Overclock.net › Forums › Overclockers Care › Overclock.net Folding@Home Team › Folding Pros care to explain Work Units to me?
New Posts  All Forums:Forum Nav:

Folding Pros care to explain Work Units to me?

post #1 of 7
Thread Starter 
Hey Guys,

So I have read about work units and I get the overall concept of them but I am definitely a bit confused about the different ones and why they are given out the way they are...

I've read about how the credit value is calculated and how the deadline is calculated, but I haven't found a good explanation why certain work units are given out compared to others.

What is the difference between a 8031 and an 8032 for example? GPU, sure. Points, sure. But what is the actual difference in terms of what work is being done? Is it testing a different disease? There seems to be dozens of different Work Units but I have no idea what the differences between all of them are besides their value.

Then I was thinking, why couldn't bigger point/value WUs be assigned to a possibly undesirable system? If it gets done before the deadline why does it matter what it's being calculated on? This one I solely ask because I am generally doing 20 WUs a day. Granted it's not all on one processing unit, but still. I would think if a GPU could spit out 30+ WUs in 10 days, why not just give it one bigger one that it could complete in the 10 day deadline? What is the value of the small WUs to Stanford in comparison to the big ones? I would think they would want to give out more of the bigger WUs.

I guess my real question on that last one is more along the lines of why do certain work units get assigned to certain processing units?



lol. I guess I am just interested in what WUs really mean and how they are distributed to machines.

Thanks guys. thumb.gif
 
Play Server
(10 items)
 
NAS
(14 items)
 
CPUCPUMotherboardRAM
Intel Xeon L5640 Intel Xeon L5640 Dell R710 Mobo Hynix 144GB DDR3 ECC PC3L-10600R 
Hard DrivePowerCaseOther
Intel X25-M G2 80GB Redundant 840W PSU Dell 710 2.5" - Google Search Appliance Dell Perc H700 Raid Controller 
OtherOther
Dell iDRAC6 Enterprise Mellanox Connectx-2 10Gb SFP+ NIC 
CPUCPUMotherboardRAM
Intel Xeon E5-2670 v1 Intel Xeon E5-2670 v1 SuperMicro X9DRL-iF 64GB Kingston ECC (kvr16r11d4/16HA) 
Hard DriveHard DriveHard DriveCooling
12x 4TB HGST 5K4000 SanDisk SSD 960GB (cache) 6x 8TB WD Red Scythe AP-29 
CoolingOSPowerCase
2x Noctua U9DXi4 FreeNAS Corsair AX750 Norco RPC-4224 
OtherOther
3x LSI 9211-8i (HBA) Mellanox Connectx-2 10Gb NIC 
  hide details  
Reply
 
Play Server
(10 items)
 
NAS
(14 items)
 
CPUCPUMotherboardRAM
Intel Xeon L5640 Intel Xeon L5640 Dell R710 Mobo Hynix 144GB DDR3 ECC PC3L-10600R 
Hard DrivePowerCaseOther
Intel X25-M G2 80GB Redundant 840W PSU Dell 710 2.5" - Google Search Appliance Dell Perc H700 Raid Controller 
OtherOther
Dell iDRAC6 Enterprise Mellanox Connectx-2 10Gb SFP+ NIC 
CPUCPUMotherboardRAM
Intel Xeon E5-2670 v1 Intel Xeon E5-2670 v1 SuperMicro X9DRL-iF 64GB Kingston ECC (kvr16r11d4/16HA) 
Hard DriveHard DriveHard DriveCooling
12x 4TB HGST 5K4000 SanDisk SSD 960GB (cache) 6x 8TB WD Red Scythe AP-29 
CoolingOSPowerCase
2x Noctua U9DXi4 FreeNAS Corsair AX750 Norco RPC-4224 
OtherOther
3x LSI 9211-8i (HBA) Mellanox Connectx-2 10Gb NIC 
  hide details  
Reply
post #2 of 7
Well I am not the greatest, but here is some of what I know...

