From: | "Willis, Ian (Ento, Canberra)" <Ian(dot)Willis(at)ento(dot)csiro(dot)au> |
---|---|
To: | pgsql-general(at)postgresql(dot)org |
Subject: | RE: Re: Fast Inserts and Hardware Questions |
Date: | 2001-03-15 02:19:25 |
Message-ID: | D21A20CD84607E409F314E31F0F68D8A0E6377@cricket-be.ento.csiro.au. |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-general |
What you may have to do some research on is how to spread the disk writes
around as you sound like you will be write bound.
WAL will be an writes always at the end of a file. From memory raid4 is
better than raid5 for this kind of write but as the WAL will remain
resonably small in comparison to the db it may be best to stripe and mirror
disks for the best performance.
Is is possible to separate the index file from the database file and place
these on separted disk sets and scsi channels this should create faster io
and index generation (anyone?).
What is the ideal block size on disk for use with postgresql with your
database (anyone?).
--
Ian Willis
-----Original Message-----
From: Orion Henry [mailto:orion(at)trustcommerce(dot)com]
Sent: Thursday, 15 March 2001 11:43 AM
To: adb
Cc: pgsql-general(at)postgresql(dot)org
Subject: [GENERAL] Re: Fast Inserts and Hardware Questions
adb wrote:
>
> Have you done any benchmarks with a prototype of your application.
>
> Based off of some of the numbers I've been seeing with my testing,
> I would not be surprised if a single pIII 1ghz box with a decent disk
> (ata 100 or scsi 160) would handle the load you describe and is way
> cheaper than some big smp box. A dual pIII with a decent raid card
> would get you even farther. Certainly these are not as expandable as
> other options out there but they are worth a look.
>
> What I would be most concered about is a table that grows by a few million
> rows a week, that to me seems like a liability in itself since
> maintenance on that table will get pretty slow after a few months.
>
> Alex.
Right now I have a prototype running on a dual PIII 533 MHz with a scsi
raid and I am getting 2 inserts / second maximum. (To be fair there is
some overhead I did not mention on each insert but not a lot). Now if
WAL or disabling fsync() can get me a 10x speedup that's 20/second which
is almost what I need. Perhaps finding a way to drop some of my index's
and other overhead could get me to 30/second - my minimum.
But yes - it seems that CPU power is the least of my worries. I just
need really fast disks and 7.1 to come out soon.
---------------------------(end of broadcast)---------------------------
TIP 2: you can get off all lists at once with the unregister command
(send "unregister YourEmailAddressHere" to majordomo(at)postgresql(dot)org)
From | Date | Subject | |
---|---|---|---|
Next Message | Willis, Ian (Ento, Canberra) | 2001-03-15 02:26:51 | RE: Re: Fast Inserts and Hardware Questions |
Previous Message | Bruce Momjian | 2001-03-15 02:18:21 | Re: Fast Inserts and Hardware Questions |