List:General Discussion« Previous MessageNext Message »
From:mac Date:July 8 2004 5:33pm
Subject:Re: INTEL vs. SPARC
View as plain text  
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

hi there,

to keep you up to date:

we tested with a simple dd-read on different machines and mount-points:

400MB file (created with "dd if=/dev/urandom of=file100Mb bs=512 
count=800000" on the intel-linux-box):

reading the file with dd to /dev/null on
a) the sparc-solaris:
local-disk: 12sec
veritas-disk: 24sec (over fiber on symmetrix)

b) the intel-linux
local-disk: 2sec

the intel-box does have some pretty modern scsi-disks in a raid-5.


the intention was to get a feeling how much the hd's will be a part of 
the problem. we didn't want to get some absolute benchmarks :-))


so it points to the disks.


but now somethin really cool:
we took a closer look on our table and tried to figure out why the 
disks are so important for our select. if you remember: it was a 
varchar-field that took us some time. the intersection of this 
varchar-field an an integer-field was one of the major results we 
wanted.
both field for themselfs had around 400.000 rows in the result-set. the 
intersection (WHERE myvarchar="mac" AND myint=1) only gives 130 to 500 
rows.
so we finally did something we should have done long before :-(
we changed the varchar into a int (creating an extra table for 
translation) and created a combined index over both fields. the first 
step brought us 5 to 8 secs speedup and the combined index just got us 
crazy. it now takes 0.01secs.....


so for one specific statement the combination of better to index types 
and combined index with better cardinality helps a lot.

but there is still the dependency to the disks.
if the intersection of both int-fields wouldn't be such a small result, 
i am pretty sure the speed would be as bad as before.



we will work on it a little bit further.
but if you have some comments on that: please help us :-)


thanks
mac
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (Darwin)

iD8DBQFA7YV9vkHn/oGTPXURAmE/AJ4gE6moPl7a6guloK9SErdCzGwIJACgmbsO
xXbcdWEQKxmfvo5ijF7ml+U=
=NFsh
-----END PGP SIGNATURE-----

Thread
Re: INTEL vs. SPARCmac2 Jul
Re: INTEL vs. SPARCmac8 Jul
  • Re: INTEL vs. SPARCmac8 Jul
    • Re: INTEL vs. SPARCDan Nelson8 Jul