First off, you were asking about the difference between the units. The difference is what kind of proteins they are folding.
An excellent link to hang on to if you are interested in knowing what you are folding is this:
http://fah-web.stanford.edu/psummaryC.html
It has all the active units listed there, and some of them have a pretty good explanation of what is being folded. Some of them are for Alzheimer's disease, some for influenza, and other things...
The description for the 8031 and 8032 state that they are larger proteins. This is possibly why they take longer to fold, I don't know... From the descriptions, both of those units look to be studying the same thing.



I don't know for sure why they don't have huge units for GPU. It is possible they just don't have any need to hand out units that big for that type of compute platform.
The Main PC
(8 items)
 
16c/32t
(9 items)
 
12c/24t
(6 items)
 
CPUMotherboardGraphicsRAM
i7 6700k ASUS Z170I PRO GAMING Mini‑ITX EVGA GTX 1080 TI FTW3 G.SKILL TridentZ F4-3200C14D-16GTZ 
Hard DriveCoolingOSPower
SAMSUNG 950 PRO M.2 Corsair H80i Windows 10 Professional x64 Corsair SF600 
CPUCPUMotherboardGraphics
Xeon E5-2670 Xeon E5-2670 Intel s2600cp2j GTX 980 
RAMHard DriveCoolingPower
Hynix 128GB PC3-12800R ECC Kingston SV300S37A Noctua NH-U9DXi4 Corsair RM850i 
Case
Rosewill RSV-L4000 
CPUCPUMotherboardGraphics
Xeon X5680 Xeon X5680 SuperMicro X8DTL-iF GTX 980 
RAMOS
24GB (6X4GB) DDR3 Memory Ubuntu 
  hide details  
Reply
The Main PC
(8 items)
 
16c/32t
(9 items)
 
12c/24t
(6 items)
 
CPUMotherboardGraphicsRAM
i7 6700k ASUS Z170I PRO GAMING Mini‑ITX EVGA GTX 1080 TI FTW3 G.SKILL TridentZ F4-3200C14D-16GTZ 
Hard DriveCoolingOSPower
SAMSUNG 950 PRO M.2 Corsair H80i Windows 10 Professional x64 Corsair SF600 
CPUCPUMotherboardGraphics
Xeon E5-2670 Xeon E5-2670 Intel s2600cp2j GTX 980 
RAMHard DriveCoolingPower
Hynix 128GB PC3-12800R ECC Kingston SV300S37A Noctua NH-U9DXi4 Corsair RM850i 
Case
Rosewill RSV-L4000 
CPUCPUMotherboardGraphics
Xeon X5680 Xeon X5680 SuperMicro X8DTL-iF GTX 980 
RAMOS
24GB (6X4GB) DDR3 Memory Ubuntu 
  hide details  
Reply
post #3 of 7
Well each work unit is a small piece of the puzzle. Folding proteins is a very processing heavy science. From my understanding one work unit may only make up a millisecond of real time protein folding. While it is hard to tell what exactly is being done for each work unit the broader problems can be found on stanfords site where they give a brief run down of what they hope to find.

As for making the WUs bigger there are several factors why it is best to keep them small. In addition the the strict science of protein folding Stanford is also doing good research in just in time computing (similar to lean manufacturing). Having one part of the puzzle solved does not help if you have to wait a while to finish the rest. This is part of the reason for deadlines i imagine. The other reason for keeping the work units small is that what happens if after 29 days of processing the unit someone has to reformat their computer, the OC turns out to be unstable and the unit is deleted, or any number of other reasons. By keeping them smaller it helps to prevent accidental loss of the information. Finally think of the space that it may take up. Uploading a bigadv can take some time depending on internet speed after 4 days of processing. Can you imagine if the WU took 30 days to complete. The upload could take hours and hours to complete.


