ZooNet is a prototype of our modular composition concept over ZooKeeper. ZooNet allows users to compose multiple ZooKeepers ensembles in a consistent fashion, facilitating applications that execute in multiple regions. In ZooNet, clients that access only local data suffer no performance penalty compared to working with a standard single ZooKeeper. Clients that use remote and local ZooKeepers show up to 7x performance improvement compared to consistent solutions available today.
ZooNet appeared in Usenix ATC'16, the paper can be found here.
Please see the following class for the main logic: ZooNet.java .
In order to run ZooNet's systest, run the following commands:
- "ant jar compile-test" (from ZK base dir).
- "ant jar" (from "src/contrib/fatjar").
- "./bin/runZooNetSystest.sh" (from ZK base dir). This script is found here.
After initialization, you'll see the following printouts once every 6 sec (of throughput and latency percentiles, per every application):
[app name]
[# of local read requests] [# of local update requests] [# of remote read requests] [# of remote update requests]
Meadian: [local reads, in ms] [local updates, in ms] [remote reads, in ms] [remote updates, in ms]
0.95: [local reads, in ms] [local updates, in ms] [remote reads, in ms] [remote updates, in ms]
0.99: [local reads, in ms] [local updates, in ms] [remote reads, in ms] [remote updates, in ms]
For the latest information about Apache ZooKeeper, please visit: