Oninit cdi / Archive
Talk to an expert →

Does "onmode -F" always work?

Posted in 2006

Summary

A user on IDS 9.40.FC5X3 / HP-UX 11.11 asked why "onmode -F" wouldn't release an apparently near-empty virtual shared memory segment (onstat -g seg showed 7 blocks used). Replies explained the behaviour rather than a bug: onmode -F first frees unused blocks back within each pool/segment, but it can only return a whole segment to the OS if exactly 1 block is in use (the segment header). Since some pool almost always holds blocks, segments usually stay. It was also noted that 9.x drains pools automatically (disable via DONTDRAINPOOLS=1 if latching costs CPU), and that defect 173929 can make onmode -F raise an assert failure, fixed in 9.40.UC8.

Auto-generated by DrWatson from the posts below — may be imperfect; read the full thread.

Topics: Server Administration, Platform-Specific Issues, Versions, Editions & End-of-Life

KERN DO — — source: IIUG Forums & Mailing Lists

Hi, Has anyone experienced this similar problem? I'm running IDS 9.40.FC5X3 on the HP-UX 11.11 -- I always have a hard time freeing unused virtual shared memory segments -- As you can see, the 2nd virtual segment doesn't seem to contain anything. Thanks for any feedback. Kern -- onstat -g seg IBM Informix Dynamic Server Version 9.40.FC5X3 -- On-Line -- Up 18 days 00:29:36 -- 1672960 Kbytes Segment Summary: id key addr size ovhd class blkused blkfree 1378386 1382041601 c000000260000000 974913536 453600 R* 237990 26 171091 1382041602 c0000002a0000000 603979776 19104 V 67951 79505 679386217 1382041603 c000000344000000 134217728 4768 V 7 32761 Total: - - 1713111040 - - 305948 112292 (* segment locked in memory)

↪ replying to KERN DO
Robert Roussey\\\\(MIS\\\\) — — source: IIUG Forums & Mailing Lists

Might be an HP-UX SHM thing. I remember having the same problem with IDS 7.3X on HP-UX 10. Bob Roussey Unix / Informix Administration Spirit Airlines Robert.Roussey@SpiritAir.com 586.741.8991 -----Original Message----- From: ids-bounces@iiug.org [mailto:ids-bounces@iiug.org] On Behalf Of KERN DO Sent: Tuesday, January 03, 2006 8:25 PM To: ids@iiug.org Subject: Does "onmode -F" always work? [6161] Hi, Has anyone experienced this similar problem? I'm running IDS 9.40.FC5X3 on the HP-UX 11.11 -- I always have a hard time freeing unused virtual shared memory segments -- As you can see, the 2nd virtual segment doesn't seem to contain anything. Thanks for any feedback. Kern -- onstat -g seg IBM Informix Dynamic Server Version 9.40.FC5X3 -- On-Line -- Up 18 days 00:29:36 -- 1672960 Kbytes Segment Summary: id key addr size ovhd class blkused blkfree 1378386 1382041601 c000000260000000 974913536 453600 R* 237990 26 171091 1382041602 c0000002a0000000 603979776 19104 V 67951 79505 679386217 1382041603 c000000344000000 134217728 4768 V 7 32761 Total: - - 1713111040 - - 305948 112292 (* segment locked in memory) ************************************************************************ ******* Forum Note: Use "Reply" to post a response in the discussion forum.

↪ replying to Robert Roussey\\\\(MIS\\\\)
John_Caron@mab.uscourts.gov — — source: IIUG Forums & Mailing Lists

