- Joined
- Mar 2, 2004
- Location
- Irvine, CA
Straight of the EMIII website.
-
Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!
The never ending HT debate
- Thread starter TalRW
- Start date
Straight of the EMIII website.
OP
- Joined
- Mar 2, 2004
- Location
- Irvine, CA
- Thread Starter
- #2
Oops sorry admins please close/delete one of these threads double posted cause it lagged cause server just came back up and I thought it didn't post.
- Joined
- Feb 23, 2004
- Location
- Charlotte, NC
sigh, points or proteins, points or proteins?
- Joined
- Nov 27, 2002
- Location
- new jersey
I don't see how my cpu handing something in 12 hours later makes a difference, chances are they're sleeping.......
- Joined
- Dec 28, 2003
- Location
- Kansas City area
Sorry, but I'm running two instances on my HT and NON-HT units period! I think if we were forced only to run one instance a lot of us would say s***w this! I'm all about the points, if it helps benefit Stanford then so be it, if it helps find a cure so be it, but I built my farms for one purpose only and that is too amass as many points, points and more points as possible.To me this is a competition where the actions of the pc's benefit ME in points and Stanford in data.
Hardcore! Hooah!
paps
Hardcore! Hooah!
paps
- Joined
- Sep 7, 2004
- Location
- Lancaster, PA
The soul purpose of F@H is to help mideical research. The points really mean jack in all honesty. It's another male inhibited race to see who has the biggest balls via their PC/PC's. In the case of points it's all about the bragging rights which in the end means nothing more then bragging which also means more controversy, debates, jealousy, and in the end fighting and flaming.
As for the time issue it's not just human eyes looking at the data. A good portion of it is other computers on the other side examining the results of your computers work. They use a distributive client only for the reason they dont have the funding or time to set up millions of PC's for number crunching they could otherwise have with a distributive client. The client has saved them years of research and wasted time. So handing in the data as quick as possible is relivant.
This is just my two cents.
In the end it's just the same arguement. If you want bragging points to help your ego then by all means run more instances of the client. If you want to help the research then by all means only run one instance. In the end the only real factor being persuaded is speed at which the cure for a disease is being produced. If you ever had a close relative or friend die of cancer or the such your ego may be a little more humbled about the situation.
Keep in mind it's because of such distributive clients that the human genome system is already mapped and is done 10 years ahead of projected schedule.
As for the time issue it's not just human eyes looking at the data. A good portion of it is other computers on the other side examining the results of your computers work. They use a distributive client only for the reason they dont have the funding or time to set up millions of PC's for number crunching they could otherwise have with a distributive client. The client has saved them years of research and wasted time. So handing in the data as quick as possible is relivant.
This is just my two cents.
In the end it's just the same arguement. If you want bragging points to help your ego then by all means run more instances of the client. If you want to help the research then by all means only run one instance. In the end the only real factor being persuaded is speed at which the cure for a disease is being produced. If you ever had a close relative or friend die of cancer or the such your ego may be a little more humbled about the situation.
Keep in mind it's because of such distributive clients that the human genome system is already mapped and is done 10 years ahead of projected schedule.
- Joined
- Apr 22, 2004
- Location
- New Hampshire
wait a second! if running 2 gives you more points... that means you finish more wu's. how is this not better for "science"?
- Joined
- Sep 7, 2004
- Location
- Lancaster, PA
veryhumid said:
It doesnt turn in the work as quickly as it should be able to. By running two instances you take longer to complete the work unit. The work unit is handed out in specific fashions to try and keep them organized and recieved in the order they would like them in. By handing them in later then expected you misorganize their research and slow it down. Also by handing in two work units almost simultanoueslyyou back up work on their end again slowing it down. By doing one work unit at a time it keeps everything flowing evenly in a timely manner and keeps it organized.
- Joined
- Apr 30, 2003
- Location
- Chicago
MLMIB said:
Exactly!
What I think that Stanford is trying to say is that on some of the slower PCs with less RAM but have Hyper Threading...(Like those 2.4 Ghz Pentium 4s) it is alot longer before work gets turned in. I have had a few instances on my 3.06 Ghz HT where it has taken each over a day to turn in 2 of those large work units. Of course it was a large work unit and they always take longer. Running the second client only delayed the whole process by a dozen hours. I can see this being a larger problem on slower PCs with less RAM.
