[TriLUG] Recommend some 'big iron'?

John F Davis johndavi at us.ibm.com
Mon Feb 4 12:55:03 EST 2002


Hello

Let me get this straight.  You compared two different benchmarks.  One
running in a JVM and the other
written as native C - and the JVM version won.  Did the the C programmer
purposely write the code
bad or something?

For my own curiosity, I would like to see the java and C used for the test.

JD

James Manning <trilug at sublogic.com>@trilug.org on 02/04/2002 12:20:40 PM

Please respond to trilug at trilug.org

Sent by:    trilug-admin at trilug.org


To:    trilug at trilug.org
cc:
Subject:    Re: [TriLUG] Recommend some 'big iron'?



[Andrew Perrin]
> > If performance is of the utmost importance, are you sure you want
> > PostgreSQL?
>
> What would you recommend? I do want the transactions, conssitency, etc.,
> that pg provides, so I'm not likely to use mySQL. And I would very much
> like the system to be all-free (as in speech), so I'd like to avoid
> oracle, etc.  Any other ideas?

Keep in mind that much like java itself, performance problems in the
past tend to hang around as false impressions in the present/future.
There have been multiple benchmarks showing postgres faster for many
workloads, sometimes by huge margins.

http://apachetoday.com/news_story.php3?ltsn=2000-08-14-008-01-PR-MR-SW

If you can get an quick idea if the kinds of data and queries you'll be
dealing with, don't listen to anyone else - benchmark it yourself!

Every DBMS will have cases where it beats up everyone else.  What's
important is what's fast and works well for your load.

If you expect large transactions and complex joins, toss those in to
your testing.  If lots of updates, do that, etc.

A close second is to get a "feel" for what (oltp, dss, etc. kinds of
tags would be a little too coarse on the granularity, but it depends
on how much time/effort you can dedicate to this call) kind of load
you'll be dealing with, find a close-ish benchmark (tpcc for oltp,
tpc-r/-h/whatever for dss, etc) and see how things work there.  Finding
benchmarks of TPC benchmarks for the free dbms's could be tougher, so
you may have to make do with what's out there, although I'd still
recommend doing your own benchmark if at all possible :)

It's kind of funny, because a guy was trying to convince me last night
of how compiled C code could do this set of operations so much faster
than Java, and when I had him actually benchmark, the ibm jre won by 30%

Never trust the opinions/benchmarks/etc of other people when it comes
to your own performance issues.  Benchmark yourself.  No else's load
will match yours.  There is no spoon.

If you actually read this entire post, you deserve more than just a
cookie - you deserve my sincere sympathy.

James
--
James Manning <jmm at sublogic.com>
GPG Key fingerprint = B913 2FBD 14A9 CE18 B2B7  9C8E A0BF B026 EEBB F6E4
_______________________________________________
TriLUG mailing list
http://www.trilug.org/mailman/listinfo/trilug





More information about the TriLUG mailing list