We have IDS 7.3.x and I have an entry in cron that runs at 6:00 each AM and it works for me. Since that entry has been made I haven't experienced any issues with memory. We are still on Sol 7 unfortunately. Hope this helps. John Caron USBC-MA Worcester 508-770-8957 Boston 617-565-6175 "Robert Rous...." <Robert.Roussey@S piritAir.com> To Sent by: ids@iiug.org ids-bounces@iiug. cc org Subject RE: Does "onmode -F" always work? 01/04/2006 08:51 [6163] AM Please respond to ids@iiug.org Might be an HP-UX SHM thing. I remember having the same problem with IDS 7.3X on HP-UX 10. Bob Roussey Unix / Informix Administration Spirit Airlines Robert.Roussey@SpiritAir.com 586.741.8991 -----Original Message----- From: ids-bounces@iiug.org [mailto:ids-bounces@iiug.org] On Behalf Of KERN DO Sent: Tuesday, January 03, 2006 8:25 PM To: ids@iiug.org Subject: Does "onmode -F" always work? [6161] Hi, Has anyone experienced this similar problem? I'm running IDS 9.40.FC5X3 on the HP-UX 11.11 -- I always have a hard time freeing unused virtual shared memory segments -- As you can see, the 2nd virtual segment doesn't seem to contain anything. Thanks for any feedback. Kern -- onstat -g seg IBM Informix Dynamic Server Version 9.40.FC5X3 -- On-Line -- Up 18 days 00:29:36 -- 1672960 Kbytes Segment Summary: id key addr size ovhd class blkused blkfree 1378386 1382041601 c000000260000000 974913536 453600 R* 237990 26 171091 1382041602 c0000002a0000000 603979776 19104 V 67951 79505 679386217 1382041603 c000000344000000 134217728 4768 V 7 32761 Total: - - 1713111040 - - 305948 112292 (* segment locked in memory) ************************************************************************ ******* Forum Note: Use "Reply" to post a response in the discussion forum. ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum.

↪ replying to KERN DO
Schmitz, Rob B [IBM — — source: IIUG Forums & Mailing Lists

My experience has shown that the number of used blocks needs to be 1 before onmode -F will free it. Anything above 1 and it will not free it. Rob Schmitz Informix DBA 913-315-4183 rob.b.schmitz@sprint.com schmitzr@us.ibm.com -----Original Message----- From: ids-bounces@iiug.org [mailto:ids-bounces@iiug.org] On Behalf Of KERN DO Sent: Tuesday, January 03, 2006 7:25 PM To: ids@iiug.org Subject: Does "onmode -F" always work? [6161] Hi, Has anyone experienced this similar problem? I'm running IDS 9.40.FC5X3 on the HP-UX 11.11 -- I always have a hard time freeing unused virtual shared memory segments -- As you can see, the 2nd virtual segment doesn't seem to contain anything. Thanks for any feedback. Kern -- onstat -g seg IBM Informix Dynamic Server Version 9.40.FC5X3 -- On-Line -- Up 18 days 00:29:36 -- 1672960 Kbytes Segment Summary: id key addr size ovhd class blkused blkfree 1378386 1382041601 c000000260000000 974913536 453600 R* 237990 26 171091 1382041602 c0000002a0000000 603979776 19104 V 67951 79505 679386217 1382041603 c000000344000000 134217728 4768 V 7 32761 Total: - - 1713111040 - - 305948 112292 (* segment locked in memory) ************************************************************************ ******* Forum Note: Use "Reply" to post a response in the discussion forum.

↪ replying to John_Caron@mab.uscourts.gov
Schmitz, Rob B [IBM — — source: IIUG Forums & Mailing Lists

My experience has shown that the number of unused blocks needs to be down to 1. Anything above 1 and onmode -F won't free it. I've never really paid any attention to this behavior from an OS perspective, but I'm pretty sure it is like that on HP-UX, AIX, and Sun. Rob Schmitz Informix DBA 913-315-4183 rob.b.schmitz@sprint.com schmitzr@us.ibm.com -----Original Message----- From: ids-bounces@iiug.org [mailto:ids-bounces@iiug.org] On Behalf Of John_Caron@.... Sent: Wednesday, January 04, 2006 8:08 AM To: ids@iiug.org Subject: RE: Does "onmode -F" always work? [6164] We have IDS 7.3.x and I have an entry in cron that runs at 6:00 each AM and it works for me. Since that entry has been made I haven't experienced any issues with memory. We are still on Sol 7 unfortunately. Hope this helps. John Caron USBC-MA Worcester 508-770-8957 Boston 617-565-6175 "Robert Rous...." <Robert.Roussey@S piritAir.com> To Sent by: ids@iiug.org ids-bounces@iiug. cc org Subject RE: Does "onmode -F" always work? 01/04/2006 08:51 [6163] AM Please respond to ids@iiug.org Might be an HP-UX SHM thing. I remember having the same problem with IDS 7.3X on HP-UX 10. Bob Roussey Unix / Informix Administration Spirit Airlines Robert.Roussey@SpiritAir.com 586.741.8991 -----Original Message----- From: ids-bounces@iiug.org [mailto:ids-bounces@iiug.org] On Behalf Of KERN DO Sent: Tuesday, January 03, 2006 8:25 PM To: ids@iiug.org Subject: Does "onmode -F" always work? [6161] Hi, Has anyone experienced this similar problem? I'm running IDS 9.40.FC5X3 on the HP-UX 11.11 -- I always have a hard time freeing unused virtual shared memory segments -- As you can see, the 2nd virtual segment doesn't seem to contain anything. Thanks for any feedback. Kern -- onstat -g seg IBM Informix Dynamic Server Version 9.40.FC5X3 -- On-Line -- Up 18 days 00:29:36 -- 1672960 Kbytes Segment Summary: id key addr size ovhd class blkused blkfree 1378386 1382041601 c000000260000000 974913536 453600 R* 237990 26 171091 1382041602 c0000002a0000000 603979776 19104 V 67951 79505 679386217 1382041603 c000000344000000 134217728 4768 V 7 32761 Total: - - 1713111040 - - 305948 112292 (* segment locked in memory) ************************************************************************ ******* Forum Note: Use "Reply" to post a response in the discussion forum. ************************************************************************ ******* Forum Note: Use "Reply" to post a response in the discussion forum. ************************************************************************ ******* Forum Note: Use "Reply" to post a response in the discussion forum.