Just a couple of my thoughts


Just read the last line about why certain processors get certain units. and the answer is we dont know/it depends. Some WUs are only able to be completed by certain cores. and only certain processors/gpus can run those cores. fermi is a15, smp is a3, bigadv is a5 i believe. So that is part of the reason. My other guess is that they have a fairly good idea about what WU is optimized for what type of environment. You only get certain ones in linux that all have a fairly consistent base point of 500. While on windows i see it vary from 300 to 3000. the other thing that comes into play is ram/number of cores. I would imagine that they have a system that has all the WUs categorized based on what type of environments they would like to send them off to
Edited by crystalhand - 3/10/12 at 6:56am
post #4 of 7
Thread Starter 
Quote:
Originally Posted by k4m1k4z3 View Post

Well I am not the greatest, but here is some of what I know...
First off, you were asking about the difference between the units. The difference is what kind of proteins they are folding.
An excellent link to hang on to if you are interested in knowing what you are folding is this:
http://fah-web.stanford.edu/psummaryC.html
It has all the active units listed there, and some of them have a pretty good explanation of what is being folded. Some of them are for Alzheimer's disease, some for influenza, and other things...
The description for the 8031 and 8032 state that they are larger proteins. This is possibly why they take longer to fold, I don't know... From the descriptions, both of those units look to be studying the same thing.
I don't know for sure why they don't have huge units for GPU. It is possible they just don't have any need to hand out units that big for that type of compute platform.

Awesome! Thank you for that link! +rep
Quote:
Originally Posted by crystalhand View Post

Well each work unit is a small piece of the puzzle. Folding proteins is a very processing heavy science. From my understanding one work unit may only make up a millisecond of real time protein folding. While it is hard to tell what exactly is being done for each work unit the broader problems can be found on stanfords site where they give a brief run down of what they hope to find.
As for making the WUs bigger there are several factors why it is best to keep them small. In addition the the strict science of protein folding Stanford is also doing good research in just in time computing (similar to lean manufacturing). Having one part of the puzzle solved does not help if you have to wait a while to finish the rest. This is part of the reason for deadlines i imagine. The other reason for keeping the work units small is that what happens if after 29 days of processing the unit someone has to reformat their computer, the OC turns out to be unstable and the unit is deleted, or any number of other reasons. By keeping them smaller it helps to prevent accidental loss of the information. Finally think of the space that it may take up. Uploading a bigadv can take some time depending on internet speed after 4 days of processing. Can you imagine if the WU took 30 days to complete. The upload could take hours and hours to complete.
Just a couple of my thoughts
Just read the last line about why certain processors get certain units. and the answer is we dont know/it depends. Some WUs are only able to be completed by certain cores. and only certain processors/gpus can run those cores. fermi is a15, smp is a3, bigadv is a5 i believe. So that is part of the reason. My other guess is that they have a fairly good idea about what WU is optimized for what type of environment. You only get certain ones in linux that all have a fairly consistent base point of 500. While on windows i see it vary from 300 to 3000. the other thing that comes into play is ram/number of cores. I would imagine that they have a system that has all the WUs categorized based on what type of environments they would like to send them off to

Interesting about the WU to real world time comparison, hmm.

Where did you get 30 days from? lol. They can keep the deadlines how they are.

I forgot all about the differences (a15/a3/a5) upsidedwnsmiley.gif haha. thanks +rep



It just seems like to me they could see that if a user is sending back so many WUs in such a fast period of time that bigger ones could be sent out that could be of greater value? Whether that is for CPUs or GPUs.

All very interesting. Going to go read that link biggrin.gif
 
Play Server
(10 items)
 
NAS
(14 items)
 
CPUCPUMotherboardRAM
Intel Xeon L5640 Intel Xeon L5640 Dell R710 Mobo Hynix 144GB DDR3 ECC PC3L-10600R 
Hard DrivePowerCaseOther
Intel X25-M G2 80GB Redundant 840W PSU Dell 710 2.5" - Google Search Appliance Dell Perc H700 Raid Controller 
OtherOther
Dell iDRAC6 Enterprise Mellanox Connectx-2 10Gb SFP+ NIC 
CPUCPUMotherboardRAM
Intel Xeon E5-2670 v1 Intel Xeon E5-2670 v1 SuperMicro X9DRL-iF 64GB Kingston ECC (kvr16r11d4/16HA) 
Hard DriveHard DriveHard DriveCooling
12x 4TB HGST 5K4000 SanDisk SSD 960GB (cache) 6x 8TB WD Red Scythe AP-29 
CoolingOSPowerCase
2x Noctua U9DXi4 FreeNAS Corsair AX750 Norco RPC-4224 
OtherOther
3x LSI 9211-8i (HBA) Mellanox Connectx-2 10Gb NIC 
  hide details  
Reply
 
