SUMMARY: L9 LTO Performance Problem

From: David Eisner <>
Date: Fri Aug 02 2002 - 12:22:48 EDT
I haven't resolved the problem yet, but I think I now know where
the bottleneck is.  Much thanks to Jay Lessert, Darren Dunham,  
Abhilash. V.M., and Peter Wallis. 

Everybody figured out that the limiting factor is probably not the tape
drive, but rather writing to the filesystem.  Part of the problem can be
creating the inodes for many small files, but even dd'ing one big file
from tape to an A1000 is slow (maxing out at about 5.8 MB/sec for dd'ing a
single 2.5GB file off tape, as I mentioned).

BUT, if I dd to /dev/null instead, I get about 14.8 MB/sec.  I also
created a 1GB file of just repeated A's, to test compression.  
dd'ing that to /dev/null from tape gave me about 28MB/sec throughput.
When I dd the same file to the A1000, I get 7.2 MB/sec throughput.
When I dd to the internal SCSI disk, I get between 6.3 MB/sec and 
7.2 MB/sec.  It's an old disk, and I'm still researching whether that's 
the best it can do, or whether there's some other bottleneck within
the system.

Regarding the A1000's, they're doing Raid 5, which I forgot to mention.
Another important fact I failed to mention (sorry), was that the
batteries have expired, disabling (at least) write-back caching.  
I'm ordering replacement batteries.

If people are interested I'll post another summary when everything
is resolved.

Thanks again!

sunmanagers mailing list
Received on Fri Aug 2 12:26:34 2002

This archive was generated by hypermail 2.1.8 : Thu Mar 03 2016 - 06:42:50 EST