Gordon,

It's worth pointing out that in 1.0 this should be greatly improved because
now we have changed vclock behavior in relation to PUTs [1].  Essentially,
client id now comes from the vnode, not externally, which leads to
smaller/static vclock sizes even in the face of frequent updates.

I can't say with complete certainty that reducing the small/big vclock won't
change your perf characteristics but I will say it is probably better to
have smaller vclocks which in turn leads to smaller objects.  Since pruning
will happen more often I would imagine your PUT latencies could potentially
go up a little.  The best way to know is to measure.

Another option is setting last_write_wins to true but this should only be
used as a last effort and only if you truly don't care about siblings [2].

-Ryan

[1]: Checkout the section titled "Get/Put Improvements" in the release
notes.  https://github.com/basho/riak/blob/riak-1.0.0/RELEASE-NOTES.org

[2]: http://wiki.basho.com/Vector-Clocks.html#Siblings

On Fri, Sep 2, 2011 at 11:33 AM, Gordon Tillman <[email protected]> wrote:

> Morning All,
>
> I have noticed that with the default vclock settings the size of the
> X-Riak-Vclock header gets really big really fast for certain objects that
> are updated frequently.  I am considering setting small_vclock to 4 and
> big_vclock to 16 instead of the default values of 10 and 50, respectivly.
>
> I know that that are various performance optimizations that are based upon
> certain default bucket settings and was wondering if this change would
> adversely affect any of those.
>
> Many thanks!
>
> --gordon tillman
> _______________________________________________
> riak-users mailing list
> [email protected]
> http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
>
_______________________________________________
riak-users mailing list
[email protected]
http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com

Reply via email to