Thanks Denis. From: Denis Magda [mailto:dma...@gridgain.com] Sent: Friday, June 24, 2016 1:00 AM To: user@ignite.apache.org Subject: Re: performance issues
Hi Pradeep, Member-member (server-server) performance is better than client-server one in the scenario because in the first case roughly a half of data will be stored on the server that executes the benchmark (meaning that there won’t be I/O at all). While in case with client-server case the client always has to send the data to a remote server. In fact if you run more servers on different physical machines and start the benchmark using the client node the performance should be better rather with the configuration with less servers. — Denis On Jun 22, 2016, at 6:06 PM, Pradeep Badiger <pradeepbadi...@fico.com<mailto:pradeepbadi...@fico.com>> wrote: Hi Denis, Thanks for your response. I ran the benchmark test on a single VM with 8 cores and 2Gb allocated to each server instance (member). All the tests were for optimistic transactions with no backup and with the default JVM settings in benchmark.properties. I see that the performance of client-member is lower than the member-member setup. The client was configured using –client in the benchmark.properties. I have attached the configuration for 1 client 1 server mode. I was trying to run it on a single server with no clients but I was not able to configure that in the benchmark test. I am not sure if it is worth to test that scenario, as all the calls would be local and latency would be lot less. Please let me know if the member-member performance is lot better compared to client-member performance. Clients Servers Threads Latency (ms) Min Avg Max 0 2 8 0.341 0.451 0.881 0 2 8 0.362 0.470 0.874 1 2 8 0.695 0.749 1.070 1 1 8 0.576 0.726 1.102 Thanks, Pradeep V.B. From: Denis Magda [mailto:dma...@gridgain.com] Sent: Thursday, June 16, 2016 6:46 PM To: user@ignite.apache.org<mailto:user@ignite.apache.org> Subject: Re: performance issues Hi, In a single server mode (embedded) there is no I/O (networking) at all. All the requests are executed locally. When you use the client node you’ll have I/O delays. The performance here can depend on several factors: - latency and throughput of your network; - CPU saturation; So take a look at these system resources usage. Also make sure that there are no long GC pauses or that GC Threads don’t consume much of CPUs. Refer to this doc for JVM tuning settings https://apacheignite.readme.io/docs/jvm-and-system-tuning Finally, share your benchmark source and configuration for validation. — Denis On Jun 16, 2016, at 11:04 PM, Pradeep Badiger <pradeepbadi...@fico.com<mailto:pradeepbadi...@fico.com>> wrote: Hi, I am trying to run the yardstick ignite benchmark test on my local VM having 8 Cores and 16GB RAM. I could see that the performance of Optimistic PUT/GET is way low for a client-server mode than what I see when running within one single server (embedded mode). Also the performance degrades with 1 additional server node. Can someone help me to optimize this? There was a blog where the author commented that the performance of the product is much similar when run in both client and server. Am I missing something here? https://gridgain.blogspot.com/2015/04/benchmarking-data-grids-apache-ignite.html?showComment=1466104423051 Thanks, Pradeep V.B. This email and any files transmitted with it are confidential, proprietary and intended solely for the individual or entity to whom they are addressed. If you have received this email in error please delete it immediately. This email and any files transmitted with it are confidential, proprietary and intended solely for the individual or entity to whom they are addressed. If you have received this email in error please delete it immediately.<benchmark.properties> This email and any files transmitted with it are confidential, proprietary and intended solely for the individual or entity to whom they are addressed. If you have received this email in error please delete it immediately.
