xfs
[Top] [All Lists]

xfsdump slow on large filesystem.

To: linux-xfs@xxxxxxxxxxx
Subject: xfsdump slow on large filesystem.
From: Eugene Melamud <melamud@xxxxxxxxxxxx>
Date: Mon, 18 Jul 2005 11:17:58 -0700 (PDT)
Reply-to: melamud@xxxxxxxxxxxx
Sender: linux-xfs-bounce@xxxxxxxxxxx
Greetings, 

I am getting really miserable results from xfsdump when mirroring large 
filesystem over network.
The command I run on the source computer is this.

xfsdump -A -J - /dev/sda1 | ssh node22 xfsrestore -J -A - /backup

Two computers are connected via 1Gbit network. Connection is good, when I test 
transfers with
rsync, I get transfer between two computers at 35Mb/sec easy on large files. 
The maximum read
speed on /dev/sda1 measured with dd is about 45 Mb/sec. The maximum write speed 
on /backup raid
disk is about 100Mb/sec.

Given that the most file on the filesystem are small, I can not expect very 
high throughput. I was
hopping for at least 10Mb/sec. What I get is 35G transferred in the last 16hrs, 
that's less than
1Mb/sec. 

There is an initial slow down when xfsdump computes tree and inode attributes 
for transfer, but
after that I would think it should really kick in. May be it has something to 
do with operating
system, ( RHEL3 on source machine, CentOS4 on destination, xfsdump version 
2.2.25-1). I know I
know,I should have gone with SUSE.

I am new to xfsdump, so may be I am doing something wrong. If I can't figure 
this out, I'll just
go back to using "cp -au" over nfs. :) Any advice is appreciated. 



<Prev in Thread] Current Thread [Next in Thread>