I want to highlight here that the performance degrades with 2 or 3 servers node configuration simply because all the nodes are running locally on a single machine. The nodes have to fight for system resources.
If to run the same benchmark having nodes on different physical machine the numbers will be better. — Denis > On Jan 6, 2017, at 12:53 AM, Nikolai Tikhonov <ntikho...@apache.org> wrote: > > Hi, > > > 1. Can some one try my test program in order to see what is the problem > > here? > > First, you don't have problem here. :) Second, your benchmark is incorrect. > Read the thread on SO for additional information how correct to write > benchmark: > http://stackoverflow.com/questions/504103/how-do-i-write-a-correct-micro-benchmark-in-java > > <http://stackoverflow.com/questions/504103/how-do-i-write-a-correct-micro-benchmark-in-java> > I've changed test methods: put/get executed 3 times (warmup) > > try (IgniteCache<Integer, Integer> cache = ignite.getOrCreateCache(cfg)) { > put(cache); > get(cache); > > put(cache); > get(cache); > > put(cache); > get(cache); > } > > and let's see how changing numbers: > > One client 2 servers > >>> Put completed. Took 24928 ms. > >>> Get completed. Took 13272 ms. > >>> Put completed. Took 14010 ms. > >>> Get completed. Took 9331 ms. > >>> Put completed. Took 13357 ms. > >>> Get completed. Took 9131 ms. > > After this changes I got the following test results: > > 1 node: > >>> Put completed. Took 2042 ms. > >>> Get completed. Took 399 ms. > > 2 nodes: > >>> Put completed. Took 7791 ms. > >>> Get completed. Took 4218 ms. > > 3 nodes: > >>> Put completed. Took 10474 ms. > >>> Get completed. Took 5412 ms. > > It's ok for me. When you start one node all changes was made locally > (absolutely without network communication), 2 nodes - half of operation will > be locally and etc. If we perform operation from client node then we will get > the following numbers: > > One client and two servers nodes > >>> Put completed. Took 13357 ms. > >>> Get completed. Took 9131 ms. > > One client and one server node. > >>> Put completed. Took 12132 ms. > >>> Get completed. Took 8714 ms. > > It shows that we have approximately the same values. > > > On Fri, Jan 6, 2017 at 7:05 AM, bintisepaha <binti.sep...@tudor.com > <mailto:binti.sep...@tudor.com>> wrote: > Sam, could you post your cache configuration? How many backups do you have? > What is the marshaller you are using? is there garbage collection happening? > > Thanks, > Binti > > > > -- > View this message in context: > http://apache-ignite-users.70518.x6.nabble.com/Performance-with-increase-in-node-tp9378p9921.html > > <http://apache-ignite-users.70518.x6.nabble.com/Performance-with-increase-in-node-tp9378p9921.html> > Sent from the Apache Ignite Users mailing list archive at Nabble.com. >