after watching a presentation about how Speedus can optimize 110% the Hazelcast performance I decided to give it a try.
I did the following setup, I downloaded the Ubuntu docker image supplied with speedus. My host operating system is MacOs with 8 CPUs.
I am using domain driven design and insert rich objects into the Hazelcast Map from 8 parallel threads. The average size of the objects after serialization is probably in the range 5kb - 10kb plus minus. For serialization I use IdentifiedDataSerializeable.
Not only that I was not able to observe any performance increase. But actually my Torusware setup had 30% performance decrease during PUT of 150 000 objects inside a distributed map.
I tested a two node setup insertion from 8 threads. The result was even worse, probably near 100% slower than my normal setup.
Is there something I am not doing right? According to the Torusware site this thing is supposed to be plug and play. I can observe in the console that SPEEDUS is loaded, but the performance goes terribly wrong.