Re: network traffic doesn't balence well on recent kernels

Nate Riffe (inkblot@geocities.com)
Sat, 5 Jun 1999 16:45:23 -0500 (CDT)


On Sat, 5 Jun 1999, Steve Bergman wrote:

> Jeff Millar wrote:
> >
> > When using the internet with late v2.2 and recent v2.3 kernels, the
> > first process to get the link seems to hog the bandwidth. My
> > configuration consists of a P2-266 running RH5.2 with many upgrades plus
> > a router/dialer running on a 486/66 and stock RH5.2. This doensn't
> > happen with W98 using the router. For example, right now I have two
> > download goings. The first started at 2.2K per second...about what my
> > phone line will allow. The second took three tries to start (timeouts
> > at the ftp site) and then started very slowly. After about 5 minutes,
> > the first connection now reports 1.9K and the second reports
> > 800....they're slowly drifting to equality. In general, once a download
> > starts, interactive use via a browser window gets very, very sluggish.
> >
> > Have I missed something on the this mail list about network tuning???
> > Reviewing config options didn't turn up anything unusual or changed
> > recently. QoS isn't selected, Not optimizing as router.
> >
>
> I've noticed this for a long time on my ppp connections. The first ftp
> transfer to get started hogs the bandwidth and subsequent ones get very
> little but very gradually improve. I would much rather have it the
> other way around so that what I am doing *right now* in the forground is
> fast and big ftp transfers get low priority. I really meant to ask
> about this a long time ago, but am just now getting around to it. BTW,
> does the reciever's tcp/ip stack have control over the priority of
> different connections or is that a sender thing only?

Hi Jeff, All,

Flow control exists on both ends of a TCP connection because the local
network at either end can get congested independently of the other. I
haven't noticed this problem (I'm running 2.2.9 and 2.3.5+devfs) on my
boxen, but I'm connected via ethernet and there are slower links upstream.
Perhaps there are retransmission timing issues with slower links?

-Nate

>
> -Steve
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.rutgers.edu
> Please read the FAQ at http://www.tux.org/lkml/
>

------------------------------------------------((\))<----------------------
Nate Riffe Duct tape by any other name is just as
inkblot@geocities.com sticky.

-----BEGIN GEEK CODE BLOCK-----
Version: 3.12
GCS/IT/M/S/O d-@ s-:- a--->- C++ UL++++BS++>$ P+>+++ L+++>+++++$ W+ N !o
K- w(---)$>-- M-(--) V(--) PS+ PE Y+ PGP>++ t(+)@ 5 X@ R tv>! b+>+++ DI++
D e>++(+++) h r++ y?
------END GEEK CODE BLOCK------

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/