Re: 2.2.0 wishlist, new version

Matthias Urlichs (
Thu, 11 Jul 1996 07:29:38 +0100

In, article <
Alan Cox <> writes:
> > > * Guaranteed stream throughput (for example, reserve 300 kB/=
sec for
> > > a
> > Dito, Alan is doing that. (Whence he geht all the time to do all th=
e things
> > he does is completely beyond me.)
> No Im doing a downward limiter so you can say "Client XX gets 64K". I=
t makes
> no guarantees that buffers are preallocated or that the user doesnt
> overcommit the link. Guaranteed throughput is an extremely complex it=

True. However, the first problem of guaranteed throughput is that nobod=
else is grabbing more bandwidth than available. Your patch will allow m=
e to
route everything else (on the local host...) through a limiter or two,
thereby solving _that_ problem.

Thus, _if_ everything else works out right, then we get something
resembling guaranteed throughput with what you're doing. Guaranteeing t=
all these other things indeed do work out right is the next problem. ;=

> Firstly its a soft realtime problem, secondly its a tricky resource
> management problem and thirdly you can't do it on a shared network bu=
> (eg ethernet).
All too true... but there are enough cases out there where all of this
isn't really a problem.

Dime is money.
Matthias Urlichs \ noris network GmbH / Xlink-POP N=FCrnberg=
Schleiermacherstra=DFe 12 \ Linux+Internet / EMail: urlichs@nor=
90491 N=FCrnberg (Germany) \ Consulting+Programming+Networking+etc=
PGP: 1024/4F578875 1B 89 E2 1C 43 EA 80 44 15 D2 29 CF C6 C7 E0 D=
Click <A HREF=3D"">here</A>. =