Analyze Data at the Speed of Business (link) ·       Press: New Release (11.1.1.6.2BP1) of Oracle Business Intelligence Enables Users to Quickly Access and Analyze Key Business Information, Anytime, Anywhere (link) ·       Mark Hurd Interviewed on USA Today about Big Data & Analytics (link) ·       Whitepaper: Mastering Big Data - CFO Strategies to Transform Insight into Opportunity (link) ·       Nov 15: Improve Asset Utilization. Achieve Greater Profitability: Oracle Enterprise Asset Management Analytics (link) ·       Replay: Oracle Enterprise Asset Mgmt Analytics and Oracle Manufacturing Analytics (link) ·       Overload to Impact: An Industry Scorecard on Big Data Business Challenges (link) ·       Webcast Replay: Overview of Oracle Endeca Informational Discovery (link) ·       OBIEE 11g: Required and Recommended Patches and Patch Sets (link)

    Read the article

  • SEO/Google: How should I handle multiple countries and domains?

    - by Valorized
    Hello. I'm the webmaster of an online shop based in Austria (Europe). Therefore we registered "example.at". We also own different other domain names like "example-shop.com" and "example.info". Currently all those domains are redirected (301) to the .at one. Still available is: "example.net" and "example.org" (and .ws/.cc), unfortunately not available: .de/.eu The .com is currently owned by one of our partners, the contract ends in 2012 but until then we have no chance to get this one. Recently I read more about geo-targeting and I noticed ONE big deal. The tld ".at" is hardly recognised in Germany (google.de) whereas it is excellently listed in Austria (google.at). As a result of the .at I cannot set the target location manually (or to unlisted). More info: https://www.google.com/support/webmasters/bin/answer.py?answer=62399&hl=en This is a big problem. I looked at Google Analytics and - although Germany is 10x as big as Austria - there are more visits from Austria. So, how should I config the domain in order to get the best results in both, Germany and Austria? I thought of some solutions: First I could stop redirecting the .info. Then there would be a duplicate of the .at one. Moreover, in Webmastertools, I could set the target location of the .info to Germany. As the .at still targets Austria, both would be targeted - however I don't now if google punishes one of them because of the duplicate content? Same as 1. but with .net or .org (I think .info is not a "nice" domain and moreover I think search engines prefer .com, .net or .org to .info). Same as 1. (or 2.) but with a rel="canonical" on the new one (pointing to the .at). Con: I don't think this will improve the situation, because it still tells google that the .at one is more important, like: "if .info points to .at, the target may still be Austria". rel="canonical" on the .at pointing to the new (.info or .net or .org). However I fear that this will have a negative impact on the listing on google.at because: "Hey, the well-known .at is not important anymore, so let's focus on the .info which is not well-known." - Therefore: bad position in search results. Redirect .at to the new (.info or .net or .org) with a 301-Redirect. Con: Might be worse than 4, we might loose Page-Rank (or "the value of the page", because google says that page rank is not important anymore). Moreover this might be even more confusing for the customers. In 3. or 4. customers don't get redirected, they do not see the canonical-meta-tag. So, dear experts, please tell me what the best option would be! Thank you very much for your advice in advance and please excuse the long question. I really appreciate this network! Please note: It's exactly the same content AND language. In Austria we speak German.