Play Server
(10 items)
 
NAS
(14 items)
 
CPUCPUMotherboardRAM
Intel Xeon L5640 Intel Xeon L5640 Dell R710 Mobo Hynix 144GB DDR3 ECC PC3L-10600R 
Hard DrivePowerCaseOther
Intel X25-M G2 80GB Redundant 840W PSU Dell 710 2.5" - Google Search Appliance Dell Perc H700 Raid Controller 
OtherOther
Dell iDRAC6 Enterprise Mellanox Connectx-2 10Gb SFP+ NIC 
CPUCPUMotherboardRAM
Intel Xeon E5-2670 v1 Intel Xeon E5-2670 v1 SuperMicro X9DRL-iF 64GB Kingston ECC (kvr16r11d4/16HA) 
Hard DriveHard DriveHard DriveCooling
12x 4TB HGST 5K4000 SanDisk SSD 960GB (cache) 6x 8TB WD Red Scythe AP-29 
CoolingOSPowerCase
2x Noctua U9DXi4 FreeNAS Corsair AX750 Norco RPC-4224 
OtherOther
3x LSI 9211-8i (HBA) Mellanox Connectx-2 10Gb NIC 
  hide details  
Reply
post #5 of 7
Quote:
Originally Posted by deafboy View Post

Awesome! Thank you for that link! +rep
Interesting about the WU to real world time comparison, hmm.
Where did you get 30 days from? lol. They can keep the deadlines how they are.
I forgot all about the differences (a15/a3/a5) upsidedwnsmiley.gif haha. thanks +rep
It just seems like to me they could see that if a user is sending back so many WUs in such a fast period of time that bigger ones could be sent out that could be of greater value? Whether that is for CPUs or GPUs.
All very interesting. Going to go read that link biggrin.gif

Haha I tend to read initially posts to fast. For some reason I thought that's what you had initially suggested. I dont think it is a bad idea to give bigger work units but at the same time I would prefer not to receive them. From what I have seen with CPUs the bigger WUs tend to be bad PPD (more stanford bureaucracy than anything) earners and are more affected by multi tasking of the rig. Plus on more then one occasion I have lost 1+days of work due to a need to restart or instability in the rig. By keeping them short the loss will be lessened if the current WU is corrupted.

The introduction of bigadv work units was an attempt to do what you are suggesting. However as technology changes so does the performance of the equipment crunching the numbers. I remember a time not to long ago (3-4 years) when the ps3 was considered a huge asset to folding. Now however it is hardly work running them as the value of the contribution is outweighed by the cost of electricity in my opinion. However stanford has to release work units that can be handled by both cutting edge hardware and 3 year old hardware at the same time. Plus the balancing of completing the projects within a certain time frame. It does not help if 99% of the puzzle is solved but a crucial section wont be available for a while
post #6 of 7
Thread Starter 
Quote:
Originally Posted by crystalhand View Post

Haha I tend to read initially posts to fast. For some reason I thought that's what you had initially suggested. I dont think it is a bad idea to give bigger work units but at the same time I would prefer not to receive them. From what I have seen with CPUs the bigger WUs tend to be bad PPD (more stanford bureaucracy than anything) earners and are more affected by multi tasking of the rig. Plus on more then one occasion I have lost 1+days of work due to a need to restart or instability in the rig. By keeping them short the loss will be lessened if the current WU is corrupted.
The introduction of bigadv work units was an attempt to do what you are suggesting. However as technology changes so does the performance of the equipment crunching the numbers. I remember a time not to long ago (3-4 years) when the ps3 was considered a huge asset to folding. Now however it is hardly work running them as the value of the contribution is outweighed by the cost of electricity in my opinion. However stanford has to release work units that can be handled by both cutting edge hardware and 3 year old hardware at the same time. Plus the balancing of completing the projects within a certain time frame. It does not help if 99% of the puzzle is solved but a crucial section wont be available for a while

Yeah, that is definitely understandable. It just seems slightly odd to me to give a WU a deadline of 10+ days and in many cases a final deadline that is much further out when the WU is often completed in 3-4 hours. Granted I know everyone isn't folding 24/7 but still.

