SUMMARY(2): 130gig file, poor filesystem access performance ..

From: Tim Chipman <>
Date: Tue Jul 22 2003 - 14:49:55 EDT
Hi all,

In light of one response I got to my summary, as well as a bit of 
testing done since that time, I thought I should recap with a bit more 
info, to be more fair/accurate.

My posting was regarding problems of poor performance observed when 
accessing (cp, ftp, netbackup) a large (130gig) file on an external 
raid5 disk brick connected to a sol8X86 box via an adaptec U160 SCSI HBA 
; such problems were NOT observed when identical file access was 
performed against a "small" (50 gig) file on the same system.

-> In my original summary, I whinged and moaned that "Sun ought to do 
something about the cadp160 driver". In fact, it is true that Sun can't 
do anything about this driver, since it is written & maintained by 
Adaptec. If, of course, sun felt like writing a replacement driver to 
usurp the function of cadp160, I wouldn't be unhappy, but I'm not 
holding my breath :-) [in fact, it was pointed out to me that based on 
my experience so far, there is no conclusive proof that cadp160 is the 
cause of my problems - which is true ; it is just one suspect on my list 
of candidates].

-> Since posting the summary, I upgraded the cadp160 driver from version 
1.1.0 as bundled with solaris8X86 10/03 to the newest available driver 
(version 1.2.1, as downloaded from adaptec's website.)  Performance for 
the "before" vs "after" for cp,ftp,netbackup are summarized:

                  Performance of "cp between-2-slices-on-disk-array" for
   driver used                 130gig file       50gig file
cadp160 driver v.1.1.0         1.47mb/sec       4.7mb/sec
cadp160 driver v.1.2.1         2.51mb/sec       4.7mb/sec

(note that nothing else changed other than the cadp160 driver version)

These ##'s suggest performance is still significantly slower for the 
same action against this large vs. small file ; however the driver 
update did appear to have some impact on performance.

Further testing (still pending) is to swap in a different (Ultra40 
adaptec HBA, which would use cadp as the driver (ie, a totally different 
driver that was written by Sun, I believe) -- and see if any difference 
in performance persists with manipulation of the large vs small file.

For the moment, however, I've deleted this monster file and will 
probably leave further testing on the back burner, since we infrequently 
create such large files. The primary concern was that it represented a 
larger underlying problem that could result in other (more severe) problems.

as an aside: deleting the 130 gig file took ~10 minutes, and for the 
duration of the process, the system was effectively "hung". In contrast, 
deleting the 50 gig file was effectively instantaneous and had no impact 
on system responsiveness. Not sure if anyone else has seen this 
behaviour or not ever. Great fun.

Finally: as recommended by one respondant to my summary, I did submit a 
problem report to adaptec, who assure me that nobody has ever had this 
kind of problem. Yay.

So. Hope this info is of slight interest to someone, if not now then 
maybe in the future.


--Tim Chipman
sunmanagers mailing list
Received on Tue Jul 22 14:49:47 2003

This archive was generated by hypermail 2.1.8 : Thu Mar 03 2016 - 06:43:16 EST