North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

RE: Jumbo Frames (was Re: MAE-EAST Moving? from Tysons corner toreston VA. )

  • From: Roeland Meyer (E-mail)
  • Date: Mon Jun 19 12:22:54 2000

> Richard A. Steenbergen: Monday, June 19, 2000 2:57 AM

> don't), we have to actually ask for it. Imagine an internet
with a
> reliable MTU negiotation mechinism, which can take advantage
> of improved
> thruput, much lower CPU usage, zero copy, page-flipping, DMA
> transfers,
> and all those other lovely things.

Yeah ... <heavy breathing>

> These are important for many reasons. Without these
> techniques, we can't
> even do line rate GigE on "common place" servers, let alone
> have any CPU
> left over to do more then just send packets.

Actually, my testing shows a falure to utilize even 100baseTX
fully. Even in a switched FDX environment (no collisions) I can't
achieve line rate without bumping the packet size up. Considering
that the smallest box is a quad-CPU SMP machine (550Mhz), I don't
think that there is a CPU shortage <grin>.

> Its easy to just
> say "we'll
> throw a server farm at it" or "we'll just get a faster
> processor", but as
> higher speed links become more common place, and as GigE
> becomes common in
> servers (when servers can actually use it effectively)

In this case, the common problem is the RDBMS host. It is very
difficult to cluster them, due to limitations with most RDBMSs.
The result is that this host sources most of the packets and with
MTU=1500 it is throttled at about 40% of line-rate, or less,
depending on transfer size.

> and
> 10GigE becomes
> commonplace for backbone links, we'll start to see these
> things matter.
> Why engineer ourselves into a corner of shortsightedness
> which only gets
> harder and harder to fix, because its "easier" to do nothing?

I don't have a 10gig-E system, but I wonder about going there
when I can't even get gig-E to work efficiently. If vendors want
to sell 10gig-E they should be concerned about exactly this
point. Joe SOHO isn't going to buy it anyway. Joe Enterprise
isn't going to spend the extra money unless he can see some real
benefit, and Joe dot-com ain't going to do it unless it is
measurably faster than gig-E (which it won't be with MTU=1500). I
can aggrigate 3-5 gig-E links to get the same troughput, by
adjusting MTU, and not pay the 10gig-E meal-ticket. BTW, the
selling feature on gig-E is link aggrigation, built into the spec
(over Fast-E), there is no similar feature enhancement for
10gig-E, AFAICT. Evenso, it is still limited by MTU size.