bitcask 42?

Alexander Sicular siculars at gmail.com
Mon Oct 4 09:17:27 EDT 2010


If you have a server that has raided disk and you don't have a battery
backed raid controller you fail at buying servers as it's virtually
the only piece of hardware that makes a server a server.

On 2010-10-04, Nico Meyer <nico.meyer at adition.com> wrote:
> One should also mention, that the overall performance of bitcask might
> presumably become worse with a larger number of partitions per node.
> If there is only one partition writing is basically linear on the disk,
> an therefore extremely fast. If you on the other hand linearly write to
> a large number of files at the same time, the disk access pattern starts
> to look more random.
> It all very much depends on how you IO subsystem can handle appending to
> many files in parallel. For example a RAID controller with a large
> battery backed cache should help a lot.
>
> Cheers,
> Nico
>
> Am Freitag, den 01.10.2010, 14:56 -0700 schrieb Dan Reverri:
>> The value of ring_creation_size dictates how many partitions your Riak
>> cluster will manage. These partitions are distributed amongst the
>> nodes of your cluster. Each node in the cluster with manage a portion
>> of the ring ( ring_creation_size / number of nodes ). A larger ring
>> means each node will be responsible for more partitions which means
>> each node will open more files (ulimit -n).
>>
>>
>> Thanks,
>> Dan
>>
>> Daniel Reverri
>> Developer Advocate
>> Basho Technologies, Inc.
>> dan at basho.com
>>
>>
>> On Fri, Oct 1, 2010 at 12:51 PM, Mojito Sorbet <mojitotech at gmail.com>
>> wrote:
>>         On Fri, 2010-10-01 at 09:24 -0700, Dan Reverri wrote:
>>
>>         >  Projects will typically define an upper bound for a
>>         cluster's size
>>         > based on available datacenter space, available power, cost,
>>         etc.
>>         > Choosing a ring_creation_size value based on this upper
>>         bound will
>>         > allow you to start small and grow into your cluster.
>>
>>
>>         Ok, but these days there is no fixed size "data center".  If I
>>         need more
>>         resources I can add them dynamically from Amazon or an
>>         equivalent cloud
>>         provider.  If I accidentally create the next Facebook and it
>>         keeps
>>         growing, it would be nice not to have to backup and restore
>>         the whole
>>         thing as it grows into multiple hundreds of terrabytes,
>>         especially
>>         considering the downtime required.
>>
>>         What is the penalty for making the ring size a lot bigger than
>>         the
>>         initial number of nodes would require?
>>
>>
>>
>>
>>         _______________________________________________
>>         riak-users mailing list
>>         riak-users at lists.basho.com
>>         http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
>>
>>
>>
>> _______________________________________________
>> riak-users mailing list
>> riak-users at lists.basho.com
>> http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
>
>
>
> _______________________________________________
> riak-users mailing list
> riak-users at lists.basho.com
> http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
>

-- 
Sent from my mobile device




More information about the riak-users mailing list