01.01.2012 в 11:23 #14323
After several decades of Journal (JO) attempting to pick the right number of disks, the fastest disks and those connected to the right IOAs, JO is getting out of the disk selection business. In 7.1, it will no longer pick specific disks for journal receivers. It will ask for storage and let storage management (SM) give it to JO wherever SM feels is best.
The disk selection process was put in place years ago when there was a greater disparity between disk speeds and there was no IOA write cache. Selecting specific disks allowed JO to queue up disk operations in parallel which allowed greater journal throughput. The down side of this has always been that it could cause disk usage to be skewed, meaning disks filled at different rates. This was a perceived, and possibly real, performance problem and generated calls to Support that we want to eliminate.
After testing this change in multiple hardware environments with different workloads, no difference in performance was observed. MF51614 is the 7.1 PTF to change this behavior, it will propagate to the field over the next few months.
Here’s what you need to know and share:
- 10 is no longer a magic number of disks for journal receivers, data goes wherever SM puts it
- There is no minimum nor maximum number of disk units for receivers
- The journal receiver size threshold no longer affects the number of disks across which receiver are spread
- This is 7.1 only and requires a PTF, it will take a while before the PTF is propagated but the intention is to have the PTF in the next 7.1 cumulative PTF package and DB group
- If you have skewed disk usage, don’t call JO, call SM
Other things to know:
- If there is a need to revert to the old behavior, we have a patch to do so
- No PTF is planned for prior releases at this time
- There is still a potential benefit in specifying a high receiver size threshold as it influences the number of I/O requests we will request in parallel
The change below is being made in InfoCenter, it replaces the reference to the journal disk selection algorithm where an additional disk arm was used for each 32MB/64MB of rcvr size threshold.
How many arms in my disk pool will journaling use? Starting with IBM i 7.1 journal receivers may be spread across all disk arms in the disk pool. Journaling no longer directs writes to specific disk arms.
The journal receiver threshold value will influence the number of parallel writes that journal will allow. The higher the journal receiver threshold value, the more parallel I/0 requests will be allowed. Allowing more parallel I/O requests may improve performance.
Для ответа в этой теме необходимо авторизоваться.