Performance tuning tips required for bind 9.6.1-P3!!!

[Shiva Raman]

Dear All

 This is in reference to the performance tuning , i had already gone through the mailing list archives , but could not find answer to my
specific query mentioned here.

 I had installed  bind as a caching name  server for test purposes  and planning to test performance that could give me around 10000 qps.

The os running Centos 5.4 64 bit , with minimal packages installed. The server is dual quad core Intel Xeon 2.53 GHz ,16 GB RAM with 300 GB Hdd( Raid 1) .

Bind version installed is bind 9.6.1-P3 . Source extracted and compiled with the following options.

./configure --enable-epoll  --enable-atomic  --enable-ipv6 --enable-chroot --with-openssl --with-randomdev=/chroot/named --disable-openssl-version-check --with-libtool --enable-threads

Chrooted bind installation is done.

Only named, ssh and ntp services are running on the servers.

Right now i am using queryperf to test the performance with sample query file of thousand entries. Right now
i am getting only 2000 to 2300 qps . I am writing querylogs to a separate partition with noatime enabled for the
partition.

   OS hardening is done by  removing unwanted services, closing all unneccesary ports and  securing the running services.

   My system is now using only 3 GB of RAM of total 16 GB.

Following is the output of "uptime;free -m" during performance testing

[root@localhost ~]# uptime;free -m
 22:19:52 up 1 day,  6:06,  3 users,  load average: 2.03, 2.06, 1.34
             total       used       free     shared    buffers     cached
Mem:         16047       3183      12864          0        238       2037
-/+ buffers/cache:        907      15140
Swap:         8189          0       8189




Following is my named.conf

acl testsetup_net {
            10.201.31.0/26; };

acl blacklistnets {
      192.0.2.0/24; 224.0.0.0/3; 10.0.0.0/8; 192.168.0.0/16;
};

// Main options defined here
options {
      directory "/conf";
      dump-file "named_dump.db";
      statistics-file "named.stats";
      pid-file "/var/run/named.pid";
      allow-recursion { localhost; ; testsetup_net; };
      allow-query { localhost; testsetup_net; };
      allow-query-cache { localhost;  testsetup_net; };
      allow-transfer { none; };
      blackhole { blacklistnets; };
      recursive-clients 20000;
      version "Not old!";
      datasize default;
      notify yes;
};

// Logging options are defined here.
logging {         // logging option for named  process
        channel "default_debug" {
            file "/logs/named.log" versions 10 size 50m;
           print-time yes;
            print-category yes;
             severity dynamic;
       };

      channel "queries" {         // logging option for queries to named
            file "/logs/query.log" versions 20 size 100m;
                print-time yes;
                print-category yes;
                severity dynamic;
        };

      category default { "default_debug"; };
     category queries { "queries"; };    // uncomment this to log queries
      category config { "default_debug"; };
      category security { "default_debug"; };
      category network { "default_debug"; };
      category lame-servers { null; };
        category edns-disabled { null; };
};


zone "." in {
      type hint;
      file "db.rootcache";
};

zone "localhost" in {
      type master;
      file "db.local";
      notify no;
};

zone "0.0.127.in-addr.arpa" in {
      type master;
      file "db.127.0.0";
      notify no;
};


   Kindly guide me for improving the bind performance from 2000 qps to nearly 10000 qps. Which are the parameters i should change for improving
the performance? Any os level parameters to be changed for improving the performance?

thanks in advance

Regards

Shiva Raman

[Dave Sparro]

On 7/13/2010 1:11 PM, Shiva Raman wrote:
> Dear All
>
> This is in reference to the performance tuning , i had already gone
> through the mailing list archives , but could not find answer to my
> specific query mentioned here.
>
>

> Right now i am using queryperf to test the performance with sample query
> file of thousand entries. Right now
> i am getting only 2000 to 2300 qps .
>

> Kindly guide me for improving the bind performance from 2000 qps to
> nearly 10000 qps. Which are the parameters i should change for improving
> the performance? Any os level parameters to be changed for improving the
> performance?
>

What does your query file look like. On of the biggest things that
affects the numbers for a caching server is the response time of the
authoritative servers that answer the queries in your file. Network
bottlenecks can be a problem too. (I remember one time I experimented
with a caching server that had a stateful firewall between it and the
Internet; effectively killed connectivity for everybody in the building)

You may want to look at resperf:

http://www.nominum.com/services/measurement_tools.php

You may be able to get some more meaningful numbers from it.

--
Dave