Hey,

I've recently started running Rosetta@Home on my home computers (and started installing them on the ~50 low-end PCs I administrate at work)

I was just wondering if there's something I'm missing with the settings. While the CPU usage is 100% my home PC only averages to about 220 RAC - and it's a 2x Quad-Core AMD Opteron(tm) Processor 2356 that runs about 10-12 hours a day. I've looked through the stats and I've seen Core Duos doing at least double that! I don't care much about the credits, but it would be a shame if I could be processing more WUs if I tweaked some settings?

RAM usage is about 8x300mb, which isn't a problem with the 8GB I have installed.

Any help would be greatly appreciated!

Corhal / Julian


PS: I'm sorry if this issue has been addressed in another thread, I've been reading through the FAQs and looking through the forums, but I couldn't find a similiar thread to it.

Welcome aboard. RAC will take some time to give you the "real number".

For instance, if I stopped ALL my PCs, the RAC would take probably weeks or months for it to actually get to 0.
My dual core machine that has been running for years now, probably has a higher RAC than your 8-core, but yours produces more credits per day. This is because your current RAC of that machine is not accurate because it's new.

Hopefully I didn't confuse you.

---

Oh, I see! I thought RAC was just the credits that I've gotten in the past few days :) There was a link that was supposed to explain how the RAC is calculated, but it didn't load for me.

Thanks for the clarification, glad it's not actually processing fewer WUs than it's supposed to :)

Yeah, RAC is probably a 30-day average, though I'm not certain so don't take my word for it..

In any case, welcome on board, thanks for bringing so many CPUs with you :)

---

As with many things credit related, there is an entire process to it. Here is a good link describing it. The key point is buried down near the bottom...

The simplest way to picture it is that Total Credit is like your odometer, and Recent Average Credit is like your speedometer.

...but your speedometer doesn't take a period of about 2 weeks to really reflect the speed of your car, so that's not a perfect analogy, esp. when a machine first starts. But over the longer term, RAC is a gauge of your machine's speed relative to other machines. The problem is you still can't compare machines, because you don't know if the other person's is running as many hours per day, is overclocked, uses all CPUs, uses 100% of CPUs, is perhaps limited by how much memory BOINC is allowed to use, etc. etc.

---

Rosetta Moderator: Mod.Sense

Thanks for the link! Good to know how the RAC is calculated exactly :)

I've also been reading more threads on the forums and it seems that Intel CPUs have the advantage on AMD CPUs in Rosetta@home, and I've been mostly comparing with those. And I realize it's not entirely possible to compare anyway, due to the reasons you stated.


Rosetta is definitely the most interesting distributed computing project I've taken part in so far :) I like the way David Baker keeps us informed through his journal and all that extra information on the site you can read up on. I didn't see much information on the folding@home site, beside the publications at least. I hope volunteers keep joining and the average TFLOPS keep going up!

Oh, and another thing I was wondering about. Does BOINC and Rosetta update itself or is there an option for it to do so? It wouldn't be a problem to update it on my home machines, but I wouldn't have the time at work to update the clients every few months, it's hard enough to find the time to install BOINC on them as it is :)

edit: Ah, it seems Rosetta gets its application updates the same way it gets the WUs - does BOINC auto update as well?

Oh, and another thing I was wondering about. Does BOINC and Rosetta update itself or is there an option for it to do so? It wouldn't be a problem to update it on my home machines, but I wouldn't have the time at work to update the clients every few months, it's hard enough to find the time to install BOINC on them as it is :)

edit: Ah, it seems Rosetta gets its application updates the same way it gets the WUs - does BOINC auto update as well?

Everything auto-updates itself. EXCEPT for BOINC itself. But I have a machine in which I installed BOINC on about 2-3 years ago, and it still works. So there's almost no need to update BOINC to run rosetta properly.

---

I concur with Chilean. So long as the version of BOINC you use is good for the project you intend to run (I mean, for example, if you needed a GPU BOINC version for some other project, then that would be the minimum BOINC version you'd need) then you really don't have to keep current on BOINC itself, unless there is some specific problem that gets fixed and pertains to your situation. If you note in the system requirements page, Rosetta requires at least BOINC version 4 something. And that recommendation hasn't changed in... well I don't think it has changed since Rosetta@home came out of beta over 3 years ago.

---

Rosetta Moderator: Mod.Sense

I concur with Chilean. So long as the version of BOINC you use is good for the project you intend to run (I mean, for example, if you needed a GPU BOINC version for some other project, then that would be the minimum BOINC version you'd need) then you really don't have to keep current on BOINC itself, unless there is some specific problem that gets fixed and pertains to your situation. If you note in the system requirements page, Rosetta requires at least BOINC version 4 something. And that recommendation hasn't changed in... well I don't think it has changed since Rosetta@home came out of beta over 3 years ago.

