RNA World

[LAURENU2]

Well Michael your server did OK It took My beating and kept on ticking
And Michael Please keep to the Right or I will you on/about tomorrow

I Have only ported 90% of my power to your project Due to Nodes feeding on other projects
But that may change if you can provide Food for my Little hoard of Nodes :Cage:

[Michael H.W. Weber]

And Michael Please keep to the Right or I will you on/about tomorrow

Well, I hope you do not mind that I chose to keep to the left instead - just to avoid this one: .


Michael.

[outlnder]

Isn't the rule of the Autobahn to "Drive Right"?

[Michael H.W. Weber]

Isn't the rule of the Autobahn to "Drive Right"?

Indeed. That's why I chose left.


Michael.

[LAURENU2]

They say Left I say Right Some one is sure to be
Thank goodness there are only 2 more to go

Have only ported 90% of my power to your project Due to Nodes feeding on other projects
But that may change if you can provide Food for my Little hoard of Nodes :Cage:

Up to about 95% NOW and I see some nodes searching for food
We need more Food Michael

I see you dishing out some fast food there 1 to 10 min WU's
How is your Server doing under the high load
Is that why some of my Nodes are looking for WU's

[Michael H.W. Weber]

Well, as stated on the project main page in the NEWS section we were fiddling around a bit with the database. It seems, however, that to our surprise there is rather a bottle neck issue with the dual HD RAID system. We are on it but for the non-Linux x64 guys it might be advisable to add a second project just in case the well runs dry overnight (although there is CMS work queued up already).

Michael.

[outlnder]

Mr. Weber, even though I am currently not doing your project, I am very pleased that you are communicative and willing to do whatever you have to do to keep it up and running.

I assure you that your attitude is few and far between.

Thank you for a job well done.

[Bigred]

I agree. It is almost unheard of for a project to be informative any place except the project web site. Keep up the great work.
And make sure you enjoy your Rosenmontag.

[Michael H.W. Weber]

Thanks guys for the nice encouragement. Even on Rosenmontag (how do you know?) we keep an eagle eye on our project as you can see.

Michael.

[Bigred]

(how do you know?)
Michael.

Take a look at my location. I'm at work in Mainz-Kastel and I can hear the parade across the river in Mainz.

[Michael H.W. Weber]

Take a look at my location. I'm at work in Mainz-Kastel and I can hear the parade across the river in Mainz.

Ahh, that is indeed an explanation.

Back to RNA World now: As soon as the current CMS WUs are over, we most likely need to wait until completion of the current CMC WUs before I can put more CMS WUs online. I will put this information as well on the main page notice board, soon.

Michael.

[Angus]

It looks like there about 7200 CMC WUs queued up (un-sent).

Since those can only be crunched on Linux 64 bit machines, and they each take 1.5 hours or so (from your server status graphs), it looks like it might be a couple of weeks before those all get crunched, since there don't seem to be very many of the required machines.

Is that correct?

[Michael H.W. Weber]

Yeah. So the best way to solve the issue would be to hook up more Linux x64 machines, right?
I could of course produce more CMS WUs by running CMC locally as I did before on my 955 BE. We will see how things develop.
But we need people to understand, that in the present development stage of the project, it is of importance to make sure our system really runs well. We are still in testing phase. Please never forget that. You saw how only recently our quite powerful new server was hit by the tremendous requests. Or the issue with the RAID system. Good things sometimes take time to optimize for the many unexpected small issues that surely arise with such a project and for that we just request a little patience.

Michael.

[edit]: By the way, the server currently helps crunching away these CMC WUs for it is a Linux x64 machine, too.

[gopher_yarrowzoo]

Micheal - just a thought you what FS you using on the Raid Ext2 or 3 if it's 3 you WILL see bottlenecking. We had that here wonder why rebuilt the server almost until we say EXT3 switched to EXT2 no journalling runs super fast

[Michael H.W. Weber]

Well, using the CMS WU shortage period for server analysis and tweaking within a day we have already modified a number of things and we think it works well now. For example we run the WU generation as RAM disk avoiding HD load which dramatically speeds up things and resolves the HD bottleneck issue. Concerning the file system, I actually assumed it must be ext2 but I will make sure by talking to our sysadmin. Anyway, thanks for the hint.

Michael.

[LAURENU2]

So how long do you think we (windows users) will be out of work
Just remember anything linux can do we can do better and more of it oke:

[Michael H.W. Weber]

It shouldn't take much longer. Approx. two thirds of the CMC WUs are already completed. Thereafter, a Windows session will start.

Michael.

[Bok]

Hi Michael,

I have some cmc units on a linux 64bit which are at 90hrs (82%) and 50hrs (43%) is this ok? Most others on the same machine have ran in a few hours at most..

Bok

[Michael H.W. Weber]

Hi Michael,

I have some cmc units on a linux 64bit which are at 90hrs (82%) and 50hrs (43%) is this ok? Most others on the same machine have ran in a few hours at most..

Yes, that is completely normal. In fact, these we call "in the monster WU range". I have one now at 90 hrs with 35% done on my AMD 955 BE Quad-Core. That one is the second largest we have in the pipeline. But often, especially these long WUs suddenly finish, so their total run time indication more frequently is overestimated compared to the small CMC WUs. We also do not fully understand why this happens. So we just have to accept it as a fact at present.

Michael.