    Read the article

  • How to find and fix performance problems in ORM powered applications

    - by FransBouma
    Once in a while we get requests about how to fix performance problems with our framework. As it comes down to following the same steps and looking into the same things every single time, I decided to write a blogpost about it instead, so more people can learn from this and solve performance problems in their O/R mapper powered applications. In some parts it's focused on LLBLGen Pro but it's also usable for other O/R mapping frameworks, as the vast majority of performance problems in O/R mapper powered applications are not specific for a certain O/R mapper framework. Too often, the developer looks at the wrong part of the application, trying to fix what isn't a problem in that part, and getting frustrated that 'things are so slow with <insert your favorite framework X here>'. I'm in the O/R mapper business for a long time now (almost 10 years, full time) and as it's a small world, we O/R mapper developers know almost all tricks to pull off by now: we all know what to do to make task ABC faster and what compromises (because there are almost always compromises) to deal with if we decide to make ABC faster that way. Some O/R mapper frameworks are faster in X, others in Y, but you can be sure the difference is mainly a result of a compromise some developers are willing to deal with and others aren't. That's why the O/R mapper frameworks on the market today are different in many ways, even though they all fetch and save entities from and to a database. I'm not suggesting there's no room for improvement in today's O/R mapper frameworks, there always is, but it's not a matter of 'the slowness of the application is caused by the O/R mapper' anymore. Perhaps query generation can be optimized a bit here, row materialization can be optimized a bit there, but it's mainly coming down to milliseconds. Still worth it if you're a framework developer, but it's not much compared to the time spend inside databases and in user code: if a complete fetch takes 40ms or 50ms (from call to entity object collection), it won't make a difference for your application as that 10ms difference won't be noticed. That's why it's very important to find the real locations of the problems so developers can fix them properly and don't get frustrated because their quest to get a fast, performing application failed. Performance tuning basics and rules Finding and fixing performance problems in any application is a strict procedure with four prescribed steps: isolate, analyze, interpret and fix, in that order. It's key that you don't skip a step nor make assumptions: these steps help you find the reason of a problem which seems to be there, and how to fix it or leave it as-is. Skipping a step, or when you assume things will be bad/slow without doing analysis will lead to the path of premature optimization and won't actually solve your problems, only create new ones. The most important rule of finding and fixing performance problems in software is that you have to understand what 'performance problem' actually means. Most developers will say "when a piece of software / code is slow, you have a performance problem". But is that actually the case? If I write a Linq query which will aggregate, group and sort 5 million rows from several tables to produce a resultset of 10 rows, it might take more than a couple of milliseconds before that resultset is ready to be consumed by other logic. If I solely look at the Linq query, the code consuming the resultset of the 10 rows and then look at the time it takes to complete the whole procedure, it will appear to me to be slow: all that time taken to produce and consume 10 rows? But if you look closer, if you analyze and interpret the situation, you'll see it does a tremendous amount of work, and in that light it might even be extremely fast. With every performance problem you encounter, always do realize that what you're trying to solve is perhaps not a technical problem at all, but a perception problem. The second most important rule you have to understand is based on the old saying "Penny wise, Pound Foolish": the part which takes e.g. 5% of the total time T for a given task isn't worth optimizing if you have another part which takes a much larger part of the total time T for that same given task. Optimizing parts which are relatively insignificant for the total time taken is not going to bring you better results overall, even if you totally optimize that part away. This is the core reason why analysis of the complete set of application parts which participate in a given task is key to being successful in solving performance problems: No analysis -> no problem -> no solution. One warning up front: hunting for performance will always include making compromises. Fast software can be made maintainable, but if you want to squeeze as much performance out of your software, you will inevitably be faced with the dilemma of compromising one or more from the group {readability, maintainability, features} for the extra performance you think you'll gain. It's then up to you to decide whether it's worth it. In almost all cases it's not. The reason for this is simple: the vast majority of performance problems can be solved by implementing the proper algorithms, the ones with proven Big O-characteristics so you know the performance you'll get plus you know the algorithm will work. The time taken by the algorithm implementing code is inevitable: you already implemented the best algorithm. You might find some optimizations on the technical level but in general these are minor. Let's look at the four steps to see how they guide us through the quest to find and fix performance problems. Isolate The first thing you need to do is to isolate the areas in your application which are assumed to be slow. For example, if your application is a web application and a given page is taking several seconds or even minutes to load, it's a good candidate to check out. It's important to start with the isolate step because it allows you to focus on a single code path per area with a clear begin and end and ignore the rest. The rest of the steps are taken per identified problematic area. Keep in mind that isolation focuses on tasks in an application, not code snippets. A task is something that's started in your application by either another task or the user, or another program, and has a beginning and an end. You can see a task as a piece of functionality offered by your application.  Analyze Once you've determined the problem areas, you have to perform analysis on the code paths of each area, to see where the performance problems occur and which areas are not the problem. This is a multi-layered effort: an application which uses an O/R mapper typically consists of multiple parts: there's likely some kind of interface (web, webservice, windows etc.), a part which controls the interface and business logic, the O/R mapper part and the RDBMS, all connected with either a network or inter-process connections provided by the OS or other means. Each of these parts, including the connectivity plumbing, eat up a part of the total time it takes to complete a task, e.g. load a webpage with all orders of a given customer X. To understand which parts participate in the task / area we're investigating and how much they contribute to the total time taken to complete the task, analysis of each participating task is essential. Start with the code you wrote which starts the task, analyze the code and track the path it follows through your application. What does the code do along the way, verify whether it's correct or not. Analyze whether you have implemented the right algorithms in your code for this particular area. Remember we're looking at one area at a time, which means we're ignoring all other code paths, just the code path of the current problematic area, from begin to end and back. Don't dig in and start