A number of points:

* The replication factor (N) * the total original data size = total data size 
stored in the cluster.  For example, if N=3, then you have 300GB of data stored 
across the cluster.  If you had 6 nodes, this would be about 50GB per node.
* Riak can be optimized in your setup for GET speed, but in general it is 
optimized for fault-tolerance and availability.  Even in the best of 
conditions, it will not perform as optimally as a web server reading static 
files off the local disk.
* The replication factor will affect both reads and writes.  If your cluster 
size is larger than your N, you should see increased throughput as you add 
nodes.

Sean Cribbs <[email protected]>
Developer Advocate
Basho Technologies, Inc.
http://basho.com/

On Apr 12, 2010, at 1:19 PM, TuX RaceR wrote:

> Hello Riak users!
> 
> Let me first present on a simple example what I can see as an application 
> that 'scales':
> If a have a web server (e.g apache) serving 100Gb of files and I get 4000 GET 
> per seconds, one way to scale the application is to copy (rsync) the 100Gb to 
> another web server and balance the load on the two nodes: that way, globally 
> my cluster of 2 nodes can serve 2 x 4000 GET per seconds.
> I can say that my architecture scales as if I multiply by 2 the number of 
> nodes, then I can multiply by 2 the number of requests that the system can 
> handle (per second).
> 
> With Riak, I am not sure to understand how the scaling works. Are we speaking 
> about a global 'key GET' rate (request per second) that scales with the 
> number of nodes added?
> My web server example above also assumed that all the data (100Gb) could fit 
> into a single node. As I understand Riak could be used to serve data too 
> large to fit on one disk. So maybe the scaling is about the data itself: a 
> web client (browser) will not see a speed difference in the response from a 
> riak cluster serving K keys with N nodes and another riak cluster serving 2*K 
> keys on 2*N nodes.
> 
> Also the number of replicas role in scaling is not clear to me: it seems to 
> me that having a lot of replicas speeds up reads but slows down writes. Is 
> there a simple scaling law for this?
> 
> Thanks in advance,
> 
> TuX
> 
> _______________________________________________
> 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