On 2008-09-03, Toni Mueller <[EMAIL PROTECTED]> wrote: > On Wed, 05.09.2007 at 00:01:09 -0600, Anthony Roberts <[EMAIL PROTECTED]> > wrote:
[EMAIL PROTECTED] :-) >> I've been tuning some networks for VoIP recently, and to get >> really good results I've found it's been necessary to do altq >> in both directions. > > This should imho be possible to look at what kind of traffic goes out > of one interface, then write appropriate altq rules. For a router, > which seems to be what you're talking about, (almost) all traffic that > enters the router on one side, leaves it on some other side. That way, > each packet needs to traverse one interface in the outgoing direction. for simple cases yes, but you missed quoting this bit: "For example, if there is more than one internal network, one can't create a single altq instance that covers them all. You can divide bandwidth between them, but you can't borrow between the different queues in this case." >> -Hosts cannot be prevented from sending me packets, so the >> potential exists for inbound bandwidth to be exausted no matter >> what I do. > > Right, but for TCP at least, you could, in theory, employ window > scaling, delaying ACKs, and ECN to make the other side send their > packets at a slower rate. This should work unless the other side is > broken, or simply a rogue site. I don't know how much overhead such a > mechanism will introduce, though. Queuing on outbound means the destination sees the packet later, so ACKs _are_ delayed, which is the reason this does actually slow down the sending rate (for TCP, anyway).
