Question when testing Caching Server with resperf
Hi, My bind version is bind-9.7.2-P3, resperf is dnsperf-1.0.1.0-1-solaris-10-sparc. On sparc 10u8 Solaris Normally, resperf tool measures maximum throughput of caching server. So i did a test follow : run with query 100-thousand -> maximum throughput ~ 9000 -> named process ~ 450 MB run with query 100-thousand -> "ran out query data" errors run with query 3-millions -> maximum throughput ~ 9000 -> named process ~ 400 MB run with query 3-millions -> maximum throughput ~ 16000 -> named process ~ 500 MB . . . run with query 3-millions -> maximum throughput ~ 16000 -> named process ~ 600 MB run with query 3-millions -> maximum throughput ~ 16000 -> named process ~ 600 MB So named cache ramains ~ 600 Mb. although many times i run the testing command. *** Now, what i want to know is, does resperf tool use the query which was used in previous command. For Example: If query 1 run 65000 first line, does the query 2 run the same query or it will run next 65000 line below. Or it ran query in file randomly. If it does not ran the same, why the cache remains 600M. But i think it have to cache, because the maximum throughput is increase from 9000 to 16000. If u have my answer please note me where it is in the document from Nominum. Thank men. Tien86. ___ bind-users mailing list bind-users@lists.isc.org https://lists.isc.org/mailman/listinfo/bind-users
Re: Question when testing Caching Server with resperf
Hi Eivind Olsen, The command to did test is /usr/local/nom/bin/resperf -d /usr/local/nom/examples/dnsperf/queryfile-example-100thousand -s localhost -P result.txt queryfile-example-100thousand or queryfile-example-3million are provided by Nominum. Each line on file is a input to query so each file have 100 000 and 3 000 000 lines( queries). So data is external data and real data. First test: When i use queryfile-example-100thousand, i did two command. Its ok with first command with result 9000 throughput. The second command is "ran out of query data". So it mean resperf use all the lines ( queries ) of file. So resperf doesn't re-use the same 65000 line that was run in the first command. May be it used 65000 line randomly, i don't know. I can' find how it works with query in Nominum document. So i need help at this. Second test: Next, i flush cache. I run command with 3million query file. And run continuesly the command and check named process by top. The process just reach to 600Mb and not increase anymore. So it mean the cache is not increase. But if the command use different query each time (as first test) why it doesnt increase? As i think, the cache will save new queries and it will help faster search. Tien 86. ___ bind-users mailing list bind-users@lists.isc.org https://lists.isc.org/mailman/listinfo/bind-users
Bind hang out when named reach to 5-600 Mb
Hi,
I install bind as a cache server on Solaris 10, Sun Sparc T5140. It has
problem,
bind always hang out when named reach to 5-600 Mb ('prstat' check). I have
several servers and all have this problem even when i install bind in zone or
try with a 64bit version. T5140's a powerful server but bind can't make use of
its power. I'm newb with bind an so i have just try some other way but useless.
What should i do to track this problem ?
Thank in advance,
Duc Tien.
___
bind-users mailing list
bind-users@lists.isc.org
https://lists.isc.org/mailman/listinfo/bind-users
Re: Bind hang out when named reach to 5-600 Mb
Can u tell me how to do this ?
If you enable query log, you might be able to see if it's actually
serving queries at that time.
I mean "it hang" is named process is still running but it cannot respond any
lookup query. CPU is almost at low rate 5-6 % or lower. It's both hang with
single thread or multithread.
datasize is m
max-cache-size is 2000m;
These are versions of bind i tried :
bash-3.00# file /usr/sbin/named
/usr/sbin/named:ELF 32-bit MSB executable SPARC Version 1, dynamically
linked, stripped
bash-3.00# which named
/usr/sbin/named
bash-3.00# file /opt/SUNWbind9.6/sbin/named
/opt/SUNWbind9.6/sbin/named:ELF 64-bit MSB executable SPARCV9 Version 1,
dynamically linked, not stripped
bash-3.00# file /usr/sbin/bind
bind-9.7.1b1/ bind9.7.0P1/
bash-3.00# file /usr/sbin/bind-9.7.1b1/named
/usr/sbin/bind-9.7.1b1/named: ELF 32-bit MSB executable SPARC Version 1,
dynamically linked, not stripped
In my circumstance, i can try installing anything as long as it can fix this
bug. I saw on sunsolve there 's a patch for bind 119783-15. Don't know is it
useful to fix this bug?
Mb i will check with other version of bind. Hope this can help.
____
From: Fajar A. Nugraha
To: khanh rua
Cc: bind-users@lists.isc.org
Sent: Thu, July 8, 2010 5:10:02 PM
Subject: Re: Bind hang out when named reach to 5-600 Mb
On Thu, Jul 8, 2010 at 4:30 PM, khanh rua wrote:
> Hi,
>
> I install bind as a cache server on Solaris 10, Sun Sparc T5140. It has
> problem, bind always hang out when named reach to 5-600 Mb ('prstat' check).
How did you determine that it "hang"?
