At 03:11 PM 03-05-2001 -0600, Steve Wolfe wrote:
>
> Since I'd rather have a screwdriver than a compiler, I'll jump in on a
>response to the original message....
Great at least a relevant response ;).
> Once you've got a gig of RAM and you're using SCSI disks (preferably
>RAID), the CPU's tend to be the bottleneck. On our system (4 Xeon 700's,
>RAID, 512 MB), the disk lights only blink *occasionally*, and we have
>hundreds of megs of file cache, no swapping. We're going to increase the
I'm wondering are the CPU's really the bottleneck here, or is the main RAM
bandwidth the bottleneck? How does one tell? (at least without comparing
performance by swapping in and out CPUs of different speeds - too bad most
CPU clocks are now locked multipliers).
Because I thought that once you got 1GB of ram AND your whole DB fits on
it, the bottleneck is likely to be the 133MHz memory bus not the 1GHz cpu.
E.g. a faster CPU won't help much unless the indexes or "to be sorted"
result sets fit in the 256K/512K 2nd level cache. In comparison I believe
SPARCs have 8MB of 2nd level cache - which should be helpful for DB stuff.
So, I was wondering what the performance would be like if you had 266MHz
DDR RAM instead.
Anyone thinking of running postgresql on Athlon servers with 266MHz DDR
RAM? Any pgbench figures (with fsync off)?
Hmm, any interleaved DDR RAM motherboards out there- when you have 533MHz
memory bandwidth the CPU caches are just to improve latency ;).
Cheerio,
Link.