I benchmarked the "raw" speed using ttcp and a windows port of ttcp called
wsttcp before I benchmarked the file transfers (I didn't post those
results).  I used the -s option in ttcp to take the drives out of the
equation as much as possible.  Throughput averaged ~11.4MB/s over 100Mb and
~60.4MB/s over 1000Mb (no matter which direction).

I purposely added the overhead of the transfer protocols and I/O into the
benchmark stats because I wanted a more realistic picture of what kind of
improvement 1000Mb gave me when transfering files (the primary reason I went
to 1000Mb).

The "raw" speed numbers were neat to have but they didn't give me enough
data about the specific task I was planning to use the network for.
According to the "raw" stats the 1000Mb is ~5 times faster than 100Mb, when
in actuallity I'm only going to see it running ~2-3 times faster when I'm
using it (unless I'm pulling through samba).  The protocol and I/O are the
limiting factors over 1000Mb whereas over 100Mb the network speed is the
limiting factor.

James  

> -----Original Message-----
> From: [EMAIL PROTECTED] 
> [mailto:[EMAIL PROTECTED] On Behalf Of -ray
> Sent: Monday, September 20, 2004 10:49 AM
> To: [email protected]
> Subject: Re: 1000Mb vs. 100Mb benchmark anomaly was RE: 
> [brlug-general]networking ignorance
> 
> 
> The other important (IMO) thing about using ttcp is that it 
> cuts out disk and I/O related bottlenecks, so you really are 
> only testing your "raw"  
> network speed.
> 
> ray
> 
> 
> On Mon, 20 Sep 2004, Shannon Roddy wrote:
> 
> > If you want to do some "raw" benchmarking, you should look 
> into nttcp.
> >  This will give you your "theoretical" max on the network.  
> We use it 
> > all the time to test just about everything.  The advantage 
> is that it 
> > will cut out the overhead of higher protocols like samba, ftp, etc.
> > and give you a true view of the network speed rather than 
> the protocol 
> > speed.
> > 
> > http://www.leo.org/~elmar/nttcp/
> > 
> > 
> > On Mon, 20 Sep 2004 03:03:47 -0500, James Kuhns 
> <[EMAIL PROTECTED]> wrote:
> > > I went ahead and rolled out the 1000Mb network this 
> weekend.  Came 
> > > up with a procedure to attempt to benchmark the file copy 
> > > performance between my two main machines before and after 
> the roll 
> > > out.  I basically used both ftp and samba mounts to 
> transfer files 
> > > between the two machines on my net.  Got a puzzling 
> anomaly with the benchmark stats.
> > > 
> > > It looks like a samba file copy (RH 9 box -> WinXP SP2) is way 
> > > slower on the 1000Mb than it was on the 100Mb (the 100Mb 
> looks to be 
> > > about 1.5 times faster than the 1000Mb), everything else is about 
> > > what I expected (~2 to 3 times faster, with ftp being 
> significantly 
> > > faster than samba).  I reran the benchmark and got basically the 
> > > same results so I did a few copies by hand and got 
> roughly the same as the benchmark.
> > > 
> > > Anyone know what could cause this?  I'm thinking maybe an old 
> > > version of samba (I think it's at 3.0.5 and the one on 
> the box is only at 2.2.8a)...
> > > The RH box is about to be wiped and Woody put on it, I'll 
> rerun the 
> > > benchmark afterwards and see if the anomaly is still there.
> > > 
> > > Here's a link to the benchmark procedure and the last set 
> of results 
> > > http://www.kuhns-la.com:9080/netbenchmark.html - notice the PULL 
> > > stats in the "Results: SAMBA File Copy Throughput in 
> MB/s" section, 
> > > averaged 5.774 MB/s on 100Mb and only 3.868 MB/s on 1000Mb. ??????
> > > 
> > > Sorry for the messy html - I was using Excel to compute 
> the stats so 
> > > I just exported it from Excel as html.
> > > 
> > > If anyone wants the benchmark scripts let me know.
> > > 
> > > James
> > > 
> > > _______________________________________________
> > > General mailing list
> > > [email protected]
> > > http://brlug.net/mailman/listinfo/general_brlug.net
> > >
> > 
> > _______________________________________________
> > General mailing list
> > [email protected]
> > http://brlug.net/mailman/listinfo/general_brlug.net
> > 
> 
> --
> =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
> Ray DeJean                                     http://www.r-a-y.org
> Systems Engineer                    Southeastern Louisiana University
> IBM Certified Specialist            AIX Administration, AIX Support
> =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
> 
> 
> 
> _______________________________________________
> General mailing list
> [email protected]
> http://brlug.net/mailman/listinfo/general_brlug.net
> 

Reply via email to