Unlock a world of possibilities! Login now and discover the exclusive benefits awaiting you.
Hi!
I know this rule was valid for 8.5 and 9.0.
Is it advised to disable Intel's hyperthreading with QV11? (servers and developer's PCs)
regards
That's very true. It depends on the size of the L1 cache as hyperthreading simply splits it in half to make a single processor appear as if it were 2. If the L1 is too small, splitting it can mean the processing pages QV is writing are too large for the available cache and they get writen to RAM instead which is many times slower. The effect is similar to running out of available physical RAM and data being writen to page files instead.
I've seen reload times half in some environments simply by turning hyperthreading off and in others not seen any noticable difference at all.
I don't believe this has changed in QV11. There is a scalability group though http://community.qlik.com/groups/qlikview-scalability, if you ask there someone might have more information on this.
I asked QT this recently and was told it hadn't changed.
Note that in some environments disabling Hyperthreading doesn't seem to make any difference but sometimes the difference it can make is vast.
That's very true. It depends on the size of the L1 cache as hyperthreading simply splits it in half to make a single processor appear as if it were 2. If the L1 is too small, splitting it can mean the processing pages QV is writing are too large for the available cache and they get writen to RAM instead which is many times slower. The effect is similar to running out of available physical RAM and data being writen to page files instead.
I've seen reload times half in some environments simply by turning hyperthreading off and in others not seen any noticable difference at all.
Thanks for the explanation Matt. I've always wondered why and under what circumstances it made a difference.
-Rob
It puzzled me for a while Rob until I spotted it was linked to the processor and finally narrowed it down to the L1.
Thanks Matt.
I got a bit of idea on it too..
Deepak