optimizing at the code level just yet. We're just analyzing. If your analysis reveals big architectural stupidity, it's perhaps a good idea to rethink the architecture at this point. For the rest, we're analyzing which means we collect data about what could be wrong, for each participating part of the complete application. Reviewing the code you wrote is a good tool to get deeper understanding of what is going on for a given task but ultimately it lacks precision and overview what really happens: humans aren't good code interpreters, computers are. We therefore need to utilize tools to get deeper understanding about which parts contribute how much time to the total task, triggered by which other parts and for example how many times are they called. There are two different kind of tools which are necessary: .NET profilers and O/R mapper / RDBMS profilers. .NET profiling .NET profilers (e.g. dotTrace by JetBrains or Ants by Red Gate software) show exactly which pieces of code are called, how many times they're called, and the time it took to run that piece of code, at the method level and sometimes even at the line level. The .NET profilers are essential tools for understanding whether the time taken to complete a given task / area in your application is consumed by .NET code, where exactly in your code, the path to that code, how many times that code was called by other code and thus reveals where hotspots are located: the areas where a solution can be found. Importantly, they also reveal which areas can be left alone: remember our penny wise pound foolish saying: if a profiler reveals that a group of methods are fast, or don't contribute much to the total time taken for a given task, ignore them. Even if the code in them is perhaps complex and looks like a candidate for optimization: you can work all day on that, it won't matter.  As we're focusing on a single area of the application, it's best to start profiling right before you actually activate the task/area. Most .NET profilers support this by starting the application without starting the profiling procedure just yet. You navigate to the particular part which is slow, start profiling in the profiler, in your application you perform the actions which are considered slow, and afterwards you get a snapshot in the profiler. The snapshot contains the data collected by the profiler during the slow action, so most data is produced by code in the area to investigate. This is important, because it allows you to stay focused on a single area. O/R mapper and RDBMS profiling .NET profilers give you a good insight in the .NET side of things, but not in the RDBMS side of the application. As this article is about O/R mapper powered applications, we're also looking at databases, and the software making it possible to consume the database in your application: the O/R mapper. To understand which parts of the O/R mapper and database participate how much to the total time taken for task T, we need different tools. There are two kind of tools focusing on O/R mappers and database performance profiling: O/R mapper profilers and RDBMS profilers. For O/R mapper profilers, you can look at LLBLGen Prof by hibernating rhinos or the Linq to Sql/LLBLGen Pro profiler by Huagati. Hibernating rhinos also have profilers for other O/R mappers like NHibernate (NHProf) and Entity Framework (EFProf) and work the same as LLBLGen Prof. For RDBMS profilers, you have to look whether the RDBMS vendor has a profiler. For example for SQL Server, the profiler is shipped with SQL Server, for Oracle it's build into the RDBMS, however there are also 3rd party tools. Which tool you're using isn't really important, what's important is that you get insight in which queries are executed during the task / area we're currently focused on and how long they took. Here, the O/R mapper profilers have an advantage as they collect the time it took to execute the query from the application's perspective so they also collect the time it took to transport data across the network. This is important because a query which returns a massive resultset or a resultset with large blob/clob/ntext/image fields takes more time to get transported across the network than a small resultset and a database profiler doesn't take this into account most of the time. Another tool to use in this case, which is more low level and not all O/R mappers support it (though LLBLGen Pro and NHibernate as well do) is tracing: most O/R mappers offer some form of tracing or logging system which you can use to collect the SQL generated and executed and often also other activity behind the scenes. While tracing can produce a tremendous amount of data in some cases, it also gives insight in what's going on. Interpret After we've completed the analysis step it's time to look at the data we've collected. We've done code reviews to see whether we've done anything stupid and which parts actually take place and if the proper algorithms have been implemented. We've done .NET profiling to see which parts are choke points and how much time they contribute to the total time taken to complete the task we're investigating. We've performed O/R mapper profiling and RDBMS profiling to see which queries were executed during the task, how many queries were generated and executed and how long they took to complete, including network transportation. All this data reveals two things: which parts are big contributors to the total time taken and which parts are irrelevant. Both aspects are very important. The parts which are irrelevant (i.e. don't contribute significantly to the total time taken) can be ignored from now on, we won't look at them. The parts which contribute a lot to the total time taken are important to look at. We now have to first look at the .NET profiler results, to see whether the time taken is consumed in our own code, in .NET framework code, in the O/R mapper itself or somewhere else. For example if most of the time is consumed by DbCommand.ExecuteReader, the time it took to complete the task is depending on the time the data is fetched from the database. If there was just 1 query executed, according to tracing or O/R mapper profilers / RDBMS profilers, check whether that query is optimal, uses indexes or has to deal with a lot of data. Interpret means that you follow the path from begin to end through the data collected and determine where, along the path, the most time is contributed. It also means that you have to check whether this was expected or is totally unexpected. My previous example of the 10 row resultset of a query which groups millions of rows will likely reveal that a long time is spend inside the database and almost no time is spend in the .NET code, meaning the RDBMS part contributes the most to the total time taken, the rest is compared to that time, irrelevant. Considering the vastness of the source data set, it's expected this will take some time. However, does it need tweaking? Perhaps all possible tweaks are already in place. In the interpret step you then have to decide that further action in this area is necessary or not, based on what the analysis results show: if the analysis results were unexpected and in the area where the most time is contributed to the total time taken is room for improvement, action should be taken. If not, you can only accept the situation and move on. In all cases, document your decision together with the analysis you've done. If you decide that the perceived performance problem is actually expected due to the nature of the task performed, it's essential that in the future when someone else looks at the application and starts asking questions you can answer them properly and new analysis is only necessary if situations changed. Fix After interpreting the analysis results you've concluded that some areas need adjustment. This is the fix step: you're actively correcting the performance problem with proper action targeted at the real cause. In many cases related to O/R mapper