Advanced search

Message boards : Announcements : Ben's back

Author Message
Profile Benjamin Wandelt
Volunteer moderator
Project administrator
Project scientist
Avatar
Send message
Joined: 24 Jun 07
Posts: 187
Credit: 15,273
RAC: 0
Message 8617 - Posted: 7 Oct 2009, 7:33:57 UTC

Hi everyone,

I wanted to apologize for being out of touch for a while. in the end of July I had a bad accident which left me without the use of my hands/arms for quite a while. Then, 3 days later, our son Elyan was born. So the last couple of months have been special in more ways than one!

I'm back now, able to write and communicate again. So you'll hear more from me again, especially in the "Astronomy and Cosmology" section of this message board.

All the best,

Ben

____________
Creator of Cosmology@Home

Brian Silvers
Send message
Joined: 11 Dec 07
Posts: 420
Credit: 270,580
RAC: 0
Message 8619 - Posted: 7 Oct 2009, 10:36:34 UTC - in response to Message 8617.

Hi everyone,

I wanted to apologize for being out of touch for a while. in the end of July I had a bad accident which left me without the use of my hands/arms for quite a while. Then, 3 days later, our son Elyan was born. So the last couple of months have been special in more ways than one!

I'm back now, able to write and communicate again. So you'll hear more from me again, especially in the "Astronomy and Cosmology" section of this message board.

All the best,

Ben


First, I'm sorry to hear this. I have been in a car crash where I probably should've been killed, but wasn't. I've also had open heart surgery due to something that would've killed me if I had waited another 1-4 hours before going to the ER. Can't relate on the kid, though I guess that is something good to offset the problems...

Second, I've been very hard on you and your project. Had someone at least mentioned that there were personal issues going on, the total silence would've been understood. It hasn't been only you that hasn't been around, but neither has Anshul or Scott, if Scott is still with you at all...

Some things that I see that need addressing right away are the errors when downloading a task. That's being caused by the parameter file not being kept on the download server after the first person downloads the task. Due to the very large memory requirements of your application and the very long runtimes, it seems that many people simply issue a Reset Project from their BOINC client and your server is never told that those tasks have been abandoned. Thus those of us who remain continually run into issues downloading work.

As I mentioned, the memory requirements are very large for participating here. I know you said earlier this year that you were seeing good results. I urge you though to consider doing something to cut the memory consumption in half. The current memory requirements make it to where anyone with less than 1GB of memory either has a very difficult time participating in your project, or they cannot participate at all due to the task taking too much memory. There have also been reports from people with 4-8GB of memory that say their system is very sluggish when processing tasks.

There are two ideas I have for the memory usage issue:

- Increase the frequency of checkpointing. Currently the application has what appears to be a very long checkpoint interval. I don't know if this is the case, but I'm guessing that if you checkpoint more frequently, the application will not need to have so much held in memory.

- Develop a CUDA or OpenCL applcation for this work.

I think developing a CUDA / OpenCL, or even Brook+/CAL application would help things out tremendously. As it stands right now, there is a lot of excess computing power available at MilkyWay@Home. Their tasks, which are easily run on systems as old as Pentium II, are processed in 30-60 seconds on the current generation ATI graphics cards. Due to the much higher complexity of the work here, I would think this would be a great project to convert completely to a GPU project.

At a minimum, I would suggest looking into the checkpointing and see if you can get the application to checkpoint more frequently.

Just some thoughts...

-Brian
____________

Rapture
Avatar
Send message
Joined: 27 Oct 07
Posts: 85
Credit: 643,762
RAC: 108
Message 8621 - Posted: 7 Oct 2009, 20:42:02 UTC - in response to Message 8617.



I wanted to apologize for being out of touch for a while. in the end of July I had a bad accident which left me without the use of my hands/arms for quite a while. Then, 3 days later, our son Elyan was born. So the last couple of months have been special in more ways than one!

I'm back now, able to write and communicate again. So you'll hear more from me again, especially in the "Astronomy and Cosmology" section of this message board.



Welcome back, Ben! I am glad to hear that you have made a great recovery. It is great to see you here again. I look forward to hearing you more often soon on this board.

Bill

.clair.
Send message
Joined: 4 Nov 07
Posts: 591
Credit: 3,824,712
RAC: 5,033
Message 8622 - Posted: 7 Oct 2009, 21:50:24 UTC

It is distressing to here of your accident Ben,
I am pleased that you are recovering and mobile.
congratulations on your son, is he your first, or an addition to the clan.

Best regards

clive

Profile Benjamin Wandelt
Volunteer moderator
Project administrator
Project scientist
Avatar
Send message
Joined: 24 Jun 07
Posts: 187
Credit: 15,273
RAC: 0
Message 8624 - Posted: 8 Oct 2009, 5:51:54 UTC - in response to Message 8622.

Thanks -

He's my first!

Ben
____________
Creator of Cosmology@Home

Profile MSE29
Send message
Joined: 3 Jul 07
Posts: 30
Credit: 1,578,663
RAC: 0
Message 8627 - Posted: 9 Oct 2009, 14:28:19 UTC

Congratulations on your birth son!

I'm glad you're back.

Emanuel
Send message
Joined: 28 Oct 07
Posts: 31
Credit: 316,100
RAC: 0
Message 8640 - Posted: 19 Oct 2009, 15:39:33 UTC - in response to Message 8619.
Last modified: 19 Oct 2009, 15:42:09 UTC

- Increase the frequency of checkpointing.
- Develop a CUDA or OpenCL applcation for this work.

The memory requirement is indeed one of the greatest issues for this project. There must be a way to break up the WUs into smaller chunks - for instance, have one WU do calculations for one half of the model, a second WU do calculations for the other half, and a third WU combine them.

That way you could also safely raise the complexity of the model - just break it up into smaller chunks. Being able to do something like this is very important for any distributed computing project, otherwise you will always be limited by the capabilities of a limited amount of top-end home computers rather than simply the amount of users willing to participate. Right now you're forcing a lot of people to stop participating, as they simply cannot use their computers while the calculations are going on.

Message boards : Announcements : Ben's back