Re: system usage stats (Was: Re: Why Not MySQL? ) - Mailing list pgsql-hackers

From Don Baccus
Subject Re: system usage stats (Was: Re: Why Not MySQL? )
Date
Msg-id 3.0.1.32.20000504074216.00f43a90@mail.pacifier.com
Whole thread Raw
In response to Re: system usage stats (Was: Re: Why Not MySQL? )  (Thomas Lockhart <lockhart@alumni.caltech.edu>)
List pgsql-hackers
At 02:26 PM 5/4/00 +0000, Thomas Lockhart wrote:

>From what I've read, the extra cache in the PII/III gives you a 5%
>boost over the Celeron (I'll guess more for some server apps).

For a db server the 100Mhz memory bus of the PII/III probably
wins more than 5% over the 66Mhz memory bus of the Celeron.  This
assumes you're database is reasonably big.  Lots of memory transfers
going on...

>Intel
>still sells the Xeon chips, which have a cache twice as big as the
>PII/III, but I'm not sure the clock has kept pace and it was always
>overpriced wrt performance.

Katmai Xeons come with cache size ranging from 512K (the same as a PII/PIII
Katmai) to 2MB.  That cache, though, is FULL SPEED.  Gets you about 10%
over a PII/III Katmai for server-type benchmarks I've seen.  But they're
expensive.

You can also go 4-way SMP with them, vs. 2-way with PII/III...

Now the new Coppermines have changed things...the Xeon Cu and PII/III
Cu both have identical 256K full-speed cache.  There's not much reason
to buy the Xeon unless you want 4-way SMP, and Intel recognizes this
apparently because the price is only 10% higher for these new parts.

>Not that anyone asked, but imho the best price/performance x86 machine
>has always been a dual processor box one or two clock jumps behind the
>fastest available.

That's what I did, a dual PII450.  I bought them (boxed, fan) for $180
each when PIII450s were $250 or so and PIII 500Es $299 (I bought one
of the latter for a home workstation).  So it's a PII without the
latest matrix instructions for graphics?  How many games will I
run on my web/db server? :)

>You get ~80% more performance for ~5% more cost
>than a uniprocessor at the fastest speed. I haven't looked recently to
>see if there are now uniprocessor machines at the low end that can
>beat the price/performance of the dual-processor setup.

The Athlons do, actually ... but I wouldn't trust the mobos for a
remote server at this point, not until they're proven stable.  They're
tricky to build and finicky about power supplies.



- Don Baccus, Portland OR <dhogaza@pacifier.com> Nature photos, on-line guides, Pacific Northwest Rare Bird Alert
Serviceand other goodies at http://donb.photo.net.
 


pgsql-hackers by date:

Previous
From: "Mitch Vincent"
Date:
Subject: Re: --enable-locale and SET command ...
Next
From: Tom Lane
Date:
Subject: Re: AW: libpq++ build now checks make rc