[Michael H.W. Weber]

Micheal - just a thought you what FS you using on the Raid Ext2 or 3 if it's 3 you WILL see bottlenecking. We had that here wonder why rebuilt the server almost until we say EXT3 switched to EXT2 no journalling runs super fast

Could you please specify a bit more in detail what exactly the problems were with EXT3? We figured that our server has EXT3 as well (to my surprise).

Michael.

[Michael H.W. Weber]

Please do not be surprised if the RNA World server is unreachable for a couple of hours. We have to run a thorough HD hardware check to exclude damage with the file system (don't worry - everything was backed up in advance and is backed up anyway regularly).
I will also release additional CMS WUs soon (this week). But details will be found on the main page as usual in the NEWS section.

Michael.

[gopher_yarrowzoo]

It to do with the Journals it does any FS transaction aka file read/write gets added in,
Bok will verify this one, we set up a Super fast new server and after very little time it slowed from 6Mbit transfer rates to Sub 1.5Mbit we pulled it apart re did it still the same redid the chipset drivers the works - see if you can't turn the Journalling off in EXT3..

[Michael H.W. Weber]

New CMS work ahead some time this weekend. Concerning the EXT3 journalling issue thanks for the information. We will look into it.

Michael.

[LAURENU2]

New CMS work ahead some time this weekend..
Michael.

But but I want it NOW .

[Michael H.W. Weber]

I will release around 300.000 CMS WUs, just for a start.

Please stick to preferentially completing the Linux x64 CMC WUs by deactivating CMS if you have Linux x64 machines that still receive CMC WUs (some don't due to our very strict HR settings, so these can work on the upcoming CMS instead if you like) otherwise we will have a WU shortage again later on. The shorter deadline of CMS will make BOINC engage preferentially the new CMS WUs. That's why I ask to transiently deactivate CMS on the relevant boxes.

Michael.

[LAURENU2]

I will release around 300.000 CMS WUs, just for a start.
Michael.

That will be a good for a breakfast
But what about Dinner

[Bigred]

That will be a good for a breakfast
But what about Dinner

There might be a few left for an afternoon snack.

[LAURENU2]

There might be a few left for an afternoon snack.

NOT IF I CAN HELP IT
First come first served My train is fueling up

[LAURENU2]

I think I was to greedy I melted one of My DSL Modems with all the short WU's
Only have 1 DSL line working now so My totals will be low today
Off to FRY's I go to get a New DSL Modem

[yoyo]

I think I was to greedy I melted one of My DSL Modems with all the short WU's
Only have 1 DSL line working now so My totals will be low today
Off to FRY's I go to get a New DSL Modem

So we are not only testing our server
It was also a testt for your modem

[Angus]

I'm getting consistent "no work available" messages again even though the server status page insists there are about 17000 WUs available.

They can't *ALL* be non-Windows tasks, can they?

[LAURENU2]

So we are not only testing our server
It was also a testt for your modem

Yes I guess so Life is one big test
I'm back on line now and see
2/21/2010 4:18:41 PM RNA World Message from server: cmcalibrate is not available for your type of computer.
2/21/2010 4:18:41 PM RNA World Message from server: (there was work but it was committed to other platforms)
On some of my Nodes

[gopher_yarrowzoo]

I think I was to greedy I melted one of My DSL Modems with all the short WU's
Only have 1 DSL line working now so My totals will be low today
Off to FRY's I go to get a New DSL Modem

You sure it wasn't maybe a line spike ....
Oh and maybe need to invest in a cooling tray for your routers :P

[LAURENU2]

You sure it wasn't maybe a line spike ....
Oh and maybe need to invest in a cooling tray for your routers :P

No line spike here home is filtered 3 times and nothing else blew out
All 4 of the gigabit routers are cool but DSL Modems does get warm under load

I push things to limit here at home 24/7 things burn out all the time.
I have even been know to take down projects I work on

[gopher_yarrowzoo]

No line spike here home is filtered 3 times and nothing else blew out
All 4 of the gigabit routers are cool but DSL Modems does get warm under load

I push things to limit here at home 24/7 things burn out all the time.
I have even been know to take down projects I work on

The routers will be cool as they have fans in them - bet the DSL's don't I know mine don't...
You do take project out sometimes either by server overload, bandwidth saturation or you just plain suck it dry.

[Michael H.W. Weber]

Well, I think we will need to install a filter on the server side that, based on the run time estimates, prevents such small WUs from being sent out to clients in the future. Although we have more than 200,000 WUs left, little is computed due to the extremely heavy MySQL load that is caused by the massive connections. So, another lesson learned, I would say. These small WUs in fact can easily run on some of the (mostly idle) server cores and will also finish within a few days.

Michael.

[LAURENU2]

Might last a few days if we could get them out of your server Michael
All I seem to get is like 1 WU maybe every 10 Min And thats for a quad
Why make the nods Wait 10 min to return for a WU that runs for a Min or less
Other projects alow WU's to be sent when there done and not wait the 10 min
like RNA does

[zombie67]

Might last a few days if we could get them out of your server Michael
All I seem to get is like 1 WU maybe every 10 Min And thats for a quad
Why make the nods Wait 10 min to return for a WU that runs for a Min or less
Other projects alow WU's to be sent when there done and not wait the 10 min
like RNA does

The server is already overwhelmed. Allowing a shorter back-off time would only make that worse.

But by not sending out the very short tasks at all, that should really improve the load on the server. And that should allow for shorter back-off times.

[yoyo]

Correct! I increased the time between requests to 20 minutes

[LAURENU2]

Correct! I increased the time between requests to 20 minutes

I guess I will pull off this project then 20 min is a True waste of my cycles
That should help with your sever load