So, to be clear, yes, the Rosetta versions update automatically and download as needed. This is just the same way as it came down when you originally attached to the project. And, as Chilean said, the BOINC client does not update itself. Instead, it issues a message to the messages tab once and a while letting you know that a newer version is available. But it is just informational. If it ain't broke, don't fix it.

---

Rosetta Moderator: Mod.Sense

It would be great if you let that system run 24x7 for a few weeks to see the real RAC. My new systems usually take 2 weeks to hit the maximum RAC.

We would love to have the help of all 8 cores.

thanks for crunching!

---

Thx!

Paul

Hey,

I've recently started running Rosetta@Home on my home computers (and started installing them on the ~50 low-end PCs I administrate at work)

I was just wondering if there's something I'm missing with the settings. While the CPU usage is 100% my home PC only averages to about 220 RAC - and it's a 2x Quad-Core AMD Opteron(tm) Processor 2356 that runs about 10-12 hours a day. I've looked through the stats and I've seen Core Duos doing at least double that! I don't care much about the credits, but it would be a shame if I could be processing more WUs if I tweaked some settings?

RAM usage is about 8x300mb, which isn't a problem with the 8GB I have installed.

Any help would be greatly appreciated!

Corhal / Julian


PS: I'm sorry if this issue has been addressed in another thread, I've been reading through the FAQs and looking through the forums, but I couldn't find a similiar thread to it.

There are a couple of things you can do but they are not critical...1st is to change the setting to "Leave applications in memory while suspended?
(suspended applications will consume swap space if 'yes') yes". This is done under Your Account and Computing Preferences, then it is in the top section. The 2nd thing is go farther down on that same page and change "Write to disk at most every 60 seconds" so it is at least 60 seconds but as much as 600 seconds, or more. This relives some of the pressure on the hard drive with little chance of anything bad happening. This writes the Rosetta, in this case, stuff to the hard drive as you go, meaning that in case of a crash you have only lost whatever the number of seconds is set to. There are cases where it has been set to zero seconds and causes extreme hard drive usage and slows down the crunching.

mikey, just to clarify, Rosetta typically doesn't attempt to write very frequently to the hard drive, so typically not a problem. But I wanted to clarify that the setting you describe is not CAUSING anything to be written. What would possibly be written is a checkpoint, which preserves work done so far, while still being in the middle of a model. Then at the end of a model a checkpoint is always taken. The application is only capable of making checkpoints at specific areas in the program. The frequency of reaching one of these areas varies with different types of proteins being studied.

Based on your setting for how often to write, the application may hold one or more checkpoints to try and comply with that. So, while a setting of say 600 seconds does as you say and helps avoid the potential for run-away writes, it does potentially ADD to the runtime that is possibly lost if the machine were to power off before the next write opportunity (only by up to 10 minutes in this case, but if you set it higher, then it would be more lost).

---

Rosetta Moderator: Mod.Sense

There are a couple of things you can do but they are not critical...1st is to change the setting to "Leave applications in memory while suspended?
(suspended applications will consume swap space if 'yes') yes". This is done under Your Account and Computing Preferences, then it is in the top section. The 2nd thing is go farther down on that same page and change "Write to disk at most every 60 seconds" so it is at least 60 seconds but as much as 600 seconds, or more. This relives some of the pressure on the hard drive with little chance of anything bad happening. This writes the Rosetta, in this case, stuff to the hard drive as you go, meaning that in case of a crash you have only lost whatever the number of seconds is set to. There are cases where it has been set to zero seconds and causes extreme hard drive usage and slows down the crunching.

Lately, I've been seeing BOINC very slow at writing suspended workunits from RAM to the disk swap space, so that various non-BOINC programs can get enough RAM to run properly. This happens both for 64-bit BOINC 6.10.18 under Vista SP2, and 32-bit BOINC 6.6.36 under Vista SP2; I haven't checked under 64-bit Vista SP1. Also, I've seen no word on whether the minirosetta problem a few versions back with a memory leak has been fixed. Also, some workunits appear to be using RAM in a way that the Windows Task Manager does not report as specific to any particular to any specific task, and therefore many non-BOINC programs cannot get enough RAM to run well; this even happans eventually (after a week or two) if I tell BOINC to limit the RAM it uses to 40% of the RAM at all times, with a total of 8 GB RAM on the machines with 64-bit BOINC and 2 GB for the one with 32-bit BOINC. Has the minirosetta memory leak been fixed? Is there a way to tell BOINC to show me how much RAM both specific workunits, and the rest of BOINC in general, are using? Is there a way to tell it to move suspended workunits to the disk swap space faster? Is there a way to ask it how much swap space it has assigned to specific workunits, or how much it has assigned to specific projects but not yet to any workunit? Is there a way to get better memory handling without going back to a BOINC version so old it will not run GPU workunits?

I'm thinking of enabling Rosetta@home participation on my other two machines, but only AFTER these memory usage problems get fixed.