↪ replying to KERN DO
Model-Bosch, Tilman — — source: IIUG Forums & Mailing Lists

Hi, onmode -F does two things: 1) It searches all memory pools and tries to free any freeblks in each memory pool. Those freeblks belong to the memory pool, but are unused. When 'onmode -F succeeds, it returns the memory to the the the free memory in the segment. You should see that the number of blkferee increases and the number of blkused decreases after running 'onmode -F' . 2) It attempts to return any segment that is completley free to the operating system. This is possible only, if the segment has only 1 blkused , as this first block of memory contains the segment header structure. If any other blk is in use, the segment can't be returned to the OS. Unfortunatley, any SHM segment will almost everytime have some blocks allocated by some user or internal pools. So you will often see that onmode -F frees memory inside a segment, but is not able to return a segment memory to the OS. Apart from that, in IDS 9.x pools are usually drained by default, i.e. the engine attempts constantly to return freeblks in pools to the the global free memory inside a SHM segment. So 'onmode -F' will not be that usefull , as memory managment of the IDS engine is more efficient anyway. The downturn of this is, that the constant attempt to free memory from pools requires latching, which in turn may cause busy waits by other threads. This can lead to waste of CPU cycles. If you are experiencing CPU resource bottlenecks (long ready queues) you may! be hitting this issue. (may have other reasons). To turn off the new memory managment in IDS , you may set the environment variable DONTDRAINPOOLS to 1 , before you start IDS (in the environment of the user who starts IDS). To avoid the increased usage of memory after setting DONTDRAINPOOLS you may choose to run 'onmdoe -F' again at intervals of your choice. One last issue to consider: in IDS 9.40 there is a defect 173929 which may cause an 'ASSERT FAILURE' when onnmode -F is run. 173929 ONMODE -F MAY CAUSE ASSERT FAIL IN MT_SET_POOL The defect will cause an assertion failure , but the engine will stay online. This is fixed in 9.40.UC8 . Best regards Tilman -----Ursprüngliche Nachricht----- Von: ids-bounces@iiug.org [mailto:ids-bounces@iiug.org] Im Auftrag von KERN DO Gesendet: 04 January 2006 02:25 An: ids@iiug.org Betreff: Does "onmode -F" always work? [6161] Hi, Has anyone experienced this similar problem? I'm running IDS 9.40.FC5X3 on the HP-UX 11.11 -- I always have a hard time freeing unused virtual shared memory segments -- As you can see, the 2nd virtual segment doesn't seem to contain anything. Thanks for any feedback. Kern -- onstat -g seg IBM Informix Dynamic Server Version 9.40.FC5X3 -- On-Line -- Up 18 days 00:29:36 -- 1672960 Kbytes Segment Summary: id key addr size ovhd class blkused blkfree 1378386 1382041601 c000000260000000 974913536 453600 R* 237990 26 171091 1382041602 c0000002a0000000 603979776 19104 V 67951 79505 679386217 1382041603 c000000344000000 134217728 4768 V 7 32761 Total: - - 1713111040 - - 305948 112292 (* segment locked in memory) ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum.

Related threads

Glossary