On Mon, 2005-11-07 at 13:58 +0100, Grzegorz Nosek wrote:
> The test cases I've found are:
> - no QoS at all
> everything runs smoothly
> - root qdisc on eth0 set to htb (or pfifo, or sfq)
> still everything nice and smooth
> - default class on eth0 set to htb (rate 1Gbit, so it's not true
> bandwith choking)
> mysql connections seem to take forever to establish (lots of
> unauthenticated user connections as they are too throttled to even log
> in and time out eventually)
Strange. Certainly in terms of firewalling, packets that don't leave
the host always travel via the loopback interface.
I'd be curious if the same thing happens without vserver (eg, by just
'entering' the servers with chroot and starting the one or two services
you are testing, suitably configured to bind as appropriate).
> Our production boxes (w/o vservers) are happily shaping packets at the
> rates we encounter so CPU power is not an issue either.
> One thing that bugs me is why do these packets even go through eth0
> queue? ip route show table local shows the local ip addresses as
> "local X.X.X.X dev eth0 proto kernel scope host src X.X.X.Y" (X.X.X.X
> being guest address and X.X.X.Y being host address)
> Have you got any ideas/suggestions? Is this a known issue?
> Best regards,
> Grzegorz Nosek
> Vserver mailing list
Vserver mailing list
Received on Mon Nov 7 20:05:00 2005