I would like to take your input on this interesting situation.... We are on VM and have 5 Million records in memory table.. When next statement executes which is resident laod from this table to another table,, It TAKES 30-45 minutes with 16 cores cpu.... But if i RUN the same on 4 cores, it take only 10-15 minutes.... I see with 16 cores all cpus seems to be in 100% most of the time,,,
Any insight will be apprecaited to understand relaod in publisher. We are on 11.0 SR2
When you mention numbers of 'cores' do you mean numbers of virtual processors ?
For the physical host how many physical cores does it have ?
The reason why I ask the above is that for a virtual server its number of virtual processors should always be less than half the number of physical host's, physical cores. Otherwise processing can end thrashing with context switching and performance plummets.
I had a scenario [non QlikView] some years ago with a physical host with 4 cores, that had a virtual server with 2 virtual processors that was performing badly. Somebody decided to double the virtual processors to 4 and its performance become even worse.
I was called in and against much opposition decreased the virtual server processors to just 1 virtual processor and performance improved dramatically. I did win the bet and was made a lovely cup of tea.
So I would say if your physical host has 32 or less physical cores then your findings above are exactly what I would expect.
Yes 8 cpu * 2. can you explain on what do you mean on phusical cpus and what we assigned to publisher. It is clear in this case that performance improved with 4 or 8 cores over 16 on vm. Why performance goes bad with more cores?