powered applications it means you'll use different features of the O/R mapper to achieve the same goal, or apply optimizations at the RDBMS level. It could also mean you apply caching inside your application (compromise memory consumption over performance) to avoid unnecessary re-querying data and re-consuming the results. After applying a change, it's key you re-do the analysis and interpretation steps: compare the results and expectations with what you had before, to see whether your actions had any effect or whether it moved the problem to a different part of the application. Don't fall into the trap to do partly analysis: do the full analysis again: .NET profiling and O/R mapper / RDBMS profiling. It might very well be that the changes you've made make one part faster but another part significantly slower, in such a way that the overall problem hasn't changed at all. Performance tuning is dealing with compromises and making choices: to use one feature over the other, to accept a higher memory footprint, to go away from the strict-OO path and execute queries directly onto the RDBMS, these are choices and compromises which will cross your path if you want to fix performance problems with respect to O/R mappers or data-access and databases in general. In most cases it's not a big issue: alternatives are often good choices too and the compromises aren't that hard to deal with. What is important is that you document why you made a choice, a compromise: which analysis data, which interpretation led you to the choice made. This is key for good maintainability in the years to come. Most common performance problems with O/R mappers Below is an incomplete list of common performance problems related to data-access / O/R mappers / RDBMS code. It will help you with fixing the hotspots you found in the interpretation step. SELECT N+1: (Lazy-loading specific). Lazy loading triggered performance bottlenecks. Consider a list of Orders bound to a grid. You have a Field mapped onto a related field in Order, Customer.CompanyName. Showing this column in the grid will make the grid fetch (indirectly) for each row the Customer row. This means you'll get for the single list not 1 query (for the orders) but 1+(the number of orders shown) queries. To solve this: use eager loading using a prefetch path to fetch the customers with the orders. SELECT N+1 is easy to spot with an O/R mapper profiler or RDBMS profiler: if you see a lot of identical queries executed at once, you have this problem. Prefetch paths using many path nodes or sorting, or limiting. Eager loading problem. Prefetch paths can help with performance, but as 1 query is fetched per node, it can be the number of data fetched in a child node is bigger than you think. Also consider that data in every node is merged on the client within the parent. This is fast, but it also can take some time if you fetch massive amounts of entities. If you keep fetches small, you can use tuning parameters like the ParameterizedPrefetchPathThreshold setting to get more optimal queries. Deep inheritance hierarchies of type Target Per Entity/Type. If you use inheritance of type Target per Entity / Type (each type in the inheritance hierarchy is mapped onto its own table/view), fetches will join subtype- and supertype tables in many cases, which can lead to a lot of performance problems if the hierarchy has many types. With this problem, keep inheritance to a minimum if possible, or switch to a hierarchy of type Target Per Hierarchy, which means all entities in the inheritance hierarchy are mapped onto the same table/view. Of course this has its own set of drawbacks, but it's a compromise you might want to take. Fetching massive amounts of data by fetching large lists of entities. LLBLGen Pro supports paging (and limiting the # of rows returned), which is often key to process through large sets of data. Use paging on the RDBMS if possible (so a query is executed which returns only the rows in the page requested). When using paging in a web application, be sure that you switch server-side paging on on the datasourcecontrol used. In this case, paging on the grid alone is not enough: this can lead to fetching a lot of data which is then loaded into the grid and paged there. Keep note that analyzing queries for paging could lead to the false assumption that paging doesn't occur, e.g. when the query contains a field of type ntext/image/clob/blob and DISTINCT can't be applied while it should have (e.g. due to a join): the datareader will do DISTINCT filtering on the client. this is a little slower but it does perform paging functionality on the data-reader so it won't fetch all rows even if the query suggests it does. Fetch massive amounts of data because blob/clob/ntext/image fields aren't excluded. LLBLGen Pro supports field exclusion for queries. You can exclude fields (also in prefetch paths) per query to avoid fetching all fields of an entity, e.g. when you don't need them for the logic consuming the resultset. Excluding fields can greatly reduce the amount of time spend on data-transport across the network. Use this optimization if you see that there's a big difference between query execution time on the RDBMS and the time reported by the .NET profiler for the ExecuteReader method call. Doing client-side aggregates/scalar calculations by consuming a lot of data. If possible, try to formulate a scalar query or group by query using the projection system or GetScalar functionality of LLBLGen Pro to do data consumption on the RDBMS server. It's far more efficient to process data on the RDBMS server than to first load it all in memory, then traverse the data in-memory to calculate a value. Using .ToList() constructs inside linq queries. It might be you use .ToList() somewhere in a Linq query which makes the query be run partially in-memory. Example: var q = from c in metaData.Customers.ToList() where c.Country=="Norway" select c; This will actually fetch all customers in-memory and do an in-memory filtering, as the linq query is defined on an IEnumerable<T>, and not on the IQueryable<T>. Linq is nice, but it can often be a bit unclear where some parts of a Linq query might run. Fetching all entities to delete into memory first. To delete a set of entities it's rather inefficient to first fetch them all into memory and then delete them one by one. It's more efficient to execute a DELETE FROM ... WHERE query on the database directly to delete the entities in one go. LLBLGen Pro supports this feature, and so do some other O/R mappers. It's not always possible to do this operation in the context of an O/R mapper however: if an O/R mapper relies on a cache, these kind of operations are likely not supported because they make it impossible to track whether an entity is actually removed from the DB and thus can be removed from the cache. Fetching all entities to update with an expression into memory first. Similar to the previous point: it is more efficient to update a set of entities directly with a single UPDATE query using an expression instead of fetching the entities into memory first and then updating the entities in a loop, and afterwards saving them. It might however be a compromise you don't want to take as it is working around the idea of having an object graph in memory which is manipulated and instead makes the code fully aware there's a RDBMS somewhere. Conclusion Performance tuning is almost always about compromises and making choices. It's also about knowing where to look and how the systems in play behave and should behave. The four steps I provided should help you stay focused on the real problem and lead you towards the solution. Knowing how to optimally use the systems participating in your own code (.NET framework, O/R mapper, RDBMS, network/services) is key for success as well as knowing what's going on inside the application you built. I hope you'll find this guide useful in tracking down performance problems and dealing with them in a useful way.  

