New DDDT application version

[sierra_bound]

Just an FYI:

After a successful beta test, the Discovering Dengue Drugs - Together application has been updated to version 5.15. The application has been modified to allow us to combine multiple work units into a single longer running unit of work.

http://www.worldcommunitygrid.org/forums/wcg/viewthread?thread=18873

[sierra_bound]

A little more information.

We loaded in the rest of the 'short' workunits yesterday (Monday). We then manually adjusted the estimated duration to be what we expected and started loading those workunits. We will continue to use the manual duration estimate until later this week when the estimate will be valid.

The 5.15 DDDT application works with both the 'short' workunits and the 'long' workkunits.

The 'long' workunits should start going out later today. These will have the 12 day deadline.

This is of interest to those of you with less powerful machines.

As for those with slower machines that struggle to complete workunits on time. We are working with BOINC to put in place the ability to send shorter workunits to slower machines (slower meaning either they are less powerful or they aren't very often) and larger workunits to more powerful machines. This will meet both our need for a lower transaction rate to keep load on the database at an acceptable level but also make best use of all of the computers that are contributing.

This issue is very important to us so please do not think that we are ignoring the problems with workunit duration and deadlines.

We are updating our BOINC server code in a little bit to include the latest code from BOINC. This is the first step. BOINC has been working to implement a better mechanism for matching workunits to the different computers. Once they finish the basic mechanism, we will be enhancing it to allow for matching workunits based on duration.

http://www.worldcommunitygrid.org/forums/wcg/viewthread?thread=18867

[[XC] 4X4N]

Looks like they have the new wu's now. I have several that are listed at about 2 hours for run time.

[Martijn]

As for those with slower machines that struggle to complete workunits on time. We are working with BOINC to put in place the ability to send shorter workunits to slower machines (slower meaning either they are less powerful or they aren't very often) and larger workunits to more powerful machines. This will meet both our need for a lower transaction rate to keep load on the database at an acceptable level but also make best use of all of the computers that are contributing.

This issue is very important to us so please do not think that we are ignoring the problems with workunit duration and deadlines.

We are updating our BOINC server code in a little bit to include the latest code from BOINC. This is the first step. BOINC has been working to implement a better mechanism for matching workunits to the different computers. Once they finish the basic mechanism, we will be enhancing it to allow for matching workunits based on duration.

Now that I like :up:

[sierra_bound]

Yes, I think that's good. It might encourage more people with slower machines to crunch.

[Bobsama]

Unfortunate with many slow machines is it's very expensive to upgrade SD RAM. Last estimates would be over $100 to upgrade the P2 400 and Celeron 600 to 256MB, which is the minimum for most crunching. That $100 would allow much more research if I saved it for a new computer or a quad-core. If you have enough RAM, then it's incentive. If you need minor upgrades, not so much anymore.

[sierra_bound]

Actually not quite that expensive.:) Here are some choices.

http://www.clubit.com/product_detail.cfm?itemno=CA4351258
http://www.clubit.com/product_detail.cfm?itemno=CA4351259
http://www.newegg.com/Product/Product.aspx?Item=N82E16820161614

512MB modules cost around $25 if you're not picky about brand. Also, there are probably team members who could sell you DDR for really cheap.

[WrigleyVillain]

I have a bunch of old DDR and PC 100/133

[sierra_bound]

My machines have begun crunching the longer DDDT work units. They appear to be about four times longer than the short ones. If I get four times the credit, then I will be a happy camper.

[[XC] 4X4N]

Hopefully they will all be like this:

[sierra_bound]

I'm getting mixed results so far. The page fault count seems higher with these longer work units. Maybe that's to be expected. I probably won't know for several days how they are affecting daily production.

[sierra_bound]

Page faults per completed WU are averaging around 10 million for me. Quorum results aren't bad on quad-cores.

[WesM63]

Wait.. who needs SDRAM? I have at least 1-2gb of it laying around. PM Me.

