Timport time on SunOS 5.8

Post Reply
bart
Posts: 251
Joined: Wed Apr 26, 2000 12:42 am

Timport time on SunOS 5.8

Post by bart »

I timported the "MASTER" table from http://registry.faa.gov/database/AR122001.zip on my sun box and it took about 15 minutes to load using "timport -csv". Its got about 320,000 rows so it loaded at about 350 recs/sec.

Seems too long to me, you too?
User avatar
John
Site Admin
Posts: 2622
Joined: Mon Apr 24, 2000 3:18 pm
Location: Cleveland, OH
Contact:

Timport time on SunOS 5.8

Post by John »

Yes, that does seem a little slow. I just tried the same timport on my home linux box and it loaded ~950 records/sec.
John Turnbull
Thunderstone Software
bart
Posts: 251
Joined: Wed Apr 26, 2000 12:42 am

Timport time on SunOS 5.8

Post by bart »

HMMM. I put the db in /tmp (aka swap). Wonder if that had an effect?
User avatar
John
Site Admin
Posts: 2622
Joined: Mon Apr 24, 2000 3:18 pm
Location: Cleveland, OH
Contact:

Timport time on SunOS 5.8

Post by John »

Don't think so. I just compared two different steps, the timport to /dev/null with no DB ops, and a create table as select * on the existing DB. Both took twice as much user time on sparc than Linux. (700 MHz PIII vs 300 MHz UltraSparc-II). The timport takes twice as long as the table copy.
John Turnbull
Thunderstone Software
bart
Posts: 251
Joined: Wed Apr 26, 2000 12:42 am

Timport time on SunOS 5.8

Post by bart »

In playing with the db I've noticed that SELECTs operate at a pretty good rate. Maybe sparcs just suck at write. Since sun is a big market share for Texis maybe some investigation is warranted.
User avatar
John
Site Admin
Posts: 2622
Joined: Mon Apr 24, 2000 3:18 pm
Location: Cleveland, OH
Contact:

Timport time on SunOS 5.8

Post by John »

We'll certainly investigate some more, although that's why I tried the timport without writing to the DB, and looked at user CPU time. It's almost as if the Sparc and Pentium get the same work done per clock cycle, and the Pentium cycles that much faster, although that's difficult to believe, so maybe there's some missing or different optimizations needed for the Sparc chip.
John Turnbull
Thunderstone Software
Post Reply