lkml.org 
[lkml]   [2006]   [Jul]   [3]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectHow to explain a strange NFS behavior
Hi, folks,

I benchmarked NFS V3 and found a strange thing. I benchmarked with
command "tar xvf linux-2.6.12.tar" and record the total time as well
as the time used on several important system calls. NFS directory is
exported in async mode. The result is listed below:

number of NFS3 NFS3
being called 1st run 2nd run Ext3
(ms) (ms)
(ms)

stat64 23 1249 64484 35
fstat64 11 2078 5399 4
utime 18436 1290737 1266746 169427
open 17371 2747423 3160002 913427
read 20802 7269517 11566103 10036876
write 54784 611536 723411 1594104
close 17386 26698315 12052766 61096
mmap2 5 17 21 15
munmap 8 122 136 82
---------------------------------------------------------------------------------------------------
Total time: 37s 25s 22s

Note that I added some fast communication mechanisms to make NFS
client and server communicate very fast.

Now the problem is: the second run of NFS was much faster than the
first run. Because I didn't flush the disk cache at the server side,
intuitively I think the performance difference should come from the
sys_read() as the tar file data should be already in disk cache for
the second run. However, the results showed that sys_read in the
second run was even slower than the first run. BUT, sys_close of the
second run was much faster! Why? I am really confused.

Please help!

Also, I ran the same benchmark on ext3, it spent more time on sys_read
and sys_write, but sys_close was much faster. I guess that's because
NFS and Ext3 use different running model. NFS always tried to cache
data at client side and flush data until file is closed. This can
send a lot of data to the server in a bursty manner. On the contrary,
the ext3 file system periodically flush data to disk and amotizes the
data write cost and make disk write executed with untar
simultaneously. Is this the real reason? If not, how to explain these
interesting running patterns of NFS and Ext3?

Again, many thanks for kind help!

xin
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2006-07-03 06:07    [W:1.028 / U:0.300 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site