Walt, www.linuximq.net version is the evolution from Devik->McHardy's IMQ. Roy started a project where he developed his own implementation of IMQ like functionality. I can't tell you more about his version because I never had a chance to give it a try.
The IMQ version from www.linuximq.net that comes from the original IMQ implementation really needs patching the kernel and iptables sources.
Andre
Walt Wyndroski wrote:
So you are saying that I do not need to patch my kernel? I do not understand that statement. I had to recompile my kernel with the imq patch as well as iptables before IMQ would work for me. The way I have always understood IMQ is that it is a virtual network device, a virtual network card if you will. Therefore it seems to me that egress would apply since iptables is only being used to redirect traffic through the virtual IMQ device. IPROUTE2/TC would then shape traffic leaving the virutal IMQ device (egress traffic). This is how I understand IMQ. If I am wrong, please set me straight.
From what you are saying, either IMQ is completely unstable or iptablesand/or the tcp/ip stack is unstable. Not that I am a guru on the internals of iptables or the Linux tcp/ip stack, but I think iptables and the Linux tcp/ip stack is most likely stable. To be fair, I must admit that I have not had a full opportunity to test out your version of IMQ either, only the prior versions.
Who's version of IMQ resides on www.linuximq.org ?
Walt Wyndroski
----- Original Message ----- From: "Roy" <roy@xxxxxx>
To: "Walt Wyndroski" <wdwrn@xxxxxxxxxxxxxxxx>; <lartc@xxxxxxxxxxxxxxx>
Sent: Tuesday, July 06, 2004 10:34 AM
Subject: Re: Simply IMQ
Probably I was the last one who changed imq code. so here is are the facts: Basicaly all imq versions are usefull under aproriate condition, whis is
do
not touch localy generated traffic. ingress nad egress terms are not correct for imq, because it is iptables module, not nic. Just my version hooks on different iptables hooks, and simply ignores all local generated traffic. It cant be crashed with incorrect rules. basicaly only advantage of my version is nore clean way to hook on
iptables,
code is same for 2.4 and 2.6 kernels, and no need to patch, stability
should
be same on both kernels. Now it is hard to say why imq crash, because crashes occur in various
places
not related to this module, it seems like memory leak, but does not like
imq
can have such bug. I suppose there is somethisng wrong with iptables or
tcp
code itself, since imq does big mess with packets by droping and
reordering
then alot.
Anyway imq does not work as I expected, basicaly all forward shaping is quite hard, I was trying to make tcp traffic predictor because else it is too late. It must be sart enough to work I need to adjust predictor delay, and
packets
size. what makes it quite hard to implement.
----- Original Message ----- From: "Walt Wyndroski" <wdwrn@xxxxxxxxxxxxxxxx>
To: <lartc@xxxxxxxxxxxxxxx>
Sent: Tuesday, July 06, 2004 4:48 PM
Subject: Simply IMQ
I've followed this list for quite a long time and have even posted a couple of times. I used the early versions of IMQ from Devik (I think that was
his
name), and it worked well. I only ever got the chance to implement it in
my
test environment. I now need to implement it in my production
environment.
My Linux core router has nine interfaces and has a 27 megabit connection
to
the internet. It is quite busy much of the time. It runs Fedora Core 1
now
but will most likely be upgraded to Fedora Core 2 in the next month or
so.
Now with all that said, here is my question. I see that maintenance of
IMQ
has been passed on a couple of times. I see some people say that IMQ is
not
stable and should not be put into a production environment. My use of
IMQ
a
year ago invovled only egress qdiscs using HTB and SFQ because the
egress
qdiscs were much more powerful and better than the ingress qdisc. The
only
problem that I ever had with IMQ was using the iptables target with both PREROUTING and POSTROUTING. I see Roy has posted that IMQ essentially crashes when doing egress shaping. Is this correct? I've always
understood
egress as outbound shaping/filtering and ingress as inbound shaping/filtering. I say that because I saw in an earlier post by Roy
that
he changed his terminology to INPUT,OUTPUT, and FORWARD. Was he not
using
the terms egress and ingress correctly? I see that the current 'big'
problem
is touching locally generated traffic. What I need to know is which
version
of IMQ is most stable for kernel 2.6? Or even kernel2.4? Is it Devera's? McHardy's? Correa's? or Roy's? Or should I just leave it alone? My
apologies
if I got names wrong.
This is probably a long email just to ask that question, but I can't
seem
to
find an answer from the list archives. I downloaded the whole 46 mb
archive
and essentially read 90% of the posts related to IMQ. I'm just trying to
get
a good understanding of what's happening with/to IMQ.
Thank you in advance for any advice.
Walt Wyndroski
_______________________________________________ LARTC mailing list / LARTC@xxxxxxxxxxxxxxx http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
_______________________________________________ LARTC mailing list / LARTC@xxxxxxxxxxxxxxx http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
_______________________________________________ LARTC mailing list / LARTC@xxxxxxxxxxxxxxx http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
_______________________________________________ LARTC mailing list / LARTC@xxxxxxxxxxxxxxx http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
