Re: response time

Thanx for the response alex.

The constant value was smaller than the working set size according to
polymix-4 default calculations based on request rate,fillable ratio,host 
count,etc.
There wasnt any other change between the two tests other than this.


regards,
Senthil

Alex Rousskov wrote:

>On Thu, 10 Jun 2004, senthil wrote:
>
>>1.I tried changing the working_set_cap to a constant value. The hits
>>time acheived was very high about 15 seconds.
>>
>>I tried again with the polymix configuration for working_set_cap.
>>But now the time for a hit was good about 1 second. Any specific
>>reasons??
>>
>
>There can be many reasons. Is your constant value larger or smaller
>than PolyMix-4 value for the request rate you are using? Smaller
>working set sizes are easier for the cache to handle. Is it possible
>that you also changed something other than working_set_cap between the
>two tests?
>
>>2.And also I have set the ProxyCacheSize to be 500MB only such that
>>the Fill phase will not extend for a long time.Does this have any
>>impact on response time and hit ratio??
>>
>
>Not properly filling the cache often has a significant effect on the
>quality/reliability of all measurements, including response time and
>measured hit ratio. Many caches perform much better when they are not
>filled. In real world, the cache will be filled after some initial
>warmup time; we want to test under realistic conditions and fill the
>cache (twice).
>
>>3.If I need to have a higher hit ratio i need to change the
>>recurrence value.am i right??
>>
>
>Yes, increasing recurrence ratio will increase _offered_ hit ratio.
>Increasing cachability ratio will have the same effect. If you want to
>increase _measured_ hit ratio, you need to find out why it is lower
>than you want and go from there.
>
>HTH,
>
>Alex.
>