Re: Pipeline Production Problems

[Robert Creager <Robert_Creager@LogicalChaos.org>]

When grilled further on (Mon, 13 Oct 2003 16:01:08 -0500),
"Thomas Droege" <tdroege2@earthlink.net> confessed:

> 
> >From the looks of things, the transmission speed goes down as I start each
> transfer process going.  I could add some extra cables so that the possible
> routes had parallel cables if the system is smart enough to make use of
> them.  

I'm wondering if the transmission speed is either CPU or 'hub' related.

I just did a few tests.  I created a 2Gbyte file from /dev/urandom, and did a
scp between my server with a 1000Mbit pipe going into a 100Mbit switch through
three other 100Mbit switches to an PII 400Mhz 128Mbyte ram in 7:40 minutes
(4.4Mbytes/sec).  Then to my AMD K62 500Mhz 664Mbyte ram through one switch in
19:58 minutes (1.7Mbytes/sec).  The AMD was definitely CPU bound. I didn't pay
attention to the P2 CPU usage.  The server is a 2xAMD 2600+ with 2Gbytes memory.

I then went to NFS, and managed the 2Gbyte copy in 3:25 minutes to the PII, or
just shy of 10Mbytes/second, which is pretty darn close to the theoretical
maximum of a 100Mbit connection.  The AMD clicked along quite nicely in 11:54
minutes (2.9Mbytes/second), with out being CPU bound.

As a last try, I did both at the same time.  The PII completed in 4 minutes for
8.5Mbytes/second, and I'm going to bed before the AMD will finish.  I am doing
the NFS mount from an IDE disk, not my SCSI disk, which might be why it took an
extra 35 seconds.

What's all this say?  My AMD has a crappy network card and Tom shouldn't be
seeing significant slowdowns unless he is running hubs, or has crappy network
cards also...  I suspect his PC's are all faster than my PII and AMD, and should
be able to maintain full bandwidth.

Cheers,
Rob

-- 
 22:05:15 up 73 days, 14:32,  5 users,  load average: 2.50, 2.53, 2.55

PGP signature