We've found that c3.large instances in Amazon AWS work well for ZooKeeper instances. We run 5 nodes of ZK across 3 Availability Zones. Load is almost always zero, around 50-200kb/s network traffic in/out, ~500-750kb/s disk writes, ~135MB RAM used per node. JVM heap settings are default which works out to ~937.5MB on a c3.large.
Each node is handling ~50 connections and doing about 100M sent/received a month. Our node count is pretty low though, only a few thousand. We use ZK for coordinating Apache Kafka as well as service discovery. Does anyone run ZK anywhere close to "hot", i.e. 60-80% machine utilization? On Thu, May 29, 2014 at 8:09 AM, James A. Robinson < [email protected]> wrote: > Not really much information, but we're running a 5-node > zookeeper cluster, all the nodes are within the same > local network. We're using it to help coordinate Apache > Kafka and we're going to be using it to help coordinate > a bit of configuration for some web crawlers we'll be running > against sites we manage. > > Jim >
