[MarkLogic Dev General] Option to not hit cache

Michael Blakeley mike at blakeley.com
Wed Aug 22 07:50:46 PDT 2012


List cache and expanded-tree cache have to be large enough for the queries you are running or you'll just get CACHEFULL errors. You can set compressed-tree to 1, but the OS will just cache more of that data for you.

Clearing the OS cache won't clear disk-level and controller-level caches. In my experience those will make a measurable difference.

The best solution is to design the workload so that it represent the real application workload. A representative workload will use the various hardware and software caches to the same extent that the real application does, so no tricks will be needed. This is why database workloads are typically large: 3x memory is a good rule of thumb. At that point the access pattern and concurrency level will dictate cache utilization.

Of course designing a good workload is hard, so folks look for shortcuts. But this shortcut ends in a ditch.

-- Mike

On 22 Aug 2012, at 07:21 , semerau at hotmail.com wrote:

> What about setting the amounts for expanded and compressed tree caches to be really small? Would that work?
> 
> Sent from my iPhone
> 
> On Aug 22, 2012, at 7:01 AM, "Wayne Feick" <Wayne.Feick at marklogic.com> wrote:
> 
>> No, there isn't. Your best bet would be to stop the server, clear OS caching from memory (e.g. unmount and remount the filesystem), and then restart the server again.
>> 
>> Wayne
>> 
>> Danny Sinang <
>> d.sinang at gmail.com
>> > wrote:
>> 
>> 
>> Hi,
>> 
>> Is there an option in xquery, cts:search, or search:search telling ML not to fetch results from the cache ?
>> 
>> We need to test the performance of some queries and we want to see its actual speed not coming from the cache.
>> 
>> Regards,
>> Danny
>> 
>> _______________________________________________
>> General mailing list
>> General at developer.marklogic.com
>> http://developer.marklogic.com/mailman/listinfo/general
> _______________________________________________
> General mailing list
> General at developer.marklogic.com
> http://developer.marklogic.com/mailman/listinfo/general



More information about the General mailing list