Basic scale-out deployment is to use multiple Ignite nodes deployed on multiple physical machines and stored data in partitioned distributed caches.
So, I would moving in this direction: - start a cluster of multiple Ignite nodes deployed on several machines. - launch the application from the other machine - there should be a warmup loop meaning that you’ve to perform this operations in a dry-run mode warming up the JVM - execute operations from multiple threads. At all, if you need to benchmark put and get it makes sense to have a look at Yardstick benchmark which is an official benchmarking framework used by Ignite. Yardstick has PutGetBenchmark that cover your case. https://github.com/apacheignite/yardstick-ignite — Denis > On Dec 5, 2016, at 11:02 AM, [email protected] wrote: > > Actual application is multiple machines and various threads doing get and > put. As a simple test which can provide average get and put time, I tried > attached test program. > > Tried same program on multiple physical machines (each with 48 CPU, 40GB > Ram) and see similar behavior. > > To rule out network latency, tried same on single physical machine (48 CPU, > 40GB Ram) and see similar behavior. > > Anybody faced this? > How are others using it? (single cache server or multiple nodes) > Are there any configurations which I need to tweak? > > -Sam > > > > -- > View this message in context: > http://apache-ignite-users.70518.x6.nabble.com/Performance-with-increase-in-node-tp9378p9404.html > Sent from the Apache Ignite Users mailing list archive at Nabble.com.
