Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 30 Jul 1999 11:51:15 -0700
From:      "David Schwartz" <davids@webmaster.com>
To:        "Joe Gleason" <freebsd.list@bug.tasam.com>, "Tom" <tom@uniserve.com>
Cc:        "Jeffrey J. Libman" <jeffrl@wantabe.com>, <freebsd-stable@FreeBSD.ORG>
Subject:   RE: routing over a dual t1 connection (fwd)
Message-ID:  <001601bedabc$805f0410$021d85d1@youwant.to>
In-Reply-To: <006801bedab6$23b107a0$0286860a@tasam.com>

next in thread | previous in thread | raw e-mail | index | archive | help
> > > > 1) It can be very easy to get working.
> > >
> > >   Funny, you just said it was tricky to get working :)
> >
> > Tricky to get working _well_. Easy to get working. :)
> >
> > Adding an extra 'ip route' command takes 5 seconds. But then
> watching your
> > traffic balance terribly and performance go down is the hitch.
>
> I have done this is a production environment, the traffic balanced
> perfectly.

	Then perhaps you'd care to share with us how you told the router whether it
should route outbound traffic to avoid inbound traffic or not do so.
Universally, the inability to control this is the largest problem that I
have. Is there some 'outbound traffic avoids inbound traffic' mode that you
can turn on or off?

	The router seems to make no distinction between shared and unshared media
(bandwidth shared between inbound and outbound) and, as far as I can tell,
uses an algorithm that is optimized for neither case.

	In addition, no single TCP connection seems to be able to use the combined
bandwidth. They appear to be assigned to a single link and pretty much stay
that way. This could be considered to be a feature, I suppose it depends
upon your application.

	And if it did use both links for a single connection, large numbers of
packets would be received out of order. This is because each link is
considered a separate IP conduit. Techniques that bond the two links into a
signle conduit do not have this problem.

	The decision of which link to assign a connection to seems made based upon
long-term load calculations (30 _seconds_ or more), so traffic patterns that
vary widely over short periods of time cannot easily be accounted for.

	I'd love to learn the tricks to fix this, because I'd prefer to use this
solution. But I've had mediocre results every time I've tried it.

	DS



To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-stable" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?001601bedabc$805f0410$021d85d1>