Re: [squid-users] Hard Drive Latency

On Mon, 30 Aug 2004, Thomas-Martin Seck wrote:

> I'd like to compare 4.10+diskd against 5.3+diskd/aufs on the same
> hardware with web polygraph. Is there a specific workload I can (or
> should) use for this (I did not work with polygraph yet and am just
> studying the manual)?

polymix4 is the best so far.

but problem is that it is not trivial to make the compare you are looking
for with polymix.. all the polymix tests assumes you have already
determined the performance this box should have and all the test says is
if the box can keep up with the specified load, not what the maximum load
the system is capable of handling is..

I have a modified polymix4 workload which I call "polyboil4" which is
quite useful in determining this. This workload dynamically increases the
load until there is errors and is quite helpful in determining the limit
of the system.

Unfortunately the license of Polygraph does not allow me to distribute
this modified workload, but I can send instructions on how to build the
workload based on polymix4.. searching.. found.. translating...

%%% How to make polyboil-3 wrokload %%%

We started from the standard workload, but instrad of #include
<polymix-3-guts.pg> vi included the file contents directly and made the
changes below. These changes can be done in the same manner in polymix-4
except that load_factor_end then has to be replaced by populus_factor_end

1. Added a definition of a "bolier" phase

Phase phBoil = {
         name = "boil";
         load_factor_end = 1.0;
         goal.duration = 4hour;
         goal.errors = 1;
};

2. Changed Inc1 making the boiler start at a lower rate than norma

Phase phInc1 = {
         name = "inc1";
         goal.duration = rampDur;
         load_factor_end = 0.05;
};

3. Changed the schedule to use the new boil phse instead of top, and
shortened it to one run.

// build schedule using some well-known phases and phases defined above
schedule(
         phWarm,
         phFRamp, phFill, phFExit,
         phInc1, phBoil
);

5. Ve also have a shortened version running with the assumption that the
cache has already been filled. To make this we replaced the "fill" phase
with the following:

Phase phFill = {
         name = "fill";
         goal.duration = 1min;
         wait_wss_freeze = yes; // will finish only if WSS is frozen
};

%%% End polyboil-3 workload sheed %%%

I have not yet written down exact instructions based on Polymix-4, but as
you can see above the difference in what has to be done is pretty minimal.

With the short version it becomes pretty quick to investigate the impact
of different changes once the initial setup of the test has been done. The
initial filling of the cache is quite time consuming but a required step
to get meaningful results.

Regards
Henrik
Received on Mon Aug 30 2004 - 09:52:55 MDT