I came across this blog entry and the accompanying presentation by Robert Milkoski about his experience switching from Linux to Oracle Solaris 11 for a distributed OpenAFS file serving environment at Morgan Stanley.
If you are an IT manager, the presentation will show you:
Running Solaris with a support contract can cost less than running Linux (even without a support contract) because of technical advantages of Solaris.
IT departments can benefit from hiring computer scientists into Systems Programmer or similar roles. Their computer science background should be nurtured so that they can continue to deliver value (savings and opportunity) to the business as technology advances.
If you are a sysadmin, developer, or somewhere in between, the presentation will show you:
A presentation that explains your technical analysis can be very influential.
Learning and using the non-default options of an OS can make all the difference as to whether one OS is better suited than another. For example, see the graphs on slides 3 - 5. The ZFS default is to not use compression.
When trying to convince those that hold the purse strings that your technical direction should be taken, the financial impact can be the part that closes the deal. See slides 6, 9, and 10. Sometimes reducing rack space requirements can be the biggest impact because it may stave off or completely eliminate the need for facilities growth.
DTrace can be used to shine light on performance problems that may be suspected but not diagnosed. It is quite likely that these problems have existed in OpenAFS for a decade or more. DTrace made diagnosis possible.
DTrace can be used to create performance analysis tools without modifying the source of software that is under analysis. See slides 29 - 32.
Microstate accounting, visible in the prstat output on slide 37 can be used to quickly draw focus to problem areas that affect CPU saturation. Note that prstat without -m gives a time-decayed moving average that is not nearly as useful.
Instruction level probes (slides 33 - 34) are a super-easy way to identify which part of a function is hot.
