[bind10-dev] an interesting take on how to do stats
Jerry Scharf
scharf at isc.org
Fri Dec 17 16:36:40 UTC 2010
Hi,
Paul Ebersman tossed an idea to me on the side and after thinking about
it I wanted to send it out to the list. This is me paraphrasing and
expanding his idea.
His basic suggestion was aimed at getting around the challenge of adding
more stats and not slowing down the server path. He asked what if the
stats engine got all its data from a pcap tap of the raw traffic and the
query processing didn't have to do anything related to stats.
At first glance, this seemed like quite a bit of duplicated processing
(packet unmarshalling, context switching, etc.) and would slow things
down even more. But on a multicore system, this could run almost
completely independent of the BIND 10 server, using it's own memory
space and not impacting the query processing "fast path" at all.
The upsides I see:
takes code out of the "fast path"
concentrates all logging work in a separate area, better modularity
ideal for high end systems with many cores/processors
The downsides I see:
doesn't work on systems that don't have pcap
could make life worse on busy servers with 1-2 cores
requires root to set up pcap tap
possibly duplicates code
what do people think?
jerry
More information about the bind10-dev
mailing list