Bind performance as caching nameserver

[Ulrich David]

You are right, I would have put recursive clients to a very large  
value in order to see what is the real max performance.

But in the real case, I never have more than 1500 recursive clients.

David

Le 28 août 08 à 23:37, Kevin Darcy a écrit :

> Why do you have a recursive-clients limit of 1500 if the point is to
> push your box to the breaking point?
>
> - Kevin
>
> Ulrich David wrote:
>> Hi,
>>
>> I'm testing some configuration for my caching name server with Bind
>> 9.4.2-P2 using resperf. The server is a dual Core Xeon at 2,3GHz 2GB
>> Ram runing gentoo linux. The goal is to have maximum performance from
>> this box.
>>
>> The problem is I don't know which performance I can expect of, in  
>> term
>> of latency and in term of response per second (without or with loss).
>>
>> Here is my results :
>> Statistics:
>>
>>   Queries sent:         187367
>>   Queries completed:    136574
>>   Queries lost:         50793
>>   Ran for:              54.994721 seconds
>>   Maximum throughput:   15492.000000 qps
>>   Lost at that point:   9.31%
>>
>> # time target_qps actual_qps responses_per_sec failures_per_sec
>> avg_latency
>>   0.250   416.67   416.00   416.00     2.00 0.007470
>>   0.750  1250.00  1250.00  1250.00    18.00 0.010431
>>   1.250  2083.33  2082.00  2082.00    40.00 0.004323
>>   1.750  2916.67  2918.00  2918.00   290.00 0.186844
>>   2.250  3750.00  3750.00  3750.00  2186.00 2.234745
>>   2.750  4583.33  4582.00  4582.00  1668.00 2.879924
>>   3.250  5416.67  5418.00  5418.00  1260.00 1.817618
>>   3.750  6250.00  6250.00  6250.00  2864.00 4.013023
>>   4.250  7083.33  7082.00  7082.00  1760.00 1.853761
>>   4.750  7916.67  7918.00  7918.00  2450.00 2.763024
>>   5.250  8750.00  8750.00  8750.00  4146.00 3.793698
>>   5.750  9583.33  9582.00  9582.00  3522.00 2.775299
>>   6.250 10416.67 10418.00 10418.00  3796.00 2.814351
>>   6.750 11250.00 11250.00 11250.00  4872.00 3.234263
>>   7.250 12083.33 12082.00 12082.00  4454.00 2.876617
>>   7.750 12916.67 12918.00 12918.00  6312.00 3.520631
>>   8.250 13750.00 13750.00 13750.00  6200.00 3.318447
>>   8.750 14583.33 14582.00 14582.00  4192.00 2.017755
>>   9.250 15416.67 15418.00 15418.00  4278.00 1.928234
>>   9.750 16250.00 16250.00 13854.00  4998.00 2.440862
>>  10.250 17083.33 17082.00 15492.00  6544.00 2.837576
>>  10.750 17916.67 17918.00 12032.00  4430.00 2.467845
>>  11.250 18750.00 18750.00 13120.00  4354.00 2.217802
>>  11.750 19583.33 19582.00 15038.00  4810.00 2.081687
>>  12.250 20416.67 20418.00 10182.00  3070.00 1.983601
>>  12.750 21250.00 21250.00  8074.00  2832.00 2.224380
>>  13.250 22083.33 22082.00  9254.00  2834.00 1.906686
>>  13.750 22916.67 22918.00  8896.00  2326.00 1.752326
>>  14.250 23750.00 23750.00  8964.00  2992.00 2.003389
>>
>> I'm quite surprised by the failures and by the average latency for
>> more than 1500 qps...
>>
>> Are these results realistic for this hardware configuration or have I
>> miss something in my config file?
>>
>> query logging is OFF and my options are :
>> options {
>> 	directory "/var/bind";
>> 	pid-file "/var/run/named/named.pid";
>> 	statistics-file "/var/bind/stats/named.stats";
>>
>> 	listen-on-v6 { none; };
>>         listen-on { myself; };
>>
>> 	version	"None of your business";
>> 	allow-query { trusted_network; };
>> 	blackhole { blacklist; };
>> 	max-cache-size 0;
>> 	recursive-clients 1500;
>> };
>>
>>
>>
>>
>>
>
>