Hi,
yeah seems pretty similar to what we are always getting from engineering ... so after having tuned a couple of 100 lpars - most of them DB boxes - I would go for following values I think ...
vmo -p -o minperm%=5
vmo -p -o maxperm%=90
vmo -p -o maxclient%=90
vmo -p -o maxperm%=80
vmo -p -o minfree=960
vmo -p -o maxfree=1088
vmo -p -o lru_file_repage=0
vmo -p -o lru_poll_interval=10
ioo -p -o j2_nBufferPerPagerDevice=1024 (if not sufficient you can go upwards 256-wise - but carefully)
ioo -p -o hd_pbuf_cnt=1024 - you can go upwards 256-wise
ioo -p -o numfsbufs=1536 (can be raised up to 2048 if needed)
ioo -p -o pv_min_pbuf=1024 (you can play here up to 1536 but carefully)
ioo -p -o j2_maxRandomWrite=32 - needs monitoring but you can go up to 128 in particular cases
ioo -p -o maxrandwrt=32
ioo -p -o j2_maxPageReadAhead=128
ioo -p -o maxpgahead=16
no -p -o tcp_sendspace=262144
no -p -o tcp_recvspace=262144
no -p -o udp_sendspace=65536
no -p -o udp_recvspace=655360
if you are running cio you can even amend
ioo -p -o lvm_bufcnt=16
But please be aware - all these values are 'best practices' but every system is different, every workload, every query. What can be really great on one system can cause a huge performance impact on another one ... so keep on monitoring.
What you have to avoid are the real problems :
fi fo pi po fr sr in sy cs us sy id wa
41 196 8 0 0 0 1110 29968 10573 17 10 70 3
75 213 11 0 0 0 1043 32556 13921 16 14 62 7
22 74 85 0 0 0 740 57171 6965 21 5 68 6
60 151 99 0 0 0 434 46722 7054 31 11 47 11
96 230 88 0 0 0 1351 32595 11736 19 10 63 8
that means your box is paging out computational pages - so you should fix this at the earliest
vmo -p -o lru_file_repage=0
72.5 numperm percentage
vmo -p -o minperm%=5
vmo -p -o maxperm%=90
vmo -p -o maxclient%=90
vmo -p -o maxperm%=80
4018 pending disk I/Os blocked with no pbuf
48890 paging space I/Os blocked with no psbuf
2740 filesystem I/Os blocked with no fsbuf
110 client filesystem I/Os blocked with no fsbuf
89345 external pager filesystem I/Os blocked with no fsbuf
with
ioo -p -o j2_nBufferPerPagerDevice=1024
ioo -p -o hd_pbuf_cnt=1024
ioo -p -o numfsbufs=1536
ioo -p -o pv_min_pbuf=1024
and afterwards you might even see a few memory pools instead of
0 memory pools
and the rest of the suggested values should at least be beneficial for performance ... and please dont make the mistake to think if a box is running cio there is no filecaching needed. There is ... depending on the amount of processes connecting on the database from any given application this can be immense. And do yourself a favour - put redo-logs into a separate filesystem with (smaller) blocksize of 512 ...
Rgds
zx
