|
Problem IC66162
|
Status: Closed |
Control dynamic downward adjustment of FCM resources |
| product: |
DB2 FOR LUW / DB2FORLUW / 950 - DB2 |
| Problem description: |
Customers with large DPF environments like are experiencing
significant performance issues and
apparent hangs due to contention when FCM resources are tuned
automatically. The system
controller thread (db2sysc) will adjust FCM resources downward
after 30 minutes or more of low
workload. During relatively quiet times, the resources can get
adjusted down to very low values.
This causes problems on systems with large workload bursts. When
the workload increases dramatically,
the FCM resources need to be adjusted upwards quickly and this
is when performance issues due to latch
contention in the dynamic adjustment code occurs. This problem
gets worse on large DPF
instances with many logical partitions configured. |
| Problem Summary: |
****************************************************************
* USERS AFFECTED: *
* All DPF environments *
****************************************************************
* PROBLEM DESCRIPTION: *
* Customers with large DPF environments like are experiencing *
* *
* significant performance issues and apparent hangs due to *
* contention when FCM resources are tuned automatically. The *
* system *
* controller thread (db2sysc) will adjust FCM resources *
* downward *
* after 30 minutes or more of low workload. During relatively *
* quiet *
* times, the resources can get adjusted down to very low *
* values. *
* This causes problems on systems with large workload bursts. *
* When *
* the workload increases dramatically, the FCM resources need *
* to be *
* adjusted upwards quickly and this is when performance issues *
* due to latch *
* contention in the dynamic adjustment code occurs. This *
* problem gets *
* worse on large DPF instances with many logical partitions *
* configured. *
****************************************************************
* RECOMMENDATION: *
* Customer should upgrade to v95 fixpack 6. *
**************************************************************** |
| Local Fix: |
Large DPF customers can set FCM resources to large fixed values.
However, the downside is that this may cause node failures in
cases when db2fcmr/db2fcms EDUs run out of FCM buffers/channels. |
| available fix packs: |
DB2 Version 9.5 Fix Pack 6a for Linux, UNIX, and Windows
DB2 Version 9.5 Fix Pack 7 for Linux, UNIX, and Windows
DB2 Version 9.5 Fix Pack 8 for Linux, UNIX, and Windows
DB2 Version 9.5 Fix Pack 9 for Linux, UNIX, and Windows
DB2 Version 9.5 Fix Pack 10 for Linux, UNIX, and Windows
|
| Solution |
Document available in v95 fixpack 6 version.
With this fix setting FCM_NUM_BUFFERS or FCM_NUM_CHANNELS to
"<large number> automatic" will make sure that
the system controller thread doesn't adjust resources down below
the specified <large number> value. |
| Workaround |
not known / see Local fix |
| BUG-Tracking |
forerunner : APAR is sysrouted TO one or more of the following: IC68267 IC68270
follow-up : |
| Timestamps |
Date - problem reported :
Date - problem closed :
Date - last modified :
| 10.02.2010
01.11.2011
01.11.2011 |
| Problem solved at the following versions (IBM BugInfos) |
9.5.FP6
|
| Problem solved according to the fixlist(s) of the following version(s) |
