On 5 October 2012 10:08, Christophe Geylens <[email protected]> wrote: > Hello, > > still very green behind the ears but i'll do my best to explain it as best > as i can. > > We are currently running sles10 SP3 under z/VM 5.4. The linux has an lvm > consisting of about 140 MOD9 disks for our db2 database. > We are in progress of migrating al our linux to sles11 so i was looking > into a way to use bigger disks as to be able to manage them easier. > > I thought about using hyperpav so i set up 2 completly similar test > systems except for the lvm : > PAV0 : 36 mod 9 disk, 12 stripes > PAV1 : 12 mod 27 disk each one in a different CU, 24 aliasses (2 for each > cu), 12 stripes > > Now when i run multiple db2 scripts at the same tiem on them, the progress > time and memory usage is about the same. > However the cpu usage of the PAV1 is 3 times as much as the PAV0 when i > check in the "ibm performance toolkit for VM" > In linux itself PAV0 stays around 30-40% while PAV1 sometimes goes as high > as 180%. > > Is this normal? If not what can i do to fix it. >
Christophe, Most important is the total CPU cost, not just the percentage... My initial reaction would be that if your I/O is not the bottleneck anymore you get the work done 5 times faster and thus consume your CPU quicker... ;-) The smaller the stripes get, the more CPU it takes to break up the I/O and merge them later. Depending on the definition of the tablespace, DB2 itself will already group the devices and stripe the tables. Doing it twice is wasting resources. In general for maximum throughput, at a given number of subchannels you're better off with large disks plus PAV than many small disks with no alias. However, most database issues I worked on were CPU constrained rather than limited by disk I/O throughput (you can't even throw data away as fast as you can read it from disk). You're probably aware that DB2 LUW does not have a friendly relation with tablespaces on ECKD devices... It will cost you at least 2 CPU seconds extra per GB of data you read. How sad it may be, if you're in a position to do things over it may make sense to consider FCP devices and do away with much the LVM stuff. Rob -- Rob van der Heij Velocity Software http://www.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 ---------------------------------------------------------------------- For more information on Linux on System z, visit http://wiki.linuxvm.org/
