[Bioclusters] gigabit ethernet performance

Justin Powell bioclusters@bioinformatics.org
Fri, 9 Jul 2004 14:57:19 +0100


I can move a 1 Gb file between one dell dual xeon box and another in 44
secs on my network (23MB/sec) using scp, but if I move from a dell to a
dual G4 xserve local drive it takes 83 secs (12 MB/sec). Strangely though
if I move it from the dell to the Xraid attached to the G4 it also only
goes at 12 MB/sec - and in that case the mac drive speed should not be the
limiting factor.

Dr Justin Powell

On Fri, 9 Jul 2004, Chris Dwan wrote:

>
> Chen,
>
> More than likely, the bottleneck in your data transfer is the speed at
> which bits are read from the physical disk in your Xserve.  You can
> measure data rates off the disk in a variety of ways.  The one I prefer
> is "iostat."  12 to 13 MB/sec sounds almost exactly like a single IDE
> drive.
>
> You can improve the speed of a data storage device by striping data
> across several drives (in a RAID set), or in a variety of other ways.
>
> Good luck with this.  Data motion is a bottleneck for many of us.
>
> -Chris Dwan
>   The BioTeam
>
> On Jul 9, 2004, at 5:41 AM, Chen Peng wrote:
>
> > Hi,
> >
> > We are runing a Xserve G4 cluster with SMC gigabit switch. I found the
> > network performance is not satisfactory that file transmission speed
> > is always limited to 12MB/s to 13 MB/s.
> >
> > To copy a 800MB file from one host (192.168.101.161) to another one
> > (192.168.101.162) using FTP, it takes almost 70 seconds when the
> > entire network is almost idle. I have verified all NIC and ports on
> > the switch, all of them appear to be "1000base TX". And the disk IO
> > speed is guaranteed to be over 40MB/s.
> >
> > We expect the average speed in gigabit ethernet should be around
> > 30MB/s. Does any one know what might be the problem?
> >
> > Thank you.
> >
> > --
> > Chen Peng
> > <chenpeng@alumni.nus.edu.sg>
> >
>
> _______________________________________________
> Bioclusters maillist  -  Bioclusters@bioinformatics.org
> https://bioinformatics.org/mailman/listinfo/bioclusters
>