On Mon, 17 Jul 2000, Ittai Gilat wrote:
> According to polymix-3 console output, inc1 phase has the following
> behaviour:
>
> 000.05| Phases:
> phase load_beg load_end rec_beg rec_end goal flags
>
> inc1 0.62 1.00 0.09 1.00 30.00min
>
> This means that the offered HR starts with 5% and after 30 min it
> should be 55%. At the same time the request rate also increases
> from 0.62 to 1.00 peak rate. The request_rate/HR factor may be too
> high for a proxy to handle. I also think that since the target HR is
> to be 55% for the whole duration of the test, HR should be already
> 55% when entering inc1. I suggest to add another small phase (it can
> be very short - 5 min?) in between fill and inc1, where HR increases
> to full 55% and the request rate is kept constant at fill rate.
I think Ittai's concern is a valid one.
To avoid extra parameters and to make the workload symmetric, the fix
can be in adding the framp2 phase that
- has the same duration as all other ramp phases
- increases HR from fill HR to peak HR
- decreases request rate from fill rate to 0.1*peak rate
(i.e., to the old start rate for the inc1 phase)
Pros: the cache is not stressed with a (high load, low HR) combo
the interesting part of the workload becomes symmetric
Cons: the test becomes 30 minutes longer
Comments? Objections?
Thanks,
Alex.
P.S. Setting framp2 duration to 5 minutes may not be a good idea as some
proxies may need more time to adjust for changing parameters of the
workload. IMO, it is better to keep all ramp phases equal.
This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:14 MDT