QlikView Server slower with more CPUs

[pover]

Hello all,

I have a machine with 1 CPU, 4 cores, 8 system threads and 8 GB RAM which calculates the same graphs in half the time as a machine with 2 CPUs, 8 cores, 32 system threads and 64 GB RAM. Both have QV Server V9 SR5 and neither machine is virtualized.

Has anybody experienced something similar?

Thanks for the help.

Regards,

Karl

[pover]

Just to give some kind of close to this issue, I'l like to add that disabling the hyperthreading didn't change the performance of the QV Application on the server. IBM delivered another server that is a slighly older model and finally the performance of QV was acceptable. After confirming another similar case that also involved a IBM server, we've come to the conclusion that the newest IBM server models don't seem to work well with QlikView. We're waiting for IBM to give us a solution.

[#3]

Hej Karl,

Windows Server 2008 is running on both IBM machines?

Do you are using the R2 edition of w2k8?

 The basic notion behind SMT parking is that the Windows scheduler will attempt to schedule threads so that all physical cores are occupied before any core gets two threads scheduled on its two front-ends (or logical cores). Since Hyper-Threading involves some cache partitioning and other forms of resource sharing, this is a potentially important feature. We've seen scheduler quirks cause poor and oddly unpredictable performance on Core i7 processors in the past. Based on our limited experience testing with Windows 7 and a cadre of SMT-enabled processors for this review, our initial impressions of SMT parking are positive. We've seen performance results for executables that rely on the Windows scheduler for thread allocation that match the performance of executables with explicit, SMT-aware thread affinity built in. Our initial sense is that SMT parking blunts some potential disadvantages of Hyper-Threading, making it more of an unqualified win, even on the desktop.

Source: http://techreport.com/articles.x/17545/2

[pover]

Yeah, they both have Win 2008 R2. IBM is continuing to do tests to see what is wrong.

Regards.

[#5]

Definitely sounds like a hardware issue. I'm glad that IBM is taking care of it. I've been working on a similarly spec'd system from Dell and performance has been wonderful. There are very few of these systems out there, so if you need a benchmark on our server I may be able to help.

[jorgefsilva]

Hi Karl, do you had any answer from IBM to resolve this issue?

I've a scenario similar to your's... do you solve it?

Thanks,

Jorge

[#7]

I think QV10 takes into account the multiple cores alot better than QV9.

[pover]

No news yet from IBM. They were able to improve the performance, but it still doesn't run as fast as my laptop. If and when I find out something, I'll post it.

We tested QV 10 and we didn't see any improvement.

Regards.

[pover]

Ok, so we've tried 2 other large industrial strength servers from another Intel family and HP and every server a process QlikView chart even slower then a smaller, older server and my laptop. You might say that the server would run better with more users, but with 2 users (56 seconds per chart) the servers take almost twice the time as 1 user (34 seconds per chart). I guess it might run better with 100 users, but this is crazy.

I apparently have a magic laptop because the same graph takes 15 seconds per chart.

Does anybody know about any of the following:

- Has anybody had any relation with the QlikTech Scalibility Center?

- Who else is having a similiar situation?

Regards.

[#10]

Hi,

Are there some complicated formulas in those charts, or just a lot of data?

Disable CPU hyperthreading. Suppose there is sizeable amount of memory, so don't let Windows automatically manage the size of the pagefile. Use a fixed amount of pagefile, ex 2 GB. The old mantra swap = two times the memory does not apply for a tool called "in memory analytics"; you really need data to fit in memory.

To do a one-to-one comparison with laptop, limit QVS service from using all CPU cores from QEMC : System : Setup: QlikView Servers : Performance : CPU affinity. Choose same amount of cores as in the laptop. Use the Windows task manager to really see the rest of the CPU cores idle.

-Alex

[#11]

Hi Karl,

here are a suggestion from my colleagues:

Some time ago we had some performance problems in with newer machines as well. It was caused by the Win2008 default power plan "Balanced". We experienced much better (script) performance when we turned the power plan to "High Performance".
Maybe it's the case in your scenario as well..
I changed the Power Plan from "Ballenced" (which is default, so check after install OS ?) to "High Performance", since Windows 2008 SR2 does not support Intel Turbo Boost in other modes.
So it takes now 23 minutes (a little slower than on our server, but that's plaubible because of lower CPU clocking / smaller cache).

from MS Performance Tuning Guidelines for Windows Server 2008 R2, Page 14:

Processor Performance Boost Policy :
Intel Turbo Boost Technology is a feature that allows Intel processors to achieve
additional performance when it is most useful (that is, at high system loads).
However, this feature increases CPU core power consumption, so we configure Turbo
Boost based on the power policy that is in use. Turbo Boost is enabled for High
Performance power plans and disabled on Balanced and Power Saver plans for the
current generation of processors. For future processors, this default setting might
change depending on the power efficiency of such features. To use the Turbo Boost
feature under the Balanced or Power Saver plans, you must configure the Processor
Performance Boost Policy parameter.
The Processor Performance Boost Policy is a percentage value from 0 to 100. The
default value of this parameter is 35 percent on Balanced and Power Saver plans. Any
value lower than 51 disables Turbo mode. To enable Turbo Mode, set this value to 51
or higher.
The following commands set Processor Performance Boost Policy to 100 on the
current power plan. Specify the policy by using a GUID string, as shown below:
Powercfg -setacvalueindex scheme_current sub_processor 45bcc044-d885-
43e2-8605-ee0ec6e96b59 100
Powercfg -setactive scheme_current

Note that you must run the powercfg -setactive command to enable the new
settings. You do not need to reboot the server.
To set this value for power plans other than the current selected plan, you can use
aliases such as SCHEME_MAX (Power Saver), SCHEME_MIN (High Performance), and
SCHEME_BALANCED (Balanced) in place of SCHEME_CURRENT. Replace "scheme
current" in the powercfg -setactive commands shown above with the desired alias to
enable that power plan. For example, to adjust the Boost Policy in the Power Saver
plan and make Power Saver the current plan, run the following commands:
Powercfg -setacvalueindex scheme_max sub_processor 45bcc044-d885-43e2-
8605-ee0ec6e96b59 100
Powercfg -setactive scheme_max

Good luck!

Rainer