We recently migrated our Aerospike cluster from version 3.11 to the newest version 4.5.1.5 and found out that the new version consumes 50% more CPU (with the same load as before)
The cluster configuration is exactly the same, except for the fact that we are using 8MB record size instead of 1MB. Can this be the cause of such a huge degradation in performance
The image is hard to make out… but it looks like you’re just looking at CPU? What do you mean it’s slower? I know that between those versions there was a change that tracked network latency better, which would show higher values in histograms than previous versions - but want to understand what you’re seeing.
More information about the histogram change that @Albot mentioned can be found here: Write Latency Increase After Upgrade to Aerospike 4.4 or higher
Slower I mean, 50% more CPU usage over the same load. Another important information is that we have very few writes on this database, so must of the usage is on read / batch read operations
More CPU does not mean slower. CPU usage and latency are not related directly. If you measured latencies to be longer, then indeed you have a slower response. I don’t see that as being the case from your JPEG.
Also notice that the CPU usage is much tighter now, so less variations and more consistent workload.
indeed, it is not 50% slower. I put it wrongly. I changed the texto to “50% more CPU consumption” Anyway, this is the problem I’d like to tackle, because 50% more CPU for us means 50% increase in costs
I’m not sure I see how that translates to increased costs, unless you’re talking about power costs. Thanks for editing the text of your forum post, but your Twitter post still talks about Aerospike being slower on 4.5, which it isn’t.
You’re using community edition, the free version of Aerospike. What you can contribute is to open a good bug report in the aerospike/aerospike-server repo. Please include your config and the spec of your hardware.
Also, when asking for help please follow up with feedback and requested information. I have personally helped you by answering your questions on stackoverflow, but you tend to leave your own issues hanging, without feedback and without selecting the correct answer. The idea is to not just get an answer for your specific issue, but for the Q&A to also be helpful to the wider community.
Are you planning to share your configuration and hardware description?