I definitely understand though doing things in smaller chunks, definitely a solid way of getting work done. There is more value in many small pieces being 100% complete than a couple large pieces being incomplete.
 
Play Server
(10 items)
 
NAS
(14 items)
 
CPUCPUMotherboardRAM
Intel Xeon L5640 Intel Xeon L5640 Dell R710 Mobo Hynix 144GB DDR3 ECC PC3L-10600R 
Hard DrivePowerCaseOther
Intel X25-M G2 80GB Redundant 840W PSU Dell 710 2.5" - Google Search Appliance Dell Perc H700 Raid Controller 
OtherOther
Dell iDRAC6 Enterprise Mellanox Connectx-2 10Gb SFP+ NIC 
CPUCPUMotherboardRAM
Intel Xeon E5-2670 v1 Intel Xeon E5-2670 v1 SuperMicro X9DRL-iF 64GB Kingston ECC (kvr16r11d4/16HA) 
Hard DriveHard DriveHard DriveCooling
12x 4TB HGST 5K4000 SanDisk SSD 960GB (cache) 6x 8TB WD Red Scythe AP-29 
CoolingOSPowerCase
2x Noctua U9DXi4 FreeNAS Corsair AX750 Norco RPC-4224 
OtherOther
3x LSI 9211-8i (HBA) Mellanox Connectx-2 10Gb NIC 
  hide details  
Reply
 
Play Server
(10 items)
 
NAS
(14 items)
 
CPUCPUMotherboardRAM
Intel Xeon L5640 Intel Xeon L5640 Dell R710 Mobo Hynix 144GB DDR3 ECC PC3L-10600R 
Hard DrivePowerCaseOther
Intel X25-M G2 80GB Redundant 840W PSU Dell 710 2.5" - Google Search Appliance Dell Perc H700 Raid Controller 
OtherOther
Dell iDRAC6 Enterprise Mellanox Connectx-2 10Gb SFP+ NIC 
CPUCPUMotherboardRAM
Intel Xeon E5-2670 v1 Intel Xeon E5-2670 v1 SuperMicro X9DRL-iF 64GB Kingston ECC (kvr16r11d4/16HA) 
Hard DriveHard DriveHard DriveCooling
12x 4TB HGST 5K4000 SanDisk SSD 960GB (cache) 6x 8TB WD Red Scythe AP-29 
CoolingOSPowerCase
2x Noctua U9DXi4 FreeNAS Corsair AX750 Norco RPC-4224 
OtherOther
3x LSI 9211-8i (HBA) Mellanox Connectx-2 10Gb NIC 
  hide details  
Reply
post #7 of 7
Ya I think thats what it comes down to. The complete problem can take many months to complete and new/old hardware will be working on it. I have a 440 and a 460 that are folding 24/7. The 440 takes 3x as long to complete each unit so with budget hardware a couple of years old that could be significantly higher (although both my cards could be considered budget). In general I have to say that Stanford is going towards larger WUs. In that past week or two they have changed the GPU work units from 1.3k points to 3.8k points with the time to complete more then doubling. So while the size of the units will definately increase over time it may not keep pace with the latest and greatest. They want to make sure that even 4 year old hardware is able to complete the work units within the deadlines. Also 24/7 folding is much different then those who choose to only fold for a couple of hours a day (more then likely while the computer is being used for other things adding to the time it takes to complete).
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Overclock.net Folding@Home Team
Overclock.net › Forums › Overclockers Care › Overclock.net Folding@Home Team › Folding Pros care to explain Work Units to me?