Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: [HSSG] Jumbo frames

Here we go again...

I was going to stay out of this debate this time, but since Brad
mentioned my name (in vain), here my 2 cents.

It looks like every few years we are prompted to revisit this issue.
The last attempt was about a year ago when we were approached
by IETF to try to come up with some sort of a joint white paper that
would provide guidance to Ethernet users for using JFs in a "safe
manner". Just as in the past, I don't believe anything came out of it.
Regardless of whether you believe that JF are necessary or not, the
truth of the matter is that there is no good way for allowing the use
of JF in 802.3 without breaking the integrity and the plug-and-play
nature of our standard.

As for my personal opinion, I view the use of JFs as useless at best
and harmful at its worst. I am not going to re-iterate all the problems
that have already been brought up by others. The list is quite long.

Just a few observations from my personal experience:
* I agree that most implementations today support JFs.
* I strongly disagree that they are widely used in real life application
   environments. The vast majority of our customers never enabled it.
* Enabling JFs does not mean that an application is using it. Although
   some applications can benefit from JFs, the vast majority do not.
   There are very few industry application benchmarks that show any
   benefits from JFs.
* Most applications running on servers today need good throughput
   for small packets. In fact, the more interesting metric for network
   throughput is measured in packets/second (rather than bits/second).
   Once you address that problem, the other one comes for free.

Bottom line, this topic for HSSG is way out of scope in my mind.

Enough said,