mlockall vs vm.swappiness

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

mlockall vs vm.swappiness

ppearcy
We recently figured out that mlockall is not working for us. Unfortunately, even with setting common.jna logging level to debug, no message was getting printed out indicating this (it really would be nice to have a definitive yes/no mlockall worked in the logs). 

For the moment, instead of locking the elasticsearch JVM heap in memory, we've set the vm.swappiness level to 0. Setting the swappiness level to 0 does effect all processes, not just elasticsearch, but these machines are single purpose search boxes, so I doubt that will be a problem. 

Anybody know if there any tradeoffs between the mlockall and vm.swappiness for dedicated search boxes? 

Thanks!
Paul

--
 
 
Reply | Threaded
Open this post in threaded view
|

Re: mlockall vs vm.swappiness

Otis Gospodnetic
Others can address the mlockall part, but I'll say that I set swappiness to 0 on servers as soon as I get my hands on them and have never regretted it, but seen only positives - no swapping.

Otis
--
Solr & ElasticSearch Support
http://sematext.com/


On Monday, January 14, 2013 2:19:59 PM UTC-5, ppearcy wrote:
We recently figured out that mlockall is not working for us. Unfortunately, even with setting common.jna logging level to debug, no message was getting printed out indicating this (it really would be nice to have a definitive yes/no mlockall worked in the logs). 

For the moment, instead of locking the elasticsearch JVM heap in memory, we've set the vm.swappiness level to 0. Setting the swappiness level to 0 does effect all processes, not just elasticsearch, but these machines are single purpose search boxes, so I doubt that will be a problem. 

Anybody know if there any tradeoffs between the mlockall and vm.swappiness for dedicated search boxes? 

Thanks!
Paul

--
 
 
Reply | Threaded
Open this post in threaded view
|

Re: mlockall vs vm.swappiness

ppearcy
Thanks Otis!

On Monday, January 14, 2013 9:21:13 PM UTC-7, Otis Gospodnetic wrote:
Others can address the mlockall part, but I'll say that I set swappiness to 0 on servers as soon as I get my hands on them and have never regretted it, but seen only positives - no swapping.

Otis
--
Solr & ElasticSearch Support
http://sematext.com/


On Monday, January 14, 2013 2:19:59 PM UTC-5, ppearcy wrote:
We recently figured out that mlockall is not working for us. Unfortunately, even with setting common.jna logging level to debug, no message was getting printed out indicating this (it really would be nice to have a definitive yes/no mlockall worked in the logs). 

For the moment, instead of locking the elasticsearch JVM heap in memory, we've set the vm.swappiness level to 0. Setting the swappiness level to 0 does effect all processes, not just elasticsearch, but these machines are single purpose search boxes, so I doubt that will be a problem. 

Anybody know if there any tradeoffs between the mlockall and vm.swappiness for dedicated search boxes? 

Thanks!
Paul

--