Re: gbde Performance - 35Mb/s vs 5.2 MB/s

From: John Stockdale <jstockdale_at_stanford.edu>
Date: Fri, 06 Jun 2003 16:26:55 -0700
Hey Everyone,

I've been MIA with regards to the current list lately, but was checking 
the backlog for gbde stuff and came across this thread. Anyway, I've 
been running gbde (thanks to Poul for helping me out along the way) on a 
SMP (2x2.0ghz Xeon) box with a 3ware RAID 5 (8 drive ATA-133 array) with 
caching enabled (8MB per drive). Its been up nicely for about a month.

Running bonnie (1024MB test file) to benchmark performance (as 
recommended by 3ware tech) gives the following results:

-------Sequential Output-------- ---Sequential Input-- --Random--
-Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
   MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
1024  4845 13.7  4781  2.9  3796  2.3 12873 35.3 16287  5.9 648.3  4.3

Additionally, during the write portion of the test g_bde _at_ nice=-20 
takes up 20-30% of one processor ( taken from top -S ) while bonnie 
takes 2-8%. I'm not quite sure what is limiting the cpu usage of gbde, 
but 30% is about as high as I ever can get it to stress any one 
processor during write. Still though not bad performance.

During read I managed to get up to 75% processor usage on gbde and up to 
  45% on bonnie ( not at the same time, for some reason the two 
processes won't thread onto different processers ;)

-John
Received on Fri Jun 06 2003 - 14:27:17 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:10 UTC