On 02/23/2013 01:54 AM, Michael Haberler wrote:
>> Hi Fritz,
>>
>> thanks for the data point!
>>
>> It confirms my theory that undeterministic recovery times from powersave 
>> states are a major (if not the) primary source of latency, and that factor 
>> is largely operating-system independent since it has no control over it
>>
>> any chance of permuting the 'C6 state' and 'cool-n-quiet' settings and redo 
>> measurements?
>>
>> - Michael
>>
>>

Certainly...for science!

Same hardware as before, but I put in a cheap Radeon 6450 PCIe card to 
give a little more performance on the desktop responsiveness.

Reset the UEFI settings, proceeding while changing only the EPU, 
Cool-n-Quiet (CnQ) and C6 settings.  Reboot between changes.

Same software as before.  Results are [servo] : [base]

+EPU  +CnQ  +C6        22534 : 20476
-EPU  +CnQ  +C6        26157 : 19378
-EPU  -CnQ  +C6        22729 : 19206
-EPU  -CnQ  -C6        8002 :  6628
+EPU  +CnQ  -C6        6385 : 4712

With the video card, the results seemed to be not as good as before, but 
since the C6 option seemed to be the key, I reverted the EPU and CnQ 
power options back on, and seemingly got a little better result.

Performing a lot of IO could make the glxgears grind down to single 
digit framerates.  The very worst test I could come up with was both 
disk and net IO with CPU load:  rsync the entire contents of the disk 
via ssh to another computer, but that would still only occasionally bump 
it up to little over 8000:8000 on the last test.

The 'Pro' version of this motherboard has a parallel port, accessible 
with a ribbon cable to a header.  I'm happy to report that EPP transfers 
appear to work correctly with a Pluto-P that I've been holding on to.

------------------------------------------------------------------------------
Everyone hates slow websites. So do we.
Make your web apps faster with AppDynamics
Download AppDynamics Lite for free today:
http://p.sf.net/sfu/appdyn_d2d_feb
_______________________________________________
Emc-users mailing list
Emc-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/emc-users

Reply via email to