realy bad network performance with PowerEdge T610

Blake Hudson blake at
Mon Feb 21 10:21:35 CST 2011

The SAS2008 cards have no cache, while linux softraid will use system
RAM as a write cache by default. Additionally, most 'desktop class'
hardware does not honor fsync (commit to disk) requests and operates in
a forced write-back mode. I'm guessing this difference in behavior is
causing the high disk activity as well as IO wait you are experiencing
on the Dell. Applications that issue frequent fsyncs (commits) will
suffer noticeably because of this.

The Dell hardware options you have chosen will write data to the disks
'safely', while the previous HW configuration probably favored speed
over data integrity. If you want to go back to speed, you can
reconfigure the disks as single drives and go back to Linux soft-raid.
You may also be able to disable fsyncs in your NFS exports. Another
option is to use a HW RAID controller with battery backed cache in
write-back mode - like the PERC5/6 or H700/800 cards - these would
enable fast operation while maintaining data integrity.

Mount/Export options with 'sync' will honor disk commits and write to
disk safely; 'async' options will prefer speed, but may result in data
corruption in the event of system crash/power loss.


-------- Original Message  --------
Subject: realy bad network performance with PowerEdge T610
From: Michael Zoet <Michael.Zoet at>
To: linux-poweredge at
Date: Monday, February 21, 2011 9:45:03 AM
> Hi to all,
> for a small office I have setup a new PowerEdge T610 server: 6 core
> Xeon X5650 2.67GHz CPU, 8 GB RAM, 2 1TB SATA drives as hardware RAID1
> on a SAS2008 controller. On the server runs Ubuntu 10.04 with Dells
> OMSA, LDAP, SAMBA, NFS, Apache and some other stuff a modern office
> needs to have.
> The server should replace an old system (Athlon Sempron  1,8 GHz,
> several SATA drives as RAID1 with Linux kernel software RAID md, 2 GB
> RAM). If I setup the new server to serve as a fileserver with NFS
> (/home dir for the Unix clients) and SAMBA (Windows client) the
> performance goes down to nothing. Which means: the clients respond
> with a big time lag. Login in to a Linux client can take about 20
> minutes, network file operation with SMB shares are awfully slow.
> Starting and working with Thunderbird is really slow even if you turn
> of the indexing. Firefox is responding very slow and the client
> systems have a high I/O load. The harddrives in the T610 make really
> annoying and loud noises. So as the hardrive heads are continously and
> fast moving around. In the end this delays the data delivery to the
> network and causes a high I/O wait on the clients...
> Has someone here any clue what I can do to make things work? Maybe
> some hardware and BIOS changes to get the harddrives behave normaly?
> With the old server (with much cheaper hardware) the users have no
> problem working on their clients. Data is delivered over the network
> as fast as possible.
> For other network services (LDAP, DNS, Mail, groupware web
> applications and so on, KVM virtualisation) the server behaves very
> good and fast. It is only NFS and SAMBA that breaks everything...
> Hope someone has some tips,
> Michael

Linux-PowerEdge mailing list
Linux-PowerEdge at
Please read the FAQ at

-------------- next part --------------
An HTML attachment was scrubbed...

More information about the Linux-PowerEdge mailing list