Propose FreeBSD 3.3 for dummynet packet loss and delays

From: Duane Wessels (wessels@ircache.net)
Date: Mon Dec 06 1999 - 17:17:52 MST


 folks,

We now feel that Dummynet on FreeBSD-3.3 is reliable enough to
use for the bake-off tests. Thus, we propose to change to make this
change to 3.3 (instead of 3.1) in order to keep the packet loss
and delays in the bake-off workload.

The dummynet parameters that we will use are:

        Polygraph Servers: 55 msec delay, 0.05% loss
        Polygraph Clients: 75 msec delay, 0.1% loss

We've found that dummynet works best when you have one "pipe" for
incoming packets, and another for outgoing packets. It can be
configured using these commands:

        ipfw -f flush
        ipfw pipe 1 config delay 75ms plr 0.001
        ipfw pipe 2 config delay 75ms plr 0.001
        ipfw add pipe 1 ip from any to 10.8.0.0/16 in
        ipfw add pipe 2 ip from 10.8.0.0/16 to any out

(you will need to change "10.8.0.0" to match your own configuration).

Since we add the low-level packet delays, we propose to decrease
the Polygraph server-side "think-time." Currently it is norm(3sec,1.5sec).
We'd like to use norm(2.5sec,1.0sec) instead.

For your convenience, we have set up a FreeBSD-3.3 kernel source
tree that has all of the changes needed for the bake-off environment.
This tree is available via rsync. You can get it with this
command:

        # rsync -a -v ftp.ircache.net::freebsd33-bakeoff-sys /usr/src/sys

Our source tree includes the following changes:

        * TCP MSL lowered to 3 seconds
        * Raised system-wide MAXFILES to 32K
        * Added "atomic = 1" patch so that writes slightly
          larger than one mbuf are not delayed.
        * Alex's "addr_hash" patch that implements interface address
          lookups with a hash table instead of a linked list.

Duane W.



This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:10 MDT