You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@cassandra.apache.org by Milind Parikh <mi...@gmail.com> on 2011/07/22 19:34:39 UTC
Predictable low RW latency, SLABS and STW GC
In order to be predicable @ big data scale, the intensity and periodicity of
STW Garbage Collection has to be brought down. Assume that SLABS (Cass 2252)
will be available in the main line at some time and assume that this will
have the impact that other projects (hbase etc) are reporting. I womder
whether avoiding GC by restarting the servers before GC will be a feasible
approach (of course while knowing the workload)
Regards
Milind
Re: Predictable low RW latency, SLABS and STW GC
Posted by Peter Schuller <pe...@infidyne.com>.
> Restarting the service will drop all the memmapped caches, cassandra caches are saved / persistent and you can also use memcachd if you want.
Well, the OS won't evict everything from page cache just because the
last process to map them exits. That said, since restarts tend to have
secondary effects on caches like streaming through all the bf and
index files, restarts are certainly detrimental to the page cache.
Also you may still see some eviction (even if it doesn't *necessarily*
happen) depending (particularly if not running with numactl set to
interleave).
--
/ Peter Schuller (@scode on twitter)
Re: Predictable low RW latency, SLABS and STW GC
Posted by aaron morton <aa...@thelastpickle.com>.
Restarting the service will drop all the memmapped caches, cassandra caches are saved / persistent and you can also use memcachd if you want.
Are you experiencing stop the world pauses? There are some things that can be done to reduce the chance of them happening.
Cheers
-----------------
Aaron Morton
Freelance Cassandra Developer
@aaronmorton
http://www.thelastpickle.com
On 23 Jul 2011, at 05:34, Milind Parikh wrote:
> In order to be predicable @ big data scale, the intensity and periodicity of STW Garbage Collection has to be brought down. Assume that SLABS (Cass 2252) will be available in the main line at some time and assume that this will have the impact that other projects (hbase etc) are reporting. I womder whether avoiding GC by restarting the servers before GC will be a feasible approach (of course while knowing the workload)
>
> Regards
> Milind
>