Real life "large" server

[Rigido]

Hi Protectors
few weeks ago I opened a pmr, then a second and, finally, a third one for performance problems on two servers.
So, why am I writing on the "Capacity Planning" section?
Both server were updated to 8.1.11 and DB & LOG (256GB) increased as we started to use DCs and abandoned VTL, storage is HDS enterprise class all flash.
First server started to have issues with expire inventory and blocked sessions, second server started to have issues with "db backup", long lasting sessions and both servers had "ACTIVE LOG FULL" condition/crash (especially one).
After days of analysis, data collection, servermon files, crash and "crash recovery", support made us set DB2_USE_ALTERNATE_PAGE_CLEANING to ON, then increase ACTIVE LOG size to 512GB (on a 600GB filesystem) and, now, to increase active log filesystem size to 2TB!
We think problems begun as servers started to delete extents and, as we are talking of one hundred SAP HANA TDP daily full/differential backups (db sizes vary from few GBs to 9TB) I think there will be millions of extents to delete every day.

So, questions are:

1. Will DB2 really use free 1,5TB on the active log filesystem?
2. Is it possible we reached a "design" limit and we should starting to think to split workload?
3. What is your biggest installation?

Thanks.

 

[dietmar]

Hi,

I could only tell that i help manage a Environment where the Database is 4,5+TB, and the Active Log have to be in an dedicated Filesystem . The Daily Rate change is between 30-40 TB . So i don't know how much u do on your environment, but this is a Windows Server. I guess with a big AIX it should be even faster .

I have been working on a customer which had a HDS 990 full Flash SAN Environment over 2 DC for their VMWARE Infrastructure. During restore i have seen bad Performance (70-80MB/s for each vmdk , Backup (read) was 300-400MB/s ) . Currently we found out that we have 3ms Write delay during hotadd if the Restore is on a Mirrored Data Store . So we tested a local NOT mirrored Data Store and we had 1ms and 140MB/s for each vmdk restore . Customer also set ALUA now but it did not help for now . I am saying this because i prefere local NVMe or dedicated local small Flash Arrays for the DB2 . Any spinning device for the Database is NO option.

good luck , Dietmar