[dns-operations] DNS perf benchmarking tools

[Christian Petrasch]

Hi Daniel,

you're right, we also have only 7% tcp queries at peak times.. a very 
little part of all queries.
But we analysed tcp query behaviour because of a evalution of new hardware 
in combination of virtualizing nameserver with different hypervisors.
So the result our analysis of the new tcp performance was relevant. 


>Further: I personally believe in benchmarking to loads that are designed 
after actual or expected loads. This is how we benchmarked the initial 
versions of nsd against the "incumbent". We replayed actual >loads in a 
test lab against the servers, captured the response streams and analysed 
them. 

Yes, i agree, that would be the best way to test. At the time of our tests 
in the past udp we tested with a query capture and dnsperf but for tcp we 
wouldn't build a big testing environment, so we decided
perftcpdns is a good alternative to had a look onto the tcp stack.

kind regards
 
-- 
Christian Petrasch 



Von:    Daniel Karrenberg <daniel.karrenberg at ripe.net>
An:     "dns-operations at dns-oarc.net" <dns-operations at dns-oarc.net>, 
Datum:  07.04.2014 10:15
Betreff:        Re: [dns-operations] DNS perf benchmarking tools
Gesendet von:   dns-operations-bounces at lists.dns-oarc.net




On 05.04.2014, at 15:34 , Francis Dupont <Francis.Dupont at fdupont.fr> 
wrote:

> - you have to very carefully define what you mean by TCP DNS performance
>  or you can finish with interesting numbers about the NUT kernel
>  (vs. name server) behavior...


Excellent point.

Let me expand: If you are interested in TCP performance, the kernel and 
particularly the networking stack has much much more influence than with 
UDP. To amplify: the best server will give poor results when running on 
the wrong system. So overall performance experienced by the querier is 
much more a total system issue.

Further: I personally believe in benchmarking to loads that are designed 
after actual or expected loads. This is how we benchmarked the initial 
versions of nsd against the "incumbent". We replayed actual loads in a 
test lab against the servers, captured the response streams and analysed 
them. 

To amplify: it makes no sense to optimise for TCP performance when this 
http://k.root-servers.org/statistics/GLOBAL/ip_protocols.html
is the current load and there is no trend away from it:

No matter what FUD and marketing hype say. Is anybody really seeing 
significant TCP load in the wild at this time?

Daniel
[attachment "signature.asc" deleted by Christian Petrasch/Denic] 
_______________________________________________
dns-operations mailing list
dns-operations at lists.dns-oarc.net
https://lists.dns-oarc.net/mailman/listinfo/dns-operations
dns-jobs mailing list
https://lists.dns-oarc.net/mailman/listinfo/dns-jobs

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/x-pkcs7-signature
Size: 8597 bytes
Desc: S/MIME Cryptographic Signature
URL: <https://lists.dns-oarc.net/pipermail/dns-operations/attachments/20140407/0131ce0f/attachment.bin>