Tuberculosis project

[Michael H.W. Weber]

The first set of WUs dealing with our tuberculosis project is now ready for processing. These WUs should have quite heterogeneous run times of up to a maximum of 10 hrs (on an Intel i7 920 2.66 GHz Quad-Core CPU) and should be devoid of RAM or traffic problems. I hope that even those that had problems with our last set of huge CMS WUs will consider participating again since I believe this is a very interesting research project.
During my travels in India I arranged for cooperation with scientific labs that have a S3 security license which is actually the second highest security status known. So, we will grow and analyze even the virulent Mycobacterium tuberculosis strains to validate our computational results.

Michael.

[laurenu2]

Michael The new set of WU are still asking for more ram then I have
cmsearch needs 3814.70 MB of Ram

[Michael H.W. Weber]

Michael The new set of WU are still asking for more ram then I have
cmsearch needs 3814.70 MB of Ram

Well, I actually doubt that, but it is possible that some WUs from the previous set are still around, so could you please check and post here which WU exactly it is that consumes the RAM and how much it does consume at maximum?

Michael.

[MonsieurPoot]

Currently it looks like for the past 12 hours, only a handful of WUs have been distributed.
I assume that most tuberculosis-WUs have now been sent out and the remaining 1000-something results are mostly from older batches still waiting for specific hosts to request them - and as the number of results still is above the threshold, no new small tuberculosis-WUs are created and queued.

- Poot

[Michael H.W. Weber]

Currently it looks like for the past 12 hours, only a handful of WUs have been distributed.
I assume that most tuberculosis-WUs have now been sent out and the remaining 1000-something results are mostly from older batches still waiting for specific hosts to request them - and as the number of results still is above the threshold, no new small tuberculosis-WUs are created and queued.

Well, this requires some corrections: First of all, thousands of Mtb WUs have been delivered. Secondly, we have hundreds of thousands of additional Mtb WUs wanting for distribution in the server pipeline and I expect not to run out of work any time soon. Hence, it should be safe to dedicate a set of machines to this project now. Any (unexpected) issues that might arise would be published as soon as possible on the NEWS page.
And yes, it is true that, for certain machines, some WUs of the old batch are still being delivered. However, we have now setup a mechanism that will prevent delivery of WUs if they are returned in a non-validatable state for a certain number of times. This is to prevent keeping machines occupied with something that obviously need some manual in-depth inspection by our team. So, as you can see we try to continuously improve this sometimes somewhat demanding project to make its use as comfortable for everybody as possible.

Michael.

[MonsieurPoot]

You must have misunderstood me.
I understand that thousands of Mtb-WUs have been delivered to host computers since nine yesterday morning - but that does not change the fact, that the number of unsent WUs came to a sudden halt 14 hours ago.
Before that the delivery worked out fine, but nothing is happening right now nor has been over night:
http://www.rnaworld.de/rnaworld/server_status.php



And as this number of unsent WUs stayed above the aforementioned threshold of 1000, no new batch of Mtb has been put in the queue and I just figured THIS to be the reason, why laurenu2 is not getting any work for his "low-RAM" machine. I personally have not received a single WU either on my Win2K-computer since yesterday because all that is left in the queue needs more RAM or a different machine/OS.
Hundreds of thousands of WUs may be waiting in the pipeline - but not in the servers queue...

- Poot

[Edit] At around 18:30 the number of unsent WUs fell below 1000 and another batch of being queued, but you should probably still check out if there is a way to avoid a dryspell like this the next time around.

[MonsieurPoot]

Hey there...

It seems like we've hit "that spot" again. For two hours, my 2GB machine has not been getting work and the number of undelivered WUs is hovering above the 1000-threshold.
The 4GB machine only receives a single WU every once in a while that at the time of being sent out to me has been aborted by other users and is thus being resent.



@yoyo: Would it not be enough for a start to let the server create a new batch of WUs at a higher threshold, like 1500 remaining WUs, so there is no dryspell in the queue?

- Poot

[yoyo]

For me it is more important to find the root cause why the wus are not send. As I saw in the log the feeder doesn't fetch them into its shmem. I must clarify this with the boinc developers what could be the reason for it.
I have the impressions, that we do have to less Linux64 systems with 4G free RAM.
yoyo

[MonsieurPoot]

My idea was just supposed to be a workaround but of course you are right, it is no solution and could make everything even worse.
Anyway, good luck finding the cause of the problem!

- Poot

[mxplm]

I have the impressions, that we do have to less Linux64 systems with 4G free RAM.
yoyo

I wanted to upgrade anyway, but since there is a RNA-WU running for the next 85h, I can't swap the RAM for the next few days :lol:

[Michael H.W. Weber]

I have the impressions, that we do have to less Linux64 systems with 4G free RAM.

None of the Mtb WUs in the pipeline should require 4 GB of RAM. I would actually be surprised if there was any that would even require 1 GB although we/I do not have any proof for that as we cannot forecast RAM usage. The genomes are only approx. 5 MB in size - not up to 500 MB as with the previous batch. The CMC files used to search these genomes cover the ENTIRE range of complexity that can be thought of, though. Still, RAM usage with CMSEARCH should mainly depend on the genome size, hence the requested RAM requirements to allow a system to process these WUs should be decreased as soon as possible to broaden the number of machines that can participate.

Michael.

[MonsieurPoot]

The new Mtc-WUs definitely barely need any RAM, but I think yoyo confirmed there are hundreds of old and RAM-hungry WUs still in the queue that simply do not make it out to host-systems for unknown reasons, which is why there are still more than 1000 WUs waiting on the server, most of which are not Mtc.

- Poot