I made some changes to the way the autotuning worked that made it
infeasible to turn it on or off in the middle of a connection, which is
why this was removed in the later versions.
The next version of the web100 patch will be against 2.4.27 when it is
release, and will have NO autotuning code. Linux 2.6.7 (now) and 2.4.27
(when released) have enough of our code merged in so that it should
perform quite well assuming you have your max buffer sizes set.
(Unfortunately it does not yet work "out of the box" like the web100 patch
does, but we are pushing on this.)
Given the current status of the code in the stock Linux, I may actually be
able to add these controls you're missing back in to the next Web100
patch. I'll put this on my list of things to investigate.
Also of note, this next patch will probably be the last against the 2.4
series. I'm moving everything to 2.6 Real Soon Now.
-John
On Wed, 28 Jul 2004, Lawrence D. Dunn wrote:
> Dear web100 folks,
> I'm wondering if there is a known bug in the
> behavior I'm seeing.
> I'm running the 2.3.7 precompiled code, alpha 1.4 userland.
>
> I disabled sender and receiver autotuning.
>
> (aside: removing the autotuning "checkbox" is a
> significant step backwards for me; John Estabrook
> notes that it's because it confused the autotuning code,
> and he's working on alternative suggestions;
> but I wanted to log the "this is a pain" comment...)
>
> When I access the Sender TX slider,
> and try to shrink the buffer size to reduce throughput, it seems
> to have no effect.
> (In older versions of web100
> code, this worked fine- manually shrinking the sender tx-buf
> after session startup would reduce the throughput).
>
> On the other hand, performing the mirror-operation on the
> receiver side, i.e. reducing Receiver RX bufs, works as expected.
> That is, tiny buffers limit the throughput.
>
> Is this a known bug?
> Or a defense against some other induced problem?
> Any estimate for a fix/workaround?
>
> It's not mission-critical, in that I can "get the point across"
> just using the receiver rx-buff. But I didn't see any documentation
> on the change from previous behavior (haven't really looked),
> and would at least like to be able to explain to folks what's going on...
>
> Thanks!
> Larry
> --
>
> _______________________________________________
> Support mailing list
> Support@xxxxxxxxxx
> http://internal.web100.org/mailman/listinfo/support
>
_______________________________________________
Discussion mailing list
Discussion@xxxxxxxxxx
http://internal.web100.org/mailman/listinfo/discussion
|