BIND 9.x caching Performance under heavy loads

[Steinar Haug]

[Srini Avirneni]

|   > Please be more specific. You seem to be saying that "bad things happen
|   > when your name servers are under heavy load". This shouldn't exactly
|   > be a surprise. What's the actual problem are you experiencing and/or
|   > trying to solve?
|   
|   Yes, it is a suprise. Its a surprise to see Bind as a process, peak
|   at 25% CPU day 1. Then, Day 2, peak climbs to over 50%. Load
|   has not changed. 
|   
|   By day 3, CPU will exceed 75% and Bind will no longer respond in
|   a meaning full way (< 500 queries/sec). This is on very fast
|   hardware.

We saw a somewhat similar behavior earlier this year, running various
9.2.x versions on uniprocessor FreeBSD boxes. We saw cache stabilizing
at considerably lower values (500 MB or so), and after a few days, CPU
usage started climbing until the boxes were basically unusable. Note
that this rising CPU usage seemed to happen *after* the cache had
stabilized.

We are currently running BIND 9.3.1rc1 as a caching name server on the
following FreeBSD 4.11 uniprocessor boxes:

1. 2.67 GHz P4, 1 GB memory, query load typically 600 - 1500 q/s
2. 800 MHz P4, 512 MB memory, query load typically 400 - 700 q/s
2. 800 MHz P4, 512 MB memory, query load typically 300 - 600 q/s

In other words our peak query rate is on the order of 1500 q/s for one
box, 2800 q/s distributed on all of these boxes.

In all cases cache stabilized at well under the max-cache-size that we
had set (max-cache-size 400 MB for the boxes with 512 MB memory, 800 MB
for the box with 1 GB memory).

Our boxes now seem to be running just fine, and we haven't seen the
same issues with CPU usage climbing that we saw earlier. One important
difference is that we turned down or off basically all logging from the
servers. Other differences that may or may not be relevant:

- Running 9.3.1rc1
- Configured with --disable-ipv6 and ISC_MEM_USE_INTERNAL_MALLOC

Steinar Haug, Nethelp consulting, sthaug at nethelp.no