I believe you've hit the rough of what the adapters and the drives are
actually capable of. I usually consider around 35% of the top end of
Ethernet to be about the best you can get, and you're there already.
You're also asking ftp to read from the drive in excess of 35MB/sec, and I
doubt the drive is capable of doing that. I just did a quick check on
Seagate's web site for the formatted internal transfer rate on a 10K u160
drive, and runs between 49 and 63 MB/sec. Given that you're going to have
file system and application overhead, you've probably hit the limit. What
made you think that you hadn't reached a disk bottleneck?
I looked for an Atlas 10K 9 GB u160 drive on Tomshardware, and found the
claim of 41MB/sec sustained transfer rate. You've got to get the data off
that drive, do some work, push it to the other side, do some more work, and
acknowledge the transfer. I think that you're lucky to get 35MB/sec.
Stripe the drives and redo the test, or transfer from /dev/null to /dev/null
on both ends with a custom application that doesn't incur ftp overhead and
see what you get.
.../Ed
Ed Wilts
Mounds View, MN, USA
mailto:[EMAIL PROTECTED]
----- Original Message -----
From: "Matthew Boeckman" <[EMAIL PROTECTED]>
> I'm running RedHat 7.1 on some Altus1240's from Penguin Computing. All
> have identical GigE Nic's. All are connected via a Cisco Catalyst 3550
> GigE switch which shows that auto-negotiation has each box happily
> connected at 1000/Full Duplex. The boxes themselves have 10K RPM u3SCSI
> drives in them. The files themselves vary from 400M-1.2Gig and were
> created with dd and help from /dev/zero.
>
> Here's the rub: When I transfer a file via ftp from one box to another, I
> seem to peak out around 35MB/s, which translates rougly to 270Mb/s. A far
> cry from 1000Mb/s. Is the list aware of any tools to do some more refined
> bandwidth tests? Or how about known issues with GigE on RH7.1 or the 2.4
> kernel?
>
> I seem to get this throttle no matter how big or small the file is that I
> transfer. I doubt that the disk is the bottleneck. Top reports system
> utilization climbs up to around .8 or so, with wu-ftpd taking 30% or so of
> the CPU, so I likewise doubt the CPU. Any thoughts?
_______________________________________________
Redhat-list mailing list
[EMAIL PROTECTED]
https://listman.redhat.com/mailman/listinfo/redhat-list