    Read the article

  • OutOfMemoryException, large Private Data

    - by Captain Comic
    Hello, In previous series: http://stackoverflow.com/questions/2543648/outofmemoryexception-stack-size-is-huge-large-number-of-threads I have a .net windows service that consumes a lot of memory. The GC heap is not big. Also the stack size is not big. What is big is something called a private data. Also I can see in task manager that my application consumes a lot something that taskmanager calls a handle. My application consumes 2326 handles. I believe that these handles are some windows handles that occupy private data. I can see that this private data is occupied by blocks marked as Thread Environment Block. What is that? Screenshot of my application memory usage by VMMap Screenshot of my application memory usage by Task Manager UPDATE I run ProcessExplorer. I have two instances of my service running at the moment. I can see that they consume a lot of virtual memory for Gen2 GC. This look suspicios. Also total reserved for GC Heap size is the same for two processes.

    Read the article

  • Get JVM to grow memory demand as needed up to size of VM limit?

    - by Ira Baxter
    We ship a Java application whose memory demand can vary quite a lot depending on the size of the data it is processing. If you don't set the max VM (virtual memory) size, quite often the JVM quits with an GC failure on big data. What we'd like to see, is the JVM requesting more memory, as GC fails to provide enough, until the total available VM is exhausted. e.g., start with 128Mb, and increase geometrically (or some other step) whenever the GC failed. The JVM ("Java") command line allows explicit setting of max VM sizes (various -Xm* commands), and you'd think that would be designed to be adequate. We try to do this in a .cmd file that we ship with the application. But if you pick any specific number, you get one of two bad behaviors: 1) if your number is small enough to work on most target systems (e.g., 1Gb), it isn't big enough for big data, or 2) if you make it very large, the JVM refuses to run on those systems whose actual VM is smaller than specified. How does one set up Java to use the available VM when needed, without knowing that number in advance, and without grabbing it all on startup?

    Read the article

  • Swing layout with miglayout and nested panels

    - by Jazzepi
    I have a Swing program using SwingLayout. The structure of the Swing components looks like this. JFrame JPanel (Cardlayout) JPanel (Miglayout) - Main panel Jpanel (Flowlayout) - Checkbox Panel JPanel (Flowlayout) - Option Panel My problem right now is that I'm not sure how to prevent the checkbox panel from growing. I don't want it to "push" out the column that it's in to the right. I want Wraplayout (which works fine on the option panel) to wrap the content of the checkbox panel when it grows too big. This is the view of it from the Windowsbuilder GUI inside of Eclipse. The panel with the label "sites" in it is the Checkbox Panel. Directly to the right of the Checkbox panel is the Option panel. The big panel containing both of them is the main panel. http://i.imgur.com/S7Njo.png This is what happens when I add too many checkboxes http://i.imgur.com/f2SZz.png My main problem is that I don't understand why setting "grow 0" on the column constraint for the first column in mainpanel doesn't prevent it from growing when the component inside gets too big as I add new checkboxes to the site panel (the site panel can have an arbitrary number of checkboxes). This is my miglayout for the main panel. mainPanel.setLayout(new MigLayout("", "[][grow]", "[][][][][][][][grow]")); Here are my component constraints for when I add the checkbox panel siteCheckBoxPanel = new JPanel(); mainPanel.add(siteCheckBoxPanel, "cell 0 0,alignx left,gapx 0px,aligny center"); siteCheckBoxPanel.setLayout(new FlowLayout(FlowLayout.CENTER, 5, 5)); I've also tried it without flow layout, and that doesn't fix anything. Any insight you could provide would be great! I'm also happy to provide more information if people have questions. FYI I've also tried "grow 0" on both the column and row constraint for the cell that the Checkbox Panel is inside.

    Read the article

  • Is a VCS appropriate for usage by a designer?

    - by iconiK
    I know that a VCS is absolutely critical for a developer to increase productivity and protect the code, no doubts about it. But what about a designer, using say, Photoshop (though it's not specific to any tools, just to make my point clearer). VCSs uses delta compression to store different versions of files. This works very well for code, but for images, that's a problem. Raster image files are binary formats, though vector image files are text (SVG comes to my mind) and pose to problem. The problem comes with .psd files (and any other image "source" file) - those can get pretty big and since I'm not familiar with the format, I'll consider them as binary files. How would a VCS work in this condition? The repository could be pretty darned big if the VCS server isn't able to diff the files efficiently (or worse, not at all) and over time this can become a really big pain when someone needs to check out the repository (or clone it if using a DVCS). Have any of you used a VCS for this purpose? How well does it work? I'm mostly interested in Mercurial, though this is a general situation that applies to any VCS.

    Read the article

  • Winforms: How to speed up Invalidate()?

    - by Pedery
    I'm developing a retained mode drawing application in GDI+. The application can draw simple shapes to a canvas and perform basic editing. The math that does this is optimized to the last byte and is not an issue. I'm drawing on a panel that is using the built-in Controlstyles.DoubleBuffer. Now, my problem arises if I run my app maximized on a big monitor (HD in my case). If I try to draw a line from one corner of the (big) canvas to the diagonally oposite other, it will start to lag and the CPU goes high up. Each graphical object in my app has a boundingbox. Thus, when I invalidate the boundingbox of a line that goes from one corner of the maximized app to the oposite diagonal one, that boundingbox is virtually as big as the canvas. When a user is drawing a line, this invalidation of the boundingbox thus happens on the mousemove event, and there is a clear lag visible. This lag also exists if the line is the only object on the canvas. I've tried to optimize this in many ways. If I draw a shorter line, the CPU and the lag goes down. If I remove the Invalidate() and keep all other code, the app is quick. If I use a Region (that only spans the figure) to invalidate instead of the boundingbox, it is just as slow. If I split the boundingbox into a range of smaller boxes that lie back to back, thus reducing the invalidation area, no visible performance gain can be seen. Thus I'm at a loss here. How can I speed up the invalidation? On a side note, both Paint.Net and Mspaint suffers from the same shortcommings. Word and PowerPoint however, seem to be able to paint a line as described above with no lag and no CPU load at all. Thus it's possible to achieve the desired results, the question is how?