If you enable query log, you might be able to see if it's actually
serving queries at that time.
Also, how is the cpu and disk usage at that time (I'm trying to see if
you have cpu or disk as bottleneck?
My guess is that:
- one of your CPU thread is at 100% usage
- named is busy serving queries alreadt, so that the new query you
issued does not get processed in a timely manner.
> I have several servers and all have this problem even when i install bind in
> zone or try with a 64bit version. T5140's a powerful server but bind can't
> make use of its power.
IMHO, it's not really named-specific problem. The thing is Sun/Oracle
T-series processors doesn't perform well with single thread loads. And
(last time I check anyway) even though bind can make use of multi
processor/threads, single thread performance still matters a lot.
So you'd probably get much better performance when running named with
say ... a generic x86 server/PC with Intel Xeon/Core i-series.
--
Fajar
___
bind-users mailing list
bind-users@lists.isc.org
https://lists.isc.org/mailman/listinfo/bind-users
Re: Bind hang out when named reach to 5-600 Mb
Hi
I attached the ouput from your command in 10s.
Last time i tested, bind did reach to 1Gb but it still hang ( don't respond any
query ). Could you help me take a look at this log?
Duc Tien.
From: Stacey Jonathan Marshall - Solaris Software
To: khanh rua ; bind-users@lists.isc.org
Sent: Fri, July 9, 2010 5:27:15 PM
Subject: Re: Bind hang out when named reach to 5-600 Mb
On 07/09/10 07:36, khanh rua wrote:
Can u tell me how to do this ?
>
>If you enable query log, you might be able to see if it's actually serving
>queries at that time.
>
I mean "it hang" is named process is still running but it cannot respond any
lookup query. CPU is almost at low rate 5-6 % or lower. It's both hang with
single thread or multithread.
>datasize is m
> max-cache-size is 2000m;
>These are versions of bind i tried :
>bash-3.00# file /usr/sbin/named
>/usr/sbin/named:ELF 32-bit MSB executable SPARC Version 1, dynamically
>linked, stripped
>bash-3.00# which named
>/usr/sbin/named
>bash-3.00# file /opt/SUNWbind9.6/sbin/named
>/opt/SUNWbind9.6/sbin/named:ELF 64-bit MSB executable SPARCV9 Version 1,
>dynamically linked, not stripped
>bash-3.00# file /usr/sbin/bind
>bind-9.7.1b1/ bind9.7.0P1/
>bash-3.00# file /usr/sbin/bind-9.7.1b1/named
>/usr/sbin/bind-9.7.1b1/named: ELF 32-bit MSB executable SPARC Version 1,
>dynamically linked, not stripped
>
running 'named -V' is the best may to show version and configure options.
>In my circumstance, i can try installing anything as long as it can fix this
>bug. I saw on sunsolve there 's a patch for bind 119783-15. Don't know is it
>useful to fix this bug?
>
>
No, that patch provides BIND 9.6.1-P3 with no specific changes.
Could you collect some lock statistics using plockstat(1M)? i.e.:
# plockstat -e 300 -p `pgrep ^named`
The -e 300 causes it to collect statistics for 5 minutes - experiment a little
and feed back what is reported.
>Mb i will check with other version of bind. Hope this can help.
>
>From: Fajar A. Nugraha
>To: khanh rua
>Cc: bind-users@lists.isc.org
>Sent: Thu, July 8, 2010 5:10:02 PM
>Subject: Re: Bind hang out when named reach to 5-600 Mb
>
>On Thu, Jul 8, 2010 at 4:30 PM, khanh rua wrote:
>> Hi,
>>
>> I install bind as a cache server on Solaris 10, Sun Sparc T5140. It has
>> problem, bind always hang out when named reach to 5-600 Mb ('prstat' check).
>
>How did you determine that it "hang"?
>If you enable query log, you might be able to see if it's actually
>serving queries at that time.
>
>Also, how is the cpu and disk usage at that time (I'm trying to see if
>you have cpu or disk as bottleneck?
>
>My guess is that:
>- one of your CPU thread is at 100% usage
>- named is busy serving queries alreadt, so that the new query you
>issued does not get processed in a timely manner.
>
>> I have several servers and all have this problem even when i install bind in
>> zone or try with a 64bit version. T5140's a powerful server but bind can't
>> make use of its power.
>
>IMHO, it's not really named-specific problem. The thing is Sun/Oracle
>T-series processors doesn't perform well with single thread loads. And
>(last time I check anyway) even though bind can make use of multi
>processor/threads, single thread performance still matters a lot.
>
>So you'd probably get much better performance when running named with
>say ... a generic x86 server/PC with Intel Xeon/Core i-series.
>
>--
>Fajar
>
>
> ___ bind-users mailing list
>bind-users@lists.isc.org https://lists.isc.org/mailman/listinfo/bind-users
--
--Stacey
plockstat.rar
Description: Binary data
___
bind-users mailing list
bind-users@lists.isc.org
https://lists.isc.org/mailman/listinfo/bind-users
