Dear Mr. Anderson,

It is indeed overwhelming to hear from the PERSON himself.


Let me go directly to my Ph.D. work “Exclusion on Collective Impact – A
Novel Task Allocation Policy for Desktop Grids and its impact on Task
Retrieval Policies”. I started my Ph.D. two years back and have gone through
many papers but the papers that really convinced me to enter in this area
are given below:

1.       Derrick Kondo et. al. paper titled “Scheduling Task Parallel
Applications for Rapid Turnaround on Enterprise Desptop Grids” introduces
following Task Allocation policies:

*Inclusion Based on Priority*
Assign a task based on work unit priority.
*Exclusion on Fixed threshold*
Discard host request on the basis of fixed constraints given in work units
or project config file
*Exclusion on Make span Prediction*
Discard host request which would be unable to process request in predicted
time.


2.       David Toth et. Al. presents following task allocation policies in
“Improving the Productivity of Volunteer Computing by Using the Most
Effective Task Retrieval Policies”, each policy has different pros and cons:

   - ·         Buffer None
   - ·         Download Early
   - ·         Buffer 1 Task
   - ·         Buffer Multiple Tasks
   - ·         Super Optimal



3.        “Performance Evaluation of Scheduling Policies for Volunteer
Computing” by Derrick Kondo, David P. Anderson and John McLeod VII. This
paper becomes important because I am also using Boinc for the
implementation.

*The objectives of Thesis *

1.       To improve upon the Task allocation Policies by presenting a policy
called “Exclusion of Collective Impact”. We feel that the Exclusion on Fixed
Threshold of CPU cycles can discard many legitimate hosts because primary
memory of the host also plays a critical role in the processing. We
established a weighted mechanism that not only includes CPU speed but also
the memory, for example see the table below:

Scoring Scheme for CPU:

*C.P.U  Speed*

*Value*

2.0Ghz

1

2.4Ghz

2

2.53Ghz

3



Scoring Scheme for Memory:

*RAM*

*Value*

1GB

1

4GB

2

8GB

3



Now the rejection / acceptance of the host take into account not only the
CPU speed but also the memory.

To support our claim we implemented Boinc and Emboinc on five machines
having different configurations and gave them a toy task but frankly we did
not get the required results. We feel that if the task is big enough than we
can get some results.  *But we are unable to find larger workunits?*

2.       To analyze the impact of task allocation polices on task retrieval
policies. We feel that all the task retrieval polices presented in
“Improving the Productivity of Volunteer Computing by Using the Most
Effective Task Retrieval Policies” by David et. al. would not work well with
any given Task Allocation Policy.

If the Task Allocation policy of “Exclusion on Collective Impact” would give
the desired results, our next task would be to select the task retrieval
policy that would perform best with the proposed task allocation policy. But
in EmBoinc we are unable to find any mechanism through which we could test
the different task retrieval policies.  *Is there any other emulator
available that could perform the required task?*



I have explained my two years work to you and have tried to be very precise.
Kindly provide your input on the following:

1.       Does the proposed task allocation policy of “Exclusion on
Collective Impact” make sense to you?

2.       Do you think that the task retrieval policy has something to do
with task allocation policy?

3.       Should I continue to work upon this for my Ph.D. thesis as I have
just six months left of my scholarship?

If yes than

1.       Where can I get the larger work units or how to make some?

2.       Is there a way to change the Task Retrieval policies in EmBoinc? If
not what should I use in place of EmBoinc.

3.       From where can I get access to the scheduling data of actual
projects for simulation and comparison? The way you did in your paper.

I sincerely appreciate your reply.






Regards,

Muhammad Khalid Khan,
Director - College of Computing & Information Sciences,
Head - IT Department.
PAF - KIET.

On Wed, May 25, 2011 at 6:29 AM, David Anderson <[email protected]>wrote:

> Khalid:
> Please post to boinc_dev (or just to me), not to boinc_alpha.
>
>
> On 24-May-2011 3:46 AM, Khalid Khan wrote:
>
>>
>> Now I am about to propose the following topic as my Ph.D. proposal
>> "The schedulling and resource allocation policies of Boinc".
>>
>
> Can you be more specific?
>
>
>
>> 1- Is it possible to implement a new schedulling and resource
>> allocation mechanism in Boinc.
>>
>
> Sure; we do it all the time.  See:
> http://boinc.berkeley.edu/dev/sim_web.php
>
>
>  2- Will the schedulling data of old policies be availalbe to do the
>> comparative analysis.
>>
>
> The current client has 2 choices each for job scheduling and
> work fetch policy, selectable on the command line.
>
> -- David
>
_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.

Reply via email to