    Read the article

  • Android Layout: Display as much ImageViews as possible without scrolling

    - by Toni4780
    I am working on an app which should display several same size images on the screen. But it should only display only so much images as possible without offering scrolling. E.g. On a "big" tablet it could display 10x10 Imageviews (screen is large, so there is much space for pictures) On a "big" phone there might be enough space to display 6x6 ImageViews, so it should only display a 6x6 array of images. On a small phone there is propably only space for 4x4 ImageViews, so it should only display this. How can I make this in Android? I know about "layout-large", ... but if i make a special fixed xml-layout for a "large" device, it would not fit all devices correct. E.g. a Galaxy Nexus is a "normal" device and so is a Nexus One, but there would be at least be space for one or two more imageview rows on a Galaxy Nexus than on a Nexus One. So do I have to measure in code somehow how big the resolution is and display some TableRows accordingly? Or is there a special way how I can manage this?

    Read the article

  • Query optimization (OR based)

    - by john194
    I have googled but I can't find answers for these questions. Your advice is appreciated. centOS on vps with 512MB RAM, nginx, php5 (fastcgi), mysql5 (myisam, not innodb). I need to optimize this app created by some ex-employee. This app is working, but it's slow. Table: t1(id[bigint(20)],c1[mediumtext],c2[mediumtext],c3[mediumtext],c4[mediumtext]) id is some random big number, and is PK Those mediumtext rows look like this: c1="|box-002877|" c2="|ct-2348|rd-11124854|hw-3949|wd-8872|hw-119037736|...etc.. " c3="|fg-2448|wd-11172|hw-1656|...etc.. " c4="|hg-2448|qd-16667|...etc." (some columns contain a lot of data, around 900 KiB, database around 300 MiB) Yes, mediumtext "is bad", and (20) is too big... but I didn't create this. Those codes can be found on any of those 4 mediumtext's... //he needs all the columns of the row containing $code, so he wrote this: function f1($code) { SELECT * FROM t1 WHERE c1 LIKE '%$code%' OR c2 LIKE '%$code%' OR c3 LIKE '%$code%' OR c4 LIKE '%$code%'; Questions: Q1. If $code is found on c1... mysql automatically stops checking and returns row=id+c1+c2+c3+c4? or it will continue (wasting time) checking c2, c3 and c4?... Q2. Mysql is working with this table on disk (not RAM) because of the mediumtext, right? is this the primary cause of slowness? Q3. That query can be cached by mysql (if using a big query_cache_size=128M value on the my.cnf)? or that's not cacheable due to the mediumtexts, or due to the "OR LIKE"...? Q4. Do you recommend rewriting this with mysql's INSTR() / LOCATE() / MATCH..AGAINST [FULLTEXT]?

    Read the article

  • New NSData with range of old NSData maintaining bytes.

    - by umop
    I have a fairly large NSData (or NSMutableData if necessary) object which I want to take a small chunk out of and leave the rest. Since I'm working with large amounts of NSData bytes, I don't want to make a big copy, but instead just truncate the existing bytes. Basically: NSData *source: < a few bytes I want to discard + < big chunk of bytes I want to keep NSData *destination: < big chunk of bytes I want to keep There are truncation methods in NSMutableData, but they only truncate the end of it, whereas I want to truncate the beginning. My thoughts are to do this with the methods: - getBytes:range: and - initWithBytesNoCopy:length:freeWhenDone: However, I'm trying to figure out how to manage memory with these. I'm guessing the process will be like this (I've placed ????s where I don't know what to do): void *buffer // Get range of bytes [source getBytes:buffer range:NSMakeRange(myStart, myLength)]; // Somehow (m)alloc the memory which will be freed up in the following step ????? // Release the source, now that I've allocated the bytes [source release]; // Create a new data, recycling the bytes so they don't have to be copied NSData destination = [[NSData alloc] initWithBytesNoCopy:buffer length:myLength freeWhenDone:YES]; Thanks for the help!

    Read the article

  • What Scheme Does Ghuloum Use?

    - by Don Wakefield
    I'm trying to work my way through Compilers: Backend to Frontend (and Back to Front Again) by Abdulaziz Ghuloum. It seems abbreviated from what one would expect in a full course/seminar, so I'm trying to fill in the pieces myself. For instance, I have tried to use his testing framework in the R5RS flavor of DrScheme, but it doesn't seem to like the macro stuff: src/ghuloum/tests/tests-driver.scm:6:4: read: illegal use of open square bracket I've read his intro paper on the course, An Incremental Approach to Compiler Construction, which gives a great overview of the techniques used, and mentions a couple of Schemes with features one might want to implement for 'extra credit', but he doesn't mention the Scheme he uses in the course. Update I'm still digging into the original question (investigating options such as Petit Scheme suggested by Eli below), but found an interesting link relating to Gholoum's work, so I am including it here. [Ikarus Scheme](http://en.wikipedia.org/wiki/Ikarus_(Scheme_implementation)) is the actual implementation of Ghuloum's ideas, and appears to have been part of his Ph.D. work. It's supposed to be one of the first implementations of R6RS. I'm trying to install Ikarus now, but the configure script doesn't want to recognize my system's install of libgmp.so, so my problems are still unresolved. Example The following example seems to work in PLT 2.4.2 running in DrEd using the Pretty Big (require lang/plt-pretty-big) (load "/Users/donaldwakefield/ghuloum/tests/tests-driver.scm") (load "/Users/donaldwakefield/ghuloum/tests/tests-1.1-req.scm") (define (emit-program x) (unless (integer? x) (error "---")) (emit " .text") (emit " .globl scheme_entry") (emit " .type scheme_entry, @function") (emit "scheme_entry:") (emit " movl $~s, %eax" x) (emit " ret") ) Attempting to replace the require directive with #lang scheme results in the error message foo.scm:7:3: expand: unbound identifier in module in: emit which appears to be due to a failure to load tests-driver.scm. Attempting to use #lang r6rs disables the REPL, which I'd really like to use, so I'm going to try to continue with Pretty Big. My thanks to Eli Barzilay for his patient help.

