eLevelDB max_open_files in 1.2.0

Mark Phillips mark at basho.com
Fri Aug 17 10:12:44 EDT 2012

Hi Dave,

On Thu, Aug 16, 2012 at 9:41 AM, Dave Brady <dbrady at weborama.com> wrote:
> Thanks, Mark!
> I have a somewhat-related follow up question: is there a recommended maximum for the amount of data held on a machine?
> I ask because at my previous company Cassandra was used, and we were advised to put no more 250 GB per physical box.  The reasoning was that in a failure situation, rebuilding any more than this amount of data would cause too great a performance degradation.

So there's no hard and fast amount of data per node that you shouldn't
be passing when using Riak. Rather, it makes more sense to keep in
mind the resources needed when a node (or set of nodes) becomes
unreachable, dead, or is sold on ebay by a sysadmin trying to raise
money for a ping pong table..  Plan appropriately for capacity and
Riak should operate as advertised . 250GB/node shouldn't be an issue
as long as you understand the resources needed to keep the cluster
going in failure and growth scenarios. (Anecdotally, I can tell you I
know of one Riak user that has >5TB/physical node on a cluster with
10s of nodes.)

Ryan Zezeski also pointed out that Eric Brewer wrote a paper in 2001
[0] that touches on this and is worth a read (much like all the papers
he writes).

Hope that helps.


[0] http://www.cs.berkeley.edu/~brewer/Giant.pdf

> --
> Dave Brady
> ----- Original Message -----
> From: "Mark Phillips" <mark at basho.com>
> To: "Dave Brady" <dbrady at weborama.com>
> Cc: riak-users at lists.basho.com
> Sent: Tuesday, August 14, 2012 7:34:10 PM GMT +01:00 Amsterdam / Berlin / Bern / Rome / Stockholm / Vienna
> Subject: Re: eLevelDB max_open_files in 1.2.0
> Hi Dave,
> On Sun, Aug 12, 2012 at 3:58 PM, Dave Brady <dbrady at weborama.com> wrote:
>> Hello,
>> First I want to thanks Basho for greatly expanding the documentation on the Wiki for configuring/tuning Riak and eLevelDB in 1.2.0!  Big improvement over 1.1.x.
>> My question is about max_open_files: here the documentation is confusing to me.
>> It says to allocate one open file per 2 MB, then divide by the number of partitions.  This is the same formula used in 1.1.x.
>> It goes on to say that if you manually set this parameter in 1.1.x, to divide that value by two for 1.2.0.
>> Should not the formula for 1.2.0, in that case, read as use one file per 4 MB?
> Long story short, the answer is "yes" :)
> With 1.2, 4MB is the advised file size, and you should be running with
> no less that 20 files/nodes. I'll take a pass at updating the docs to
> make this a bit easier to understand. Thanks for pointing that out.
> Mark
>> Thankks!
>> --
>> Dave Brady
>> _______________________________________________
>> riak-users mailing list
>> riak-users at lists.basho.com
>> http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com

More information about the riak-users mailing list