On 6/13/07, Eric and Barbara Sammons <[EMAIL PROTECTED]> wrote:
Looking into performance problems, in search of some ideas of what to look for. I believe the issue may be related to the type of DASD being used.
Fortunately there's instrumentation on the platform, and you need a performance monitor to analyze your configuration and workload to understand what is holding you back.
Scenario, dd command to generate a 1G file if=/dev/zero takes 15s on x86_64 platform. The same command takes 15m on the s390x platform. In my environment the command takes 16s. My environment leverages MOD-3 DASD while the environment in question has MOD-9 and MOD-27 devices. I know historically the MOD-9 is typically a slower device but I don't know much about the MOD-27.
The "slow" was for real 3390's. Modern DASD only emulates the number of cylinders on the logical volumes, so a 3390-27 is not necessarily slower than a 3390-3. However, since there is only one I/O active at the same time to the logical volume, the subchannel might become a bottleneck. Or phrased differently, when your workload is such that it could run multiple I/O's in parallel, you might be able to take advantage of providing the disk space in multiple logical volumes (and thus multiple subchannels). Second best could be PAV in some situations. Your test is in general not enough to understand the capabilities. After all, the mainframe was meant to run a lot in parallel and it would be better to measure how many servers could be doing this test at the same time without impacting each other heavily.
Is it truly possible that the performance of these devices is bad enough to see such a difference between MOD-3 and x86_64 v. MOD-9 and MOD-27? What else should / can I look at to determine the root cause of the performance issues?
Your numbers look weird enough to be suspicious. Most likely your x86 platform has a huge amount of memory so it never even did a single I/O to disk for your benchmark. If you were given a fairly small Linux server on the mainframe, it suddenly had to do I/O and wait for that to complete. Very different cases. If you do 1 GB in 15 min, that's 1 MB/s. Still rather slow. First you'd need to see whether this is ESCON or FICON. There may also be other things holding back the virtual machine (like paging, CPU contention, failure to drop from queue, etc). Collect performance data and see what's going on. Rob -- Rob van der Heij Velocity Software, Inc http://velocitysoftware.com/ ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