    Read the article

  • ECM (Niche Vs Mass Market)

    - by Luj Reyes
    Hi Everyone, I recently started a little company with a couple of guys. Ours is the typical startup, a lot of ideas, dreams, talent and work hours :P. Our initial business plan was to develop a DM (Document Manager) with several features found on DropBox and other tools but with a big differentiator. Then we got in the team this Business Guy (I must say that several of us could be called 'Business Guys' but we are mainly hackers, he is just Another 'Networking Guy'), and along with him came this market analysis for a DM aimed at a very specific and narrow niche. We have many elements to believe in his market study and the idea is the classic "The market is X million, so if we grab a 10%...", and the market is really there to grab because all big providers deemed it too little and fled, let's say that the market is 5 million USD and demand very specific features. If we decide to go for this niche product we face a sales cycle of about 7 months, and the main goal of these revenue is to develop more ambitious projects. (Institutional VC is out of the question if you want to keep a marginal ownership of your company in my country). The only overlap between the niche and the mass market product features is the ability to store documents; everything else requires that we focus all of our efforts towards one or the other. I've studied a lot about the differences between Mass and Niche Markets, but I want to hear from people with actual experience. So everything comes down to this: If you have a really “saleable” idea what is the right thing to do: to go for the niche or go for the big prize and target primarily the mass market? Thanks for your input

    Read the article

  • Finding a picture in a picture with java?

    - by tarrasch
    what i want to to is analyse input from screen in form of pictures. I want to be able to identify a part of an image in a bigger image and get its coordinates within the bigger picture. Example: would have to be located in And the result would be the upper right corner of the picture in the big picture and the lower left of the part in the big picture. As you can see, the white part of the picture is irrelevant, what i basically need is just the green frame. Is there a library that can do something like this for me? Runtime is not really an issue. What i want to do with this is just generating a few random pixel coordinates and recognize the color in the big picture at that position, to recognize the green box fast later. And how would it decrease performance, if the white box in the middle is transparent? The question has been asked several times on SO as it seems without a single answer. I found i found a solution at http://werner.yellowcouch.org/Papers/subimg/index.html . Unfortunately its in C++ and i do not understand a thing. Would be nice to have a Java implementation on SO.

    Read the article

  • Unable to access Java-created file -- sometimes

    - by BlairHippo
    In Java, I'm working with code running under WinXP that creates a file like this: public synchronized void store(Properties props, byte[] data) { try { File file = filenameBasedOnProperties(props); if ( file.exists() ) { return; } File temp = File.createTempFile("tempfile", null); FileOutputStream out = new FileOutputStream(temp); out.write(data); out.flush(); out.close(); file.getParentFile().mkdirs(); temp.renameTo(file); } catch (IOException ex) { // Complain and whine and stuff } } Sometimes, when a file is created this way, it's just about totally inaccessible from outside the code (though the code responsible for opening and reading the file has no problem), even when the application isn't running. When accessed via Windows Explorer, I can't move, rename, delete, or even open the file. Under Cygwin, I get the following when I ls -l the directory: ls: cannot access [big-honkin-filename] total 0 ?????????? ? ? ? ? ? [big-honkin-filename] As implied, the filenames are big, but under the 260-character max for XP (though they are slightly over 200 characters). To further add to the sense the my computer just wants me to feel stupid, sometimes the files created by this code are perfectly normal. The only pattern I've spotted is that once one file in the directory "locks", the rest are screwed. Anybody ever run into something like this before, or have any insights into what's going on here?

    Read the article

  • Table cell doesn't obey vertical-align CSS declaration when it contains a floated element

    - by mikez302
    I am trying to create a table, where each cell contains a big floated h1 on the left side, and a larger amount of small text to the right of the big text, vertically centered. However, the small text is showing up at the top of each cell, despite that it has a "vertical-align: middle" declaration. When I remove the big floated element, everything looks fine. I tested it in recent versions of IE, Firefox, and Safari, and this happened in every case. Why is this happening? Does anyone know of a way around it? Here is an example: <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd"> <html><head> <meta http-equiv='Content-Type' content='text/html; charset=UTF-8'> <title>vertical-align test</title> <style type="text/css"> td { border: solid black 1px; vertical-align: middle; font-size: 12px} h1 { font-size: 40px; float: left} </style> </head> <body> <table> <tr> <td><h1>1</h1>The quick brown fox jumps over the lazy dog.</td> <td>The quick brown fox jumps over the lazy dog.</td> </tr> </table> </body></html> Notice that the small text in the first cell is at the top for some reason, but the text in the 2nd cell is vertically centered.