Sierra, should I change to these units on my Quads?

[sierra_bound]

I think HCC now has fewer page faults. I need to see daily production numbers for next several days before deciding whether these longer DDDT work units are good or bad.

[[XC] 4X4N]

I think if you are going for points only, the faah seems pretty consistent, but maybe the ddt's give the most points per day. I only care a little bit about the points, just want what is fair :) Now that the cancer units seem to be fine, I only run them and ddt.

[sierra_bound]

The quorum results I've seen so far indicate the longer work units are probably good point-producers. But it's hard to tell, just by looking at half a day's results. Also, the validation lag is worse when you start a new project or switch from crunching shorter to longer work units.

My reasons for crunching DDDT have more to do with the diseases involved. As I have mentioned elsewhere, I believe Dengue will be the next big viral threat to the US. West Nile virus is already prevalent in several states including mine. And Hepatitis C is common among AIDS patients.

Earlier this year the National Institutes of Health issued a press release on the threat.

http://www3.niaid.nih.gov/news/newsreleases/2008/dengue.htm

[[XC] 4X4N]

I'm with you on reasons for running ddt. I have a friend who's wife got west nile. It took months for the doctors to figure out what was wrong with her. Now she is crippled for life.

[sierra_bound]

Yes, people have become rather paranoid about mosquitoes where I live. Last year, 18 people died from West Nile virus in California. That's actually down from four years ago (29 deaths). I give officials credit for actively attacking the problem. For the past couple of years, they have done both aerial and ground spraying in my area.

[D_A]

Has anyone heard/noticed if they've fixed that odd issue where the DDDT unit would lock up at about 50 seconds in? It was kinda rare but I seem to get them occasionally, like a few a month. That one locked unit doesn't cost much in points it's the others waiting in line behind it that don't get crunched that hurt.

[sierra_bound]

The only thing unusual I've noticed is that when the completion time hits zero, another 10-15 seconds of CPU time ticks off before the work unit status changes to "upload".

[D_A]

I've only ever seen it on my AMD machines (sorry, I got no quads). My Intels have never done it.

[[XC] Jaco]

I've had errors on 4 DDDT workunits this morning.
When you check the quorum results (pic2), you'll see it's most probably a bad WU because others have had errors too.
Sierra , you must have had errored WU's? since you do DDDT exclusively.

[sierra_bound]

Got one error on a DDDT unit on my Clovertown rig. Three and a half hours of crunching wasted on that one.

There are a couple of threads on problems with the longer units.

http://www.worldcommunitygrid.org/forums/wcg/viewthread?thread=18973
http://www.worldcommunitygrid.org/forums/wcg/viewthread?thread=18979

[sierra_bound]

I decided to download some HCC units for comparison purposes. Big difference in page faults between DDDT and HCC. Memory usage is also double for DDDT.

[[XC] Jaco]

Got one error on a DDDT unit on my Clovertown rig. Three and a half hours of crunching wasted on that one.

There are a couple of threads on problems with the longer units.

http://www.worldcommunitygrid.org/forums/wcg/viewthread?thread=18973
http://www.worldcommunitygrid.org/forums/wcg/viewthread?thread=18979

note the replies of Didactylos :shakes:

[sierra_bound]

Jaco;2827879']note the replies of Didactylos :shakes:
He is one of the reasons why I almost never post there.

[[XC] 4X4N]

Just checked and it looks like I have errors on all 3 rigs. Also, page faults and vm is way high. Might have to think about switching off of these for a few days until these issues are worked out.

[sierra_bound]

I have had only one error, something about an "unhandled exception". Unfortunately it happened as a WU was finishing. Yours look like the errors were at the beginning, so at least you didn't have much wasted CPU time.

[sierra_bound]

Quorum results have been mixed on my Clovertown rig. Sometimes good, sometimes not so good. The first result is HCC. Page faults on the HCC unit were around 250,000 compared to 10-12 million for DDDT.