SOLVED: RDNG performance optimisation on a Windows VMware deployment
For at least 4 years we have been suffering slow performance for RM on our RDNG platform. It has undergone a couple of reoganisations and upgrades; to some extent these helped but the user experience still was not great.
We have a small user base, but hundreds of thousands of requirements - so we don't match the typical IBM tuning examples; we use Windows Servers, which again does not match the IBM sample tuning guides.
WHat follows is our experience, make any changes to your system at your own risk.
Here is what fixed our system:
- Determine how many real phyiscal cores the server has; in our case 2x6
- Allocate no more than the real physical cores your server has (DO NOT hyperthread / count 1 core as 2)
- Allocate the cores to each server as appropriate, for us it was 6 for RM, 2 DCC, 2, DB2, 2 Proxy
- ENSURE they are dedicated/reserved
-
***I'll say this again, check, with your own eyes, they are allocated as Reserved/Dedicated becuase your IT admin may not understand what this actually means***
-
Determine how much real RAM the server has, in our case 124G available
- Allocate RESERVED RAM (see 3.2 above); in our case, RM 90G, DCC 12G, DB2 16G, Proxy 4G
-
Lots of RAM is required for RM to Cache the Indices; having said that, caching does not solve the RM performance problem, so see 8
-
Use SSDs,
- if, like us you don't have SSDs available (or maybe even if you do, becuase RAM is still a lot faster than an SSD), set up a RAM DISK for the INDICES on your RM server.
- NOTE: this is NOT explicitly supported by IBM, however, it does not negate support.
-
Set up a RAM DISK large enough to store your indices (we used imdisk)
-
Copy the indices (the jts/rm/ccm etc folders) to the ram disk on system start from a persistent store
- ensure you backup your RAMDISK indices regularly to persistent store (so if you need to restart your system, the reindex-update is quick)
-
Set your system %TEMP% variables to your RAMDISK/temp
- This process alone improved our RM performance 3-10x - depending on operation.
- We can now, finally, run reports on large modules without them timing out, we can create JRS reports with 100000+ items without failing.
-
Use FireFox, Chrome's caching seems to cause us issues.
- Awesome!
Good luck!
Comments
Davyd Norris
Oct 25 '18, 3:39 a.m.Great news! Although I do remember asking 3.2 about a dozen times myself :-)