Greetings,

I've been reading about ZK (and Hedwig) off and on for a few days, and
I have a small application where ZK might be useful, but there is some
doubt.

My present app has very low performance requirements; I'm looking to
maintain no more than a couple of thousand tiny znodes as configuration
items and coordination bits for a monitoring application.  I'm experimenting
with cross-site failover capability for various tiers of this application.
One of the requirements, however, is to be robust against a complete outage
of any one datacenter (client has been burned by such an outage recently...)

Is it possible to turn some knobs that would relax the latency requirements
in the ensemble voting process in return for giving up performance?  I really
don't need more than a few tens of writes per second for this app.

Thanks.

Hi David, I'm not sure I understand your question. When you say "relax  
latency requirements" and "giving up performance", do you mean that  
you don't mind having higher latency and lower throughput? If so, then  
I think you can get it straight away with ZooKeeper. Depending on how  
many data centers you have, and how far apart they ar, you may  
consider using a different quorum configuration using this mechanism:

http://hadoop.apache.org/zookeeper/docs/r3.3.2/zookeeperHierarchicalQuorums.html
http://hadoop.apache.org/zookeeper/docs/r3.3.2/zookeeperAdmin.html

-Flavio
On Dec 5, 2010, at 6:47 PM, David Carmean wrote:

> Greetings,
>
> I've been reading about ZK (and Hedwig) off and on for a few days, and
> I have a small application where ZK might be useful, but there is some
> doubt.
>
> My present app has very low performance requirements; I'm looking to
> maintain no more than a couple of thousand tiny znodes as  
> configuration
> items and coordination bits for a monitoring application.  I'm  
> experimenting
> with cross-site failover capability for various tiers of this  
> application.
> One of the requirements, however, is to be robust against a complete  
> outage
> of any one datacenter (client has been burned by such an outage  
> recently...)
>
> Is it possible to turn some knobs that would relax the latency  
> requirements
> in the ensemble voting process in return for giving up performance?  
> I really
> don't need more than a few tens of writes per second for this app.
>
> Thanks.
>
>

flavio
junqueira

research scientist

[EMAIL PROTECTED]
direct +34 93-183-8828

avinguda diagonal 177, 8th floor, barcelona, 08018, es
phone (408) 349 3300    fax (408) 349 3301