[cadynce] slides for telecon

[Adam Porter]

Hi Gautam,

	Here's what I can do quickly.

	
On May 22, 2007, at 10:35 AM, Gautam Thaker wrote:

>
> I have a question. we would like to try to do model checking to see  
> if we can actually generate 1000 different *expected* mean end to  
> end times for appstring A in GT-4. Towards this end, where can we  
> find the 1000 allocations that you actually used in these 1000  
> tests? On the wiki I can only find the 300 allocations that CMU  
> provided for GT-4. (I believe you started with these 300 and did  
> perturbations to get 1000 runs, right? It would be good if all the  
> 1000 allocations are on the wiki and a table showing:
>

We have 1014 configs tested in the final data set (we threw out a  
bunch of earlier runs as we worked through various bugs).
About 614 configs are unique  (don't have the exact numbers in front  
of me). About 400 are replications or permuations.


> Allocation_#   CUTS_Measured_Mean_E2E_time       
> Model_Predicted_E2E_Time
>

I've attached a link to a file (called A.Mean) with 2 columns
www.cs.umd.edu/~aporter/Cadynce/A.Mean

	Allocation_#  CUTS_Measured_Mean_E2E_time

The Allocation_# is of the form  "name_UUID". For example  
allocation_p2_9_627CDA9755582E8ED5D56CBB5C97DD9A or
structure_packed_4E677BDF51B3D6246B5BD28DEC73E873.

If two runs have the same name, but different UUIDs then they are  
either replications of each other or permutations.

There can be a slight mismatch between my Allocation_# and those  
provided by CMU because I generated some configs by hand.

Therefore, I'm also attaching a link to a second file (called  
configs.txt). It has
www.cs.umd.edu/~aporter/Cadynce/configs.txt

	Allocation_#  150_component_columns

The allocation_# is the same ID that appears in the first file

Each component column gives the node to which the component was  
assigned.

The order of the component columns is listed in the first line of  
this file.


> This will be 3 columns in all.  (As you know we have been focusing  
> on producing such 3 columns using not the mean but the worst case  
> time, a problem that is much tougher due to jitter in these large  
> systems.) But for now let us concentrate on the mean performance  
> in  time shared system where all priorities are equal. (Of course  
> we don't quite know the time shared quanta, or do we? This could  
> have some impact on column 3.)

We can generate worst case times if you want. It will just take some  
time to update my data processing scripts. Let me know what you
prefer.

I don't know the time shared quanta offhand. Will have to see what it  
would take to find out.

> IF we start with the CMU's 300 allocations of GT-4 and generate 3rd  
> column of data for each of these 300 allocations may be we can  
> generate a table of 300 rows for now and make a graphic for  
> possible use on 24th.

Yes. That would be great!

>
> Gautam
> PS I believe your processor pool consists of identical machines,  
> right? Thus, no surprise that processor permutation did not make  
> any difference, though good to see it.)
>

Yes. They are supposed to be identical machines, but this data  
confirms it.


Adam