[Pdns-users] High latency on recursion without cache

[Ciro Iriarte]

2015-01-27 13:21 GMT-03:00 Ciro Iriarte <cyruspy at gmail.com>:

> 2015-01-27 6:59 GMT-03:00 bert hubert <bert.hubert at powerdns.com>:
>
> On Sun, Jan 25, 2015 at 10:33:34AM -0300, Ciro Iriarte wrote:
>> > Also, the test traffic was stopped, so the trace file should be complete
>> > and cleaner!.
>>
>> Ciro,
>>
>> I don't see anything that is wrong here. From a cold cache, it takes 11
>> queries to resolve 2.centos.pool.ntp.org.
>>
>> Your network used up the following amounts of time on those queries:
>>  in 266ms
>>  in 226ms
>>  in 184ms
>>  in 226ms
>>  in 233ms
>>  in 267ms
>>  in 223ms
>>  in 201ms
>>  in 224ms
>>  in 51ms
>>  in 199ms
>>
>> Which together is around 2 seconds.
>>
>> If there is a problem, the problem is that your network is pretty far away
>> from most servers it appears.
>>
>> If you redo your query with the latest PowerDNS test version (3.7.0-RC1)
>> you'll get slightly better timing output with --trace, which perhaps could
>> tell you a little more.
>>
>> On a high-latency network (and your fastest response to anything in this
>> trace was 51ms, even if I look at the other queries too), having a warm
>> cache is super important.
>>
>> Good luck!
>>
>>         Bert
>>
>
> Hi Bert, thanks for the analysis!. I double checked all the configuration
> (routing/pdns/linux) and everything seems to be OK. It's obviously not a
> PDNS thing as a trace using dig still gives pretty bad times
>
> ; <<>> DiG 9.8.2rc1-RedHat-9.8.2-0.30.rc1.el6 <<>> +trace
> 2.centos.pool.ntp.org
> ;; Received 241 bytes from 186.16.16.16#53(186.16.16.16) in 1317 ms <---
> this goes to localhost for recursion
> ;; Received 441 bytes from 199.7.83.42#53(199.7.83.42) in 5628 ms
> ;; Received 153 bytes from 199.19.57.1#53(199.19.57.1) in 2881 ms
> ;; Received 189 bytes from 128.175.13.17#53(128.175.13.17) in 8346 ms
> ;; Received 187 bytes from 94.242.223.210#53(94.242.223.210) in 644 ms
>
> Complete trace:
>    http://pastebin.com/tvKqhq2e
>
> What I find weird is that a query to the server 199.7.83.42 takes more
> than 5 seconds, but a plain ping request gives a RTT of about 177ms. Maybe
> the servers are overloaded and I'm over-reacting :P
>
> --- 199.7.83.42 ping statistics ---
> 5 packets transmitted, 5 received, 0% packet loss, time 4187ms
> *rtt min/avg/max/mdev = 177.556/178.459/181.044/1.334 ms*
>
>
> The same goes for the other destinations:
>
> --- 199.19.57.1 ping statistics ---
> 5 packets transmitted, 5 received, 0% packet loss, time 4183ms
> *rtt min/avg/max/mdev = 177.125/177.302/177.644/0.565 ms*
>
> --- 128.175.13.17 ping statistics ---
> 5 packets transmitted, 5 received, 0% packet loss, time 4194ms
> *rtt min/avg/max/mdev = 187.530/188.211/189.673/0.931 ms*
>
>
> --- 94.242.223.210 ping statistics ---
> 5 packets transmitted, 5 received, 0% packet loss, time 4276ms
> *rtt min/avg/max/mdev = 265.467/269.252/272.820/2.637 ms*
>
> The only other thing I could think of is some kind of QoS issue and to
> blame the carrier. It's time to poke the networking guys...
>
> Regards,
>
> --
> Ciro Iriarte
> http://iriarte.it
> --
>

For the record, a second query for the same fqdn responds a lot faster
which implies there's some cache involved now.

; <<>> DiG 9.8.2rc1-RedHat-9.8.2-0.30.rc1.el6 <<>> +trace
2.centos.pool.ntp.org
;; Received 241 bytes from 186.16.16.16#53(186.16.16.16) in 7 ms
;; Received 441 bytes from 192.33.4.12#53(192.33.4.12) in 204 ms
;; Received 153 bytes from 199.19.53.1#53(199.19.53.1) in 183 ms
;; Received 189 bytes from 128.175.13.17#53(128.175.13.17) in 187 ms
;; Received 187 bytes from 46.234.32.107#53(46.234.32.107) in 478 ms


Regards,

-- 
Ciro Iriarte
http://iriarte.it
--
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.powerdns.com/pipermail/pdns-users/attachments/20150127/62daf605/attachment-0001.html>