On Tuesday, 15 November 2016 07:09:13 GMT Jerry Lundström wrote:
> On 11/15/16 06:18, Sara Dickinson wrote:
> > - It requires around an order of magnitude less CPU to compress the C-DNS
> > file.
> Can you clarify this, do you mean that it takes longer to compress PCAP
> then C-DNS? If so, do you have numbers? And how did you get them?
We looked at compressing raw PCAP versus the same data encoded as C-DNS as
part of the C-DNS design process. We ran
$ /usr/bin/time xz -9 <input>
and logged the reported user and system CPU times and maximum resident set
size. The idea was to get an idea of the resource consumption of the
compression, and look at the obvious question of whether it was better than
just compressing PCAP.
We used 'xz -9' to get some idea of the upper bound of achievable compression.
We looked at 5 minute capture samples for 3 different servers, with loadings I
might roughly describe as light, medium and heavy.
You'll need fixed font to comfortably view the following table of results:
File size CPU
Traffic Format Input Output VM space User System
Light PCAP 47.92M 4.14M 490.56M 8.27 7.29
Light C-DNS 5.12M 1.78M 113.44M 3.39 1.19
Medium PCAP 245.77M 17.96M 690.05M 102.53 15.91
Medium C-DNS 23.57M 6.88M 276.47M 9.38 0.11
Heavy PCAP 677.61M 45.25M 690.05M 180.15 6.61
Heavy C-DNS 62.85M 16.83M 620.37M 29.36 0.95
This is really what one would expect. C-DNS presents the compression with much
smaller input, so less to chew through, and groups data likely to be similar
together in its header tables. So the compressor is given an easier job, and
needs fewer resources.
--
Jim Hague - j...@sinodun.com Never trust a computer you can't lift.
_______________________________________________
DNSOP mailing list
DNSOP@ietf.org
https://www.ietf.org/mailman/listinfo/dnsop
