Re: Difference between CPU and core.
Date: Mon, 17 Dec 2012 23:05:18 -0600
Message-ID: <CACNsJnesqAfNWBMx3U21a3z9HVt571NPTu-1bvx8_93+fhoQRw_at_mail.gmail.com>
Most likely you are using this processor http://ark.intel.com/products/53579/Intel-Xeon-Processor-E7-4870-30M-Cache-2_40-GHz-6_40-GTs-Intel-QPI and you have the hyperthreading turned off on server #1.. CPUs
40
Cores
40
Sockets
4
the CPUs in AWR are the Logical CPUs (show parameter cpu_count) which the OS also treats 1 thread as 1 cpu (cat /proc/cpuinfo)... but ultimately without instance caging turned on the cpu_count parameter and the number of CPUs on /proc/cpuinfo should be the same
On the workload level (AWR load profile).. a system let's say on the
example (photo)
https://lh5.googleusercontent.com/-oL5_-FgDcjs/UM_yHxBiquI/AAAAAAAABxo/4GR4QFGDhn4/s2048/threadsVScores.png
1socket, 4cores, 8threads which shows as 8 Logical CPUs
(both CPU_COUNT and /proc/cpuinfo) can just do a max LIOs/sec of 1,400,000M
and even if you saturate past the number of Logical CPUs you would
only be on that rate.
To measure the response time effect as you saturate up to the max CPUs
count you have to measure the session level numbers.. so here
as you use more and more CPUs the lesser LIOs/elapsed you can do and once
you reach the max CPUs count you start to see some wait on run queue
which ultimately affects the session response times.
The four test cases shows the effects of having HT and TurboBoost on or
off. In my case when I turned off the HT it only left me with 4 CPUs
although I can still handle the same workload the response time effects
manifest on the CPU wait (run queue) once you get past the max CPUs count
.. that's why HTonTurboOn is the best case
And.. I would suggest you go to this tiddler http://goo.gl/iUJyn and open this cool document by RedHat http://goo.gl/MKsml which clearly explains the HT stuff.. (redhat kbase doc DOC-7715 - multi-processor, multi-core or supports hyperthreading)
-- Karl Arao karlarao.wordpress.com karlarao.tiddlyspot.com twitter.com/karlarao -- http://www.freelists.org/webpage/oracle-lReceived on Tue Dec 18 2012 - 06:05:18 CET