Цитат от Віталій Тимчишин <tivv00@gmail.com>:
> May be you have very bad disk access times (e.g. slow random access)? In
> this case everything should be OK while data in cache and awful, when not.
> Could you check disk IO speed && IO wait while doing slow & fast query.
>
No, I think all is ok with disks. On my test server I have 8 SATA in
RAID 10 and on my production server I have 16 SATA in RAID10 dedicated
for pg data and also 8 SATA in RAID 10 for OS and pg_x_log and I do
not have any IO wait.
It is true, disks are much slower compared to RAM.
> BTW: In this case, increasing shared buffers may help. At least this will
> prevent other applications & AFAIK sequence scans to move your index data
> from cache.
I will try to increase this value.
I think recomendation in docs was 1/4 from RAM, and on production
server I have it setup to 1/4 from RAM ( 32 GB).
Will os not cache the data from shared buffers for second time ?
The next step will be to move to pg 8.4, but I i twill tak etime for testing.
>
> Best regards, Vitalii Tymchyshyn
>
regards,
ivan.
-------------------------------------
http://www.tooway.bg/
