[bind10-dev] auth server benchmarks and resource usage

[Jeremy C. Reed]

The types of data or DNS hosting scenarios I am (or will be) testing 
are:

1) single small zone with only three records

2) hosting company model with 10,000 zones with 15 records each

3) larger hosting company with 50,000 zones with 15 records each

4) huge hosting company with 100,000 zones with 15 records each

5) TLD with one zone with 5 million records

Then duplicates of all of these as DNSSEC signed. And then later these 
with NSEC3 too.

And also:

6) built-in data (version.bind) (I used custom queryperf for this. This 
reminds me: I need to submit a patch for queryperf to allow choosing the 
class.)

For each, do these types of queryperf tests:

a) queries resulting in success

b) queries with DNSSEC OK bit

c) queries resulting in NXDOMAIN

d) queries with DNSSEC OK bit resulting in NXDOMAIN

e) queries resulting in SERVFAIL or FORMERR (may need to force loading 
of corrupted data in data source)

(Note for DNSSEC, I do no validation as I am not testing that.)

Then do a series of different amounts of these, starting around 50,000 
queries and scaling up to long running tests (few minutes). This was 
useful to see that existing caching cleanup every 30 seconds had 
significant slow down.

And do each of these tests with caching on or off (-n).

Later will extend to do tests per different datasource backends;
currently only options are builtin, sqlite3, and sqlite3+caching.

I am also recording vmstat and per-process memory and cpu stats data 
(before and after).

As applicable I am able to run some of these tests against non-BIND10 
servers also.

>From this data, I plan to generate XML and/or HTML reports.

Any performance tests or resource usage data collection you'd like to me 
to consider? Any suggestions or advice for above?

(By the way, I have simple script that you give it a branch name and 
revision and it checks out, builds, installs, and does performance and 
resource testing.)