[Beowulf] raw results

Gerry Creager n5jxs gerry.creager at tamu.edu
Fri Oct 1 07:23:34 PDT 2004

One note:  Please review RFC2544 regarding packet loss and mitigation 
for small packets.  We tested a number of switches about 18 months ago 
using an Anritsu MD1230 automated tester, and saw a lot of packet loss 
in switches that were not crafted for small packets.  Something to consider.

There are manufacturers who have engineered for both jumbo and small 
packets.  The tuning for this is not trivial: the problems for small 
packets are not well translated to jumbo frames.  As packet size 
decreases, overhead increases and the packet transmission rate goes 'way 

Jumbo frames may adversely impact switch fabric memory if you're testing 
store-and-forward devices not designed with sufficient memory for jumbos 
originally.  Option 'B' is fragmentation, removing the benefit of jumbos 

Thanks, all the same, for posting your results.  We're always interested 
in independent reports and independent methods!


Robert G. Brown wrote:
> On Fri, 1 Oct 2004 050675 at student.unife.it wrote:
>>Hi all,
>>   someone in this list (Robert J. Brown, probably), a few months ago
>>asked me to post one or twice a month if I had interesting results with
>>raw ethernet on which I'm making my first level degree.
>>Now I've some interesting results: packets loss have been decreased (even
>>if under 300 bytes payload I experience so many loss, but probably the
>>problem is in the used 32 bits architecture or in the fact that I use a
>>2.4.x kernel without NAPI, for the moment).
>>For any other payload value (in a range between 300 and 1500 bytes) I've
>>no losses, even with jumbo frames (achieved throughput 111 MB/sec in a
>>point-to-point connection with Gbit ethernet cards).
>>I've also experienced NAPI, but with a kernel 2.4, not 2.6 (next goal).
>>Thank you for your attention and help in these months.
> Cool!
>    robert >>G<< brown  (rgb:-)

Gerry Creager -- gerry.creager at tamu.edu
Network Engineering -- AATLT, Texas A&M University	
Cell: 979.229.5301 Office: 979.458.4020
FAX:  979.847.8578 Pager:  979.228.0173
Office: 903A Eller Bldg, TAMU, College Station, TX 77843

More information about the Beowulf mailing list