Summary: Memory management question.

From: Jim Kurtenbach <KurtenbachJ_at_crlcorp.com>
Date: Tue, 05 Feb 2002 09:43:23 -0600

Thanks to Andy Cohen, Alan (alan_at_nabeth.cxo.cpqcorp.net) and others.

The command sys_check -perf >perfmon.html creates a spiffy html document
that makes recommendations for your system tuning. It also gives you
hyperlinks to more detailed information on the web. What a discovery!!

Additionally, there is a white paper for tuning tru64 with a large memory
configuration (more than 2gig physical memory) which had a recommended
sysconfigtab for a large database.
http://www.tru64unix.compaq.com/docs/best_practices/sys_bps.html

It turns out that it is at least a two step process.

        Step 1: Tune the kernel to:
                Limit the amount of ubc-buffer in /etc/sysconfigtab
                        vm:
                                        ubc-maxpercent=50

                Increase the amount of shared memory. (Oracle uses this
for the SGA)
                A quick glance at the Oracle installation manual will guide
you on some parameters.
                        ipc:
                                        shm-max =2139095040
                                        shm-mni =2048
                                        shm-seg =512
                                        sem-mni =2048
                                        sem-msl =2048
                                        sem-opm =400
                                        sem-ume =400
                                        ssm-theshold =0

        Step 2: Check the init<dbase>.ora
                Since you have more memory, you can increase the size of the
SGA.

Note: Our Oracle DBA's had been increasing the size of the SGA in the init
files for years, the TRU64 shared memory had never been increased from the
default. As a result, paging activity had gone through the roof.

Thanks to all!

Jim

> -----Original Message-----
> From: Jim Kurtenbach
> Sent: Monday, January 28, 2002 2:44 PM
> To: tru64-unix-managers_at_ornl.gov
> Cc: Stephen Ripley; Donald P. Theune; Ray A. Smith
> Subject: Memory management question.
>
> Dear managers,
>
> I have a AlphaServer 8200 running 4.0G with 6gig of ram with 4- 466mhz
> cpu's. This server runs several instances of Oracle.
>
> I am experience some performance slowness. Recently, I swapped a 1 gig
> memory board for a 4gig (went from 3gig to 6gig) and didn't see any
> appreciable performance improvement.I've experimented changing the vm
> parameters ubc-maxpercent from 100 to 75 to 50 (at the next boot). Also
> I've changed the ubc-borrowpercent to 10. My thoughts are that too much
> memory is being used as disk cache and not enough for processes.
>
> Not being anywhere near an expert, I was wondering if a couple of you
> experts could point me in the right direction for diagnosing this.
>
> Also, I read some references in the archives regarding a patch for the vm
> subsystem in version 4.x. Could someone point me in the direction of
> where to find patches for tru64?
>
>
> Thanks in advance!
>
> Jim
>
>
> Here's a couple of the things I'm looking at.
>
>
>
> # vmstat -P
>
> Total Physical Memory = 6144.00 M
> = 786432 pages
>
>
> Physical Memory Clusters:
>
> start_pfn end_pfn type size_pages / size_bytes
> 0 256 pal 256 / 2.00M
> 256 786348 os 786092 / 6141.34M
> 786348 786432 pal 84 / 672.00k
>
> Physical Memory Use:
>
> start_pfn end_pfn type size_pages / size_bytes
> 256 273 unixtable 17 / 136.00k
> 273 284 scavenge 11 / 88.00k
> 284 768 text 484 / 3.78M
> 768 873 data 105 / 840.00k
> 873 1088 bss 215 / 1.68M
> 1088 1229 kdebug 141 / 1.10M
> 1229 1234 cfgmgmt 5 / 40.00k
> 1234 1235 locks 1 / 8.00k
> 1235 2474 unixtable 1239 / 9.68M
> 2474 2487 pmap 13 / 104.00k
> 2487 15052 vmtables 12565 / 98.16M
> 15052 786348 managed 771296 / 6025.75M
> ============================
> Total Physical Memory Use: 786092 / 6141.34M
>
> Managed Pages Break Down:
>
> free pages = 98
> active pages = 52134
> inactive pages = 104878
> wired pages = 93592
> ubc pages = 520996
> ==================
> Total = 771698
>
> WIRED Pages Break Down:
>
> vm wired pages = 5130
> ubc wired pages = 56121
> meta data pages = 23582
> malloc pages = 4892
> contig pages = 134
> user ptepages = 3218
> kernel ptepages = 235
> free ptepages = 7
> ==================
> Total = 93319
> #
> #
> # swapon -sv
> Swap partition /dev/rz9b (default swap):
> Allocated space: 131072 pages (1024MB)
> In-use space: 50175 pages ( 38%)
> Free space: 80897 pages ( 61%)
>
> Swap partition /dev/rz21c:
> Allocated space: 523755 pages (4091MB)
> In-use space: 50416 pages ( 9%)
> Free space: 473339 pages ( 90%)
>
>
> Total swap allocation:
> Allocated space: 654827 pages (5115MB)
> Reserved space: 291198 pages ( 44%)
> In-use space: 100591 pages ( 15%)
> Available space: 363629 pages ( 55%)
> #
>
> James Kurtenbach
> Clinical Reference Laboratory
> 8433 Quivira
> Lenexa, KS 66215
> (913) 492 3652 x 1651
> kurtenbachj_at_crlcorp.com
>
Received on Tue Feb 05 2002 - 15:43:31 NZDT

This archive was generated by hypermail 2.4.0 : Wed Nov 08 2023 - 11:53:43 NZDT