    Read the article

  • Do you use Styrofoam balls to model your systems?

    - by Nick D
    [Objective-C] Do you still use Styrofoam balls to model your systems, where each ball represents a class? Tom Love: We do, actually. We've also done a 3D animation version of it, which we found to be nowhere near as useful as the Styrofoam balls. There's something about a physical, conspicuous structure hanging from the ceiling right in the middle of a development project that's regularly updated to provide not only the structure of the system that you're building, but also the current status of each one of the classes. We've done it on 19 projects the last time I've counted. One of them was 1,856 classes, which is big - actually, probably bigger than it should be. It was a big commercial project, so it needed to be somewhat big. Masterminds of Programming It is the first time I've read or heard about using styrofoam balls to model classes. Is that a commonly used technique? And, how does that sort of modeling help us to design better the system? If you have any photos to share which can show us how the classes are represented it'd be great!

    Read the article

  • capturing CMD batch file parameter list; write to file for later processing

    - by BobB
    I have written a batch file that is launched as a post processing utility by a program. The batch file reads ~24 parameters supplied by the calling program, stores them into variables, and then writes them to various text files. Since the max input variable in CMD is %9, it's necessary to use the 'shift' command to repeatedly read and store these individually to named variables. Because the program outputs several similar batch files, the result is opening several CMD windows sequentially, assigning variables and writing data files. This ties up the calling program for too long. It occurs to me that I could free up the calling program much faster if maybe there's a way to write a very simple batch file that can write all the command parameters to a text file, where I can process them later. Basically, just grab the parameter list, write it and done. Q: Is there some way to treat an entire series of parameter data as one big text string and write it to one big variable... and then echo the whole big thing to one text file? Then later read the string into %n variables when there's no program waiting to resume? Parameter list is something like 25 - 30 words, less than 200 characters. Sample parameter list: "First Name" "Lastname" "123 Steet Name Way" "Cityname" ST 12345 1004968 06/01/2010 "Firstname+Lastname" 101738 "On Account" 20.67 xy-1z 1 8.95 3.00 1.39 0 0 239 8.95 Items in quotes are processed as string variables. List is space delimited. Any suggestions?

    Read the article

  • Native Endians and Auto Conversion

    - by KnickerKicker
    so the following converts big endians to little ones uint32_t ntoh32(uint32_t v) { return (v << 24) | ((v & 0x0000ff00) << 8) | ((v & 0x00ff0000) >> 8) | (v >> 24); } works. like a charm. I read 4 bytes from a big endian file into char v[4] and pass it into the above function as ntoh32 (* reinterpret_cast<uint32_t *> (v)) that doesn't work - because my compiler (VS 2005) automatically converts the big endian char[4] into a little endian uint32_t when I do the cast. AFAIK, this automatic conversion will not be portable, so I use uint32_t ntoh_4b(char v[]) { uint32_t a = 0; a |= (unsigned char)v[0]; a <<= 8; a |= (unsigned char)v[1]; a <<= 8; a |= (unsigned char)v[2]; a <<= 8; a |= (unsigned char)v[3]; return a; } yes the (unsigned char) is necessary. yes it is dog slow. there must be a better way. anyone ?

    Read the article

  • How to map a test onto a list of numbers

    - by Arthur Ulfeldt
    I have a function with a bug: user> (-> 42 int-to-bytes bytes-to-int) 42 user> (-> 128 int-to-bytes bytes-to-int) -128 user> looks like I need to handle overflow when converting back... Better write a test to make sure this never happens again. This project is using clojure.contrib.test-is so i write: (deftest int-to-bytes-to-int (let [lots-of-big-numbers (big-test-numbers)] (map #(is (= (-> % int-to-bytes bytes-to-int) %)) lots-of-big-numbers))) This should be testing converting to a seq of bytes and back again produces the origional result on a list of 10000 random numbers. Looks OK in theory? except none of the tests ever run. Testing com.cryptovide.miscTest Ran 23 tests containing 34 assertions. 0 failures, 0 errors. why don't the tests run? what can I do to make them run?

    Read the article

  • Java: fastest way to do random reads on huge disk file(s)

    - by cocotwo
    I've got a moderately big set of data, about 800 MB or so, that is basically some big precomputed table that I need to speed some computation by several orders of magnitude (creating that file took several mutlicores computers days to produce using an optimized and multi-threaded algo... I do really need that file). Now that it has been computed once, that 800MB of data is read only. I cannot hold it in memory. As of now it is one big huge 800MB file but splitting in into smaller files ain't a problem if it can help. I need to read about 32 bits of data here and there in that file a lot of time. I don't know before hand where I'll need to read these data: the reads are uniformly distributed. What would be the fastest way in Java to do my random reads in such a file or files? Ideally I should be doing these reads from several unrelated threads (but I could queue the reads in a single thread if needed). Is Java NIO the way to go? I'm not familiar with 'memory mapped file': I think I don't want to map the 800 MB in memory. All I want is the fastest random reads I can get to access these 800MB of disk-based data. btw in case people wonder this is not at all the same as the question I asked not long ago: http://stackoverflow.com/questions/2346722/java-fast-disk-based-hash-set

    Read the article

< Previous Page | 41 42 43 44 45 46 47 48 49 50 51 52  | Next Page >