Thanks for all the answers! :)

It's pretty odd, my AMD seems to be getting a much lower Granted Credit for the same kind of tasks than the Intel running linux. I get about 9-15 Granted Credits for the current lr5_combined_smooth_torsion on the AMD, but about 40-50 on the Intel. The Claimed Credits are about the same! I understand how the credit system works, I'm just wondering why the granted credits are so much lower! I don't understand that at all. (Calliope and rainerf-pc for anyone checking my profile)


Another thing, doesn't really have to do with number crunching though, is that the webpage is in german for me and I don't see any way to set it to english. I'm from germany, so he probably got it from my header/IP or something, I would much prefer it to be in english though. Am I just missing the button to switch or isn't there one? :)

robertmiles, I have no way to address your question about a memory leak that Windows task manager doesn't show. In fact, since no Rosetta task runs for weeks, I don't even follow the logic of how you conclude that Rosetta has the magic power to override the operating system and BOINC core client that it runs within to result in memory consumption beyond the established limits. I mean how do you conclude that one of your OTHER applications on that machine isn't performing the same magic?

My other thought is that R@h has zero control over whether it is swapped out or not. This decision is solely that of the operating system. However, if the machine has significant idle time, and BOINC is configured to use all CPUs at 100% if it can, then the operating system is going to tend to keep it in memory, because it is the main thing running. The main thing running should be kept in RAM.

You can't have everything. If your machine gets sluggish with BOINC running, then you will have to reduce the number of CPUs it is allowed to use, or control the hours of the day it is allowed to run. From where I sit, I've got two active tasks right now, both have run for over 6 hours, and both have a peak memory usage of about 270MB, which is less then my Safari session that has been up for 2 days.

Every task is going to have different needs for memory, and large memory systems such as those you mentioned will potentially get tasks that are reserved only for large memory machines, because the tasks are known to require more memory to run well.

So, I believe what may have happened is that your machine ran happily with many tasks for many days and then landed a few with high memory requirements and the machine's performance for the rest of your applications began to suffer. This does not mean there is a memory leak, simply that those tasks require more memory then the ones you had run previously.

---

Rosetta Moderator: Mod.Sense

corhal, if you click languages on the homepage you can set a specific language. Otherwise it defaults to the browser's language setting.

---

Rosetta Moderator: Mod.Sense

mikey, just to clarify, Rosetta typically doesn't attempt to write very frequently to the hard drive, so typically not a problem. But I wanted to clarify that the setting you describe is not CAUSING anything to be written. What would possibly be written is a checkpoint, which preserves work done so far, while still being in the middle of a model. Then at the end of a model a checkpoint is always taken. The application is only capable of making checkpoints at specific areas in the program. The frequency of reaching one of these areas varies with different types of proteins being studied.

Based on your setting for how often to write, the application may hold one or more checkpoints to try and comply with that. So, while a setting of say 600 seconds does as you say and helps avoid the potential for run-away writes, it does potentially ADD to the runtime that is possibly lost if the machine were to power off before the next write opportunity (only by up to 10 minutes in this case, but if you set it higher, then it would be more lost).

You are absolutely correct, thank you for clarifying what I was trying to say but didn't!

Thanks for all the answers! :)

It's pretty odd, my AMD seems to be getting a much lower Granted Credit for the same kind of tasks than the Intel running linux. I get about 9-15 Granted Credits for the current lr5_combined_smooth_torsion on the AMD, but about 40-50 on the Intel. The Claimed Credits are about the same! I understand how the credit system works, I'm just wondering why the granted credits are so much lower! I don't understand that at all. (Calliope and rainerf-pc for anyone checking my profile)

Hi Corhal

We can't see your computer names - you have to be logged in as you for that ;)

I would expect your Opteron to be producing higher granted credits than it is though. I would expect the memory speed to make bigger than average difference on that system, but I'd still expect it to get more credits per core, per hour...

---

Hi Corhal

We can't see your computer names - you have to be logged in as you for that ;)

I would expect your Opteron to be producing higher granted credits than it is though. I would expect the memory speed to make bigger than average difference on that system, but I'd still expect it to get more credits per core, per hour...

Hey dcdc

yes, I'm very confused about that as well. It's doing the tasks pretty quickly and 8 at a time (as I set it to use 100%) - it's just getting such a low amount of Granted Credits per task! 9-15c for the same task that the Intel gets 40-50c for. As I understood the Granted Credit is a formula that is mostly based on the average of Claimed Credits from all the other machines that processed this type of task before. So why would it make a difference which kind of PC is processing it? Could it be an error in the credit system? It'd explain why so many AMD users are talking about their PCs earning a lot less credits in Rosetta@home than in other BOINC projects!

Mod.Sense: Thanks a lot! I must've overlooked that option. Much better for me to read it in english than german :)