zookeeper-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From <mallikharjun.vem...@thomsonreuters.com>
Subject RE: Optimal ZooKeeper configuration for a cluster of size, N
Date Tue, 09 Oct 2012 06:02:44 GMT
Thanks for the response.

Session timeout is our major concern. I understand that we can give a higher value for session
timeout to be on safer side. But, in our environment when we give a bigger value for session
timeout, our system would be wasting time in retries before giving up the 'task'. As a result
we end up syncing huge amount of data (stored in our system, not in zookeeper) when the system
reconnects and resumes its 'task'. Hence we want to keep the session timeout value as small
as possible to minimize our syncing overhead. In other words, we would like to keep the timeout
value just enough for retrying majority of the servers (or may be a little more as a buffer).
I would like to know this value for 3 & 5 ZooKeeper server deployment. 

Thanks in advance.

-----Original Message-----
From: Patrick Hunt [mailto:phunt@apache.org] 
Sent: Saturday, October 06, 2012 7:59 PM
To: user@zookeeper.apache.org
Subject: Re: Optimal ZooKeeper configuration for a cluster of size, N

The defaults should be fine for what you described. The initLimit would need to be increased
if the data being stored is very large, the syncLimit increased if you were running across
say a wan link (btw
servers) and leaderServes is only really useful if you have a large number of writes/clients
and you want the leader to focus on coordination (rather than that plus serving client requests.)


On Thu, Oct 4, 2012 at 2:41 AM,  <mallikharjun.vemana@thomsonreuters.com> wrote:
> We are using ZooKeeper 3.4.4 for our deployment. I would like to know what would be optimum
config setting for various configuration parameter viz., initLimit, syncLimit, leaderServes,
etc. As of now we are using default parameters for these parameters. But, I would like to
understand how each one of them depend on the size of the zookeeper cluster.
> Say I have N servers in the cluster, I would like to know how each of these values can
be configured (in terms of N) for optimum results. Also, I would like to understand how session
timeout would effect the performance of the cluster.
> In our deployment,
> - we are basically using the zookeeper service to detect failures in our system.
> - we are storing very minimal data in the cluster.
> - we are planning to have 3 or 5 servers in the cluster.
> Can anyone please throw some light on these config settings? I did go through the admin
guide. But, I would like to understand each one of them from the perspective of zookeeper
cluster size. Please help.
> Thanks,
> Arjun.

View raw message