[clue] Link speed test

[Jim Ockers]

Hi9,

I guess I would caution anyone trying this method of determining network 
performance that the disk might be the slowest part of the system. I 
think the only way to saturate a gigabit ethernet connection while 
writing to disk, for example, is to have RAIDed hard drives on separate 
controllers such that the disk throughput could be as fast as the network.

I think the whole idea of iSCSI is that the network might be way faster 
than the physical disk, so you aren't really losing much performance 
compared to a real disk, because the real disk locally connected is 
about the same speed as a real disk connected via iSCSI. If you have 
wizzy super-fast iSCSI target servers with fancy RAID arrays, your iSCSI 
"disk" performance might actually be faster than a locally connected 
disk, because the filer can give you I/O throughput of many disks.

You should use caution when trying to characterize network performance 
in any way that involves a physical disk. As Dave notes, the performance 
is as slow as the slowest part of the whole system, and I'm postulating 
that the storage device might be way slower than the network, especially 
if you have gigabit ethernet. If you can write to /dev/null on the 
remote system that might be better, because there is no disk I/O. For 
example this will copy 500MB over the network using ssh but not use any 
disk I/O on either system:

dd if=/dev/zero bs=1024k count=500 | ssh user at remote "dd of=/dev/null"

I'm not sure if ssh is the best tool so if you really want to know your 
link speed, you should probably just bite the bullet and use use a 
purpose-written package like netperf. By the way I've used variable size 
ping packets in the past to get an idea of link speed for slow links 
like malfunctioning T1s.

Jim
-- 
Jim Ockers, P.E., P.Eng. (ockers at ockers.net)
Contact info: http://www.ockers.net/

David L. Willson wrote:
> Here are the commands I use to test sustained read and write throughput.
>
> # write speed, presumes device to test is exclusively mounted at /mnt (or that other sources of I/O are inconseqential)
> dd if=/dev/zero of=/mnt/test.dd bs=4M count=256
>
> # read speed, same presumptions
> dd if=/mnt/test.dd of=/dev/null
>
> Both commands output a summary of I/O speeds when they complete.
>
> These commands to work equally well to test network throughput, but network throughput is the slowest of the network and any involved storage devices, minus a bit for management. I'll be trying netcat and netperf as time allows.
>
> --
> David L. Willson
> Trainer, Engineer, Enthusiast
> RHCE Network+ A+ Linux+ LPIC-1 Ubuntu
> Mobile 720-333-LANS(5267)
>
> This is a good time for a r3VOLution.
>
> ----- Original Message -----
>   
>> So, to test hard drive transfer speeds, I use dd. I'll send the
>> commands when I'm on a real computer.
>>
>> What do you use to test network speed?
>>
>> --
>> David L. Willson
>> Trainer, Engineer, Enthusiast
>> RHCE Network+ A+ Linux+ LPIC-1 Ubuntu
>> Mobile 720-333-LANS(5267)
>>
>> This is a good time for a r3VOLution.
>> _______________________________________________
>> clue mailing list: clue at cluedenver.org
>> For information, account preferences, or to unsubscribe see:
>> http://cluedenver.org/mailman/listinfo/clue
>>
>>     
> _______________________________________________
> clue mailing list: clue at cluedenver.org
> For information, account preferences, or to unsubscribe see:
> http://cluedenver.org/mailman/listinfo/clue
>   


-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://cluedenver.org/pipermail/clue/attachments/20121217/4808cf11/attachment.html