[serm]

Hi Guys

We have a system running 5.3 TL4, with 15GB Memory and 4GB of paging. We are getting Fork errors and other DB2 errors - can someone tell me the IBM recommended sizes for paging under 5.3? Previously I would set it to the same as Memory - up to a maximum of 4GB.

I have been told one formula for working it out as:

512MB + ((actual RAM - 256) X1.25)

Is there no maximum recomendation?

Or if you could point me at a decent Whitepaper..?
Many thanks

[serm]

Sorry - as a follow on question - if paging should be set (in a virtual environment) to the same as Computational Memory , should this value be the maximum value the PHYP can allocate - ie. not the desired level of memory?

Thanks again!!

[alexisl]

I don't know the magic formula but I can tell you that this is what we have:
Total(MB) 19456.0MB physical 28544.0MB paging which is about 1.5 times the physical memory and it is working good right now

[cdelgadop]

Hi

Could you please post the output of lsps -a command ? I think you have the default settings for the VMM (Virtual Memory Manager) of AIX 5L and those values are very bad for DB I/O patterns (either Oracle or DB2).

You can check for this values by issuing:

vmo -a | egrep"maxperm|maxclient|minperm"

by default they're:

maxperm=80
maxclient=80
minclient=20

As per AIX 5.2 i use to config by applying some SAP technotes recommendations for High Used SAP Central Instances DBs Servers:

maxperm=8
maxclient=8
minclient=3

As per AIX 5.3 there was a change and now new values are recommended since there are new variables that can control the lrud (Least Recently Used Daemon):

minperm=3
maxperm=90
maxclient=90
lru_file_repage=0
strict_maxclient=1
strict_maxperm=0

You can change this values in teh current exec of AIX by using the following smitty fastpath:

smitty "ChShCurVM"


Hope this helps

[serm]

Originally Posted by cdelgadop

Hi

Could you please post the output of lsps -a command ? I think you have the default settings for the VMM (Virtual Memory Manager) of AIX 5L and those values are very bad for DB I/O patterns (either Oracle or DB2).

You can check for this values by issuing:

vmo -a | egrep"maxperm|maxclient|minperm"

by default they're:

maxperm=80
maxclient=80
minclient=20

As per AIX 5.2 i use to config by applying some SAP technotes recommendations for High Used SAP Central Instances DBs Servers:

maxperm=8
maxclient=8
minclient=3

As per AIX 5.3 there was a change and now new values are recommended since there are new variables that can control the lrud (Least Recently Used Daemon):

minperm=3
maxperm=90
maxclient=90
lru_file_repage=0
strict_maxclient=1
strict_maxperm=0

You can change this values in teh current exec of AIX by using the following smitty fastpath:

smitty "ChShCurVM"


Hope this helps

Hi
Here is the output of VMO and lsps:

root > lsps -a
Page Space Physical Volume Volume Group Size %Used Active Auto Type
hd6 hdisk0 rootvg 4096MB 9 yes yes lv

Does this figure appear low if the RAM is generally set (by PLM) at 15GB?


root > vmo -a | egrep "maxperm|maxclient|minperm"
maxclient% = 15
maxperm = 568340
maxperm% = 15
minperm = 378891
minperm% = 10
strict_maxclient = 1
strict_maxperm = 1

I do not have the lru_file_repage turned off yet..will do so shortly so that it will page only if file pages are above minperm.

Thanks again.

[alanmoon]

Serm,
I used to use a factor of 1.5 x real memory to get the paging space size. Since memory amounts have ballooned I use a 1.1 x real memory. I was always under the impression that you needed at least a 1 for 1 ratio between real memory and paging.

[b00tsie]

If you like I can mail you an article with the latest performance guidelines.
Hope this will help you out...


These are the settings explained in this article:

Adjust the nextboot in /etc/tunables

vmo:
lru_poll_interval = "10"
lru_file_repage = "0"
maxfree = "1088"
minfree = "960"
minperm% = "5"

no:
udp_recvspace = "655360"
udp_sendspace = "65536"
tcp_recvspace = "262144"
tcp_sendspace = "262144"
sb_max = "1310720"
rfc1323 = "1"

nfso:
nfs_tcp_socketsize = "60000"
nfs_socketsize = "60000"

ioo:
# j2_nPagesPerWriteBehindCluster = "32" #default
# j2_nRandomCluster = "0" #default
# numclust = "1" #default
lvm_bufcnt = "9"
# lvm_bufcnt = "12" # use this with much raw I/O
numfsbufs = "2048"
pv_min_pbuf = "1024"
j2_nBufferPerPagerDevice = "1024"
maxrandwrt = "32"
j2_maxRandomWrite = "32"
maxpgahead = "16"
j2_maxPageReadAhead = "128"

[csubudhi]

Hi Guys,

I have Oracle server running in cluster . AIX 5.3 ML4 and Oracle 10g.
When ever I run net backup for offline tape backup, servers gets reboot.
Immd after net backup trigger, it reboots. Net backup servr tries to reconnect and takes the backup successful. Why first time it reboots ?

we have 11 GB RAM and in each LPAR.

Can you help how to over come ? May be I have to tune the AIX parameters or 10g Parameters ?

[Mark Taylor]

lru_file_repage functionality is also introduced in 5.2 FYI ..

Paging space and how much do you need .. there is no formula because each server is different .. it used to be that you had at least as much as realmem and a bit more, but with the introduction of deffered paging space allocation then you do not really need that i.e. a server running with 64GB realmem which is only under full load using 48GB of realmem is not going to need 96GB of paging space .. correctly tuned VMM will mean it requires very little (( i am talking about paging and not dump devices ))

lru_file_repage is the big hitter for this, we have been waiting for this flag for years .. gone are the days of lowering maxperm to below numperm .. that was a pain ..

Rgds
Mark Taylor