Re: [PERFORM] Postgresql on an AMD64 machine

2005-06-08 Thread Sam Vilain

Joshua D. Drake wrote:
Yes - we have seen with oracle 64 bit that there can be as much as a 
10% hit moving
from 32 - but we make it up big time with large db-buffer sizes that 
drastically
Well for Opteron you should also gain from the very high memory 
bandwidth and the fact that it has I believe "3" FP units per CPU.


Sure.  But you get those benefits in 32 or 64-bit mode.

Sam.

---(end of broadcast)---
TIP 3: if posting/reading through Usenet, please send an appropriate
 subscribe-nomail command to [EMAIL PROTECTED] so that your
 message can get through to the mailing list cleanly


Re: [PERFORM] Prefetch

2005-05-10 Thread Sam Vilain
Matt Olson wrote:
Other databases like Oracle and DB2 implement some sort of row prefetch.  Has
there been serious consideration of implementing something like a prefetch
subsystem?  Does anyone have any opinions as to why this would be a bad idea
for postgres?
Postges is great for a multiuser environment and OLTP applications.  However,
in this set up, a data warehouse, the observed performance is not what I
would hope for.
Oracle doesn't pre-fetch data to get its fast results in this case.
pre-fetching doesn't give you the 100 times speed increases.
Bitmap indexes are very important for data mining.  You might want to see
   http://www.it.iitb.ac.in/~rvijay/dbms/proj/
I have no idea how well developed this is, but this is often the biggest
win with Data Warehousing.  If it works, you'll get results back in seconds,
if it doesn't you'll have plenty of time while your queries run to reflect on
the possibility that commercial databases might actually have important features
that haven't even penetrated the awareness of most free database developers.
Another trick you can use with large data sets like this when you want results
back in seconds is to have regularly updated tables that aggregate the data
along each column normally aggregated against the main data set.
Of couse, Pg doesn't have the nice features that make this just work and make
queries against your data source faster (called "OLAP Query rewrite" in
Oracle), so you'll have to put a lot of work into changing your application to
figure out when to use the summary tables.  As far as I know it doesn't have
materialized views, either, so updating these summary tables is also a more
complex task than just a single REFRESH command.
Maybe some bright person will prove me wrong by posting some working
information about how to get these apparently absent features working.
You might also want to consider ditching RAID 5 and switching to plain
mirroring.  RAID 5 is a helluva performance penalty (by design).  This is
why they said RAID - fast, cheap, reliable - pick any two.  RAID 5 ain't
fast.  But that's probably not your main problem.
Sam.
Regards,
Matt Olson
Ocean Consulting
http://www.oceanconsulting.com/
---(end of broadcast)---
TIP 1: subscribe and unsubscribe commands go to [EMAIL PROTECTED]
---(end of broadcast)---
TIP 2: you can get off all lists at once with the unregister command
   (send "unregister YourEmailAddressHere" to [EMAIL PROTECTED])


Re: [PERFORM] Whence the Opterons?

2005-05-08 Thread Sam Vilain
Mischa Sandberg wrote:
After reading the comparisons between Opteron and Xeon processors for Linux,
I'd like to add an Opteron box to our stable of Dells and Sparcs, for 
comparison.
IBM, Sun and HP have their fairly pricey Opteron systems.
The IT people are not swell about unsupported purchases off ebay.
Anyone care to suggest any other vendors/distributors?
Check out the Tyan Transport systems.  Tyan are an ex Sparc clone 
manufacturer, and
released the second available Opteron board - widely considered the first 
serious
Opteron board to hit the market.
Sam.
---(end of broadcast)---
TIP 3: if posting/reading through Usenet, please send an appropriate
 subscribe-nomail command to [EMAIL PROTECTED] so that your
 message can get through to the mailing list cleanly