Operation cancelled Error

[Ben]

Hi Jeremy,

Thanks for your kind response.
> On Thu, 24 May 2012, Ben wrote:
>
>>> version: 9.7.3-P3-RedHat-9.7.3-8.P3.el6_2.2
>>> CPUs found: 8
>>> worker threads: 8
>>> number of zones: 19
>>> debug level: 0
>>> xfers running: 0
>>> xfers deferred: 0
>>> soa queries in progress: 0
>>> query logging is ON
>>> recursive clients: 6400/29900/30000
>>> tcp clients: 0/100
>>> server is up and running
>>>
>>>
>>> i constanly watch rndc status command , and at recuresive-clients tab ,
>>> first values increases maximum up to 6000-6500, why it is not going to
>>> maximum which i define 30000..?
> I don't know why it never reached the maximum. resperf should try to
> scale up to attempting 100,000 questions in its last second. (At 60th
> second I think; the final 40 seconds is waiting for responses.) It only
> tries 74038 during its total time, but I am not sure what is limiting
> it.
>
> Maybe your datafile is not unique enough? Maybe your source port range
> is not large enough? So then BIND 9 is matching existing requests and
> dropping.
My source port range is
cat /proc/sys/net/ipv4/ip_local_port_range
1024    65535

I downloaded data file from resperf provider site.
> It depends a lot on the dataset. (I think I have seen around 17,000
> queries with resperf and as low as 236 qps -- in this case it was
> depending on number of ACLs.)
I do not using more acl for testing purpose.
> I don't know why you have the burst of "operation canceled". (The
> ISC_R_CANCELED can happen from different problems.)
Please suggest us that what are reasons generate "operation canceled" 
error comes in named.run log file
>>> rndc status shows 8 worker process, when i checked  by pgrep named , it
>>> shows only single instance.so does it need to show 8 instance or ?
> 8 worker threads is different than 8 processes.
>
>>> Currently we use bind as caching name server , so why rndc status shows
>>> number of zones 19..?
> The 19 zones are built-in zones. (See the ARM for the list.)
>
> By the way, to set some comparison maximum baseline you can try having
> resperf query the built-in zones. (It won't be real recursive work, but
> should show you some potential maximum qps.)
>
Is there anything which we need to mind on OS kernel tuning parameters 
or from bind configuration side to achieve more QPS?

By the way, what is highest benchmark for bind with no. of QPS in 
production servers?

I would request you , if someone has getting high QPS with bind in 
production servers, kindly suggest your inputs.


>    Jeremy C. Reed
>    ISC
Regards,
Ben