- Joined
- Apr 22, 2004
- Location
- New Hampshire
IrQ said:
ahhh, thank you for clarifying that. But it's not like an HT computer is going to be struggling for deadlines. 12 hours later is probably a good estimate
- Joined
- Nov 1, 2001
- Location
- New Iberia, LA
TommyHolly said:
I think Vijay is full of poop on this subject anyways; otherwise they would shorten the time deadline on all their wu's if they have to get the results that fast. If he's so darn worried that HT and 2 clients will slow his results down, then why doesn't he worry about the sorry way his assignment server logic assigns giant wu's to slower machines. I don't know how many times I've seen people post about starting folding with a 700-1000 MHz machine and right off the bat that poor machine draws a 200-300 point wu that will take their machine a month or more of constant crunching to finish. That's pretty damn frustrating right there to a new team member and doesn't keep them motivated to keep folding at all, to see any results for a month in the way of points under their username. I was working on my younger brother's machine yesterday, which is an old Dell P3 866 machine and after I reinstalled Win98 on it(yuck
) I installed the graphical client on it and the very first wu it drew is a 220+ point gromacs wu, which looks like it will take 100+ hours of constant crunching to finish and most probably a lot longer than that since he will also use this machine for his normal computer needs. When Stanford decides to give us a little more control over what kind of wu a particular machine can work on, then I might consider only running 1 client on my HT machines.
- Joined
- Feb 20, 2001
Firstly, Dr Pande is not "full of poop".
The only reason this project is running is because someone, at the end of the day, is footing the bill. And chances are, the only reason they continue to foot the bill is because they want to see tangible results.
The way SETI and FaH work is different. SETI forms workunits from data collected and then sends those work units to computers to look for certain trends and so forth. Each work unit is independant of one another. In FaH, new work is formed from previous work. If a project is showing promising progress then they will further explore this area by releasing a new project. It is in the interests of the Science that Workunits have deadlines, and are reqiuired as soon as possible. This keeps the project moving at a good pace and ensures that progress is being made and thus the money behind the project stays interested.
I agree however, the best thing would be to make the cores SMP aware which would essentially put an end to the debate.
The only reason this project is running is because someone, at the end of the day, is footing the bill. And chances are, the only reason they continue to foot the bill is because they want to see tangible results.
The way SETI and FaH work is different. SETI forms workunits from data collected and then sends those work units to computers to look for certain trends and so forth. Each work unit is independant of one another. In FaH, new work is formed from previous work. If a project is showing promising progress then they will further explore this area by releasing a new project. It is in the interests of the Science that Workunits have deadlines, and are reqiuired as soon as possible. This keeps the project moving at a good pace and ensures that progress is being made and thus the money behind the project stays interested.
I agree however, the best thing would be to make the cores SMP aware which would essentially put an end to the debate.
- Joined
- Oct 18, 2003
I just cannot see how it is that important to run one instance on a HT machine to get the WU done faster. I mean, with 2 processes on an HT processor, you get more WU's done over the same period of time, so I think he does not have a valid argument. Maybe he is saying to just run one instance because with two, people can notice a slowdown on their computer so they might remove it.
- Joined
- Feb 20, 2001
Quailane said:
Running two clients on a HTT enabled CPU will mean each unit is handed in later than it would have been had it been run on its own on the same CPU. You are right, more work is done in the same time period, but each work unit is returned later than it otherwise would have been. Thus it is better for the science that WUs are returned as quickly as possible.
- Joined
- Oct 12, 2001
- Location
- Los Angeles
First, let's agree people fold for points and for science. Some put a higher priority on the first, some the latter, we should not judge as to people's reasons why they fold. In my book, if you fold you are good with me Anyway, I think what Vijay is trying to say is that it is better for the project to return the WU's ASAP, with 2 instances on HT this is not the case. No it is not slow, but it is not as fast as running one unit. I mostly run 2 instances on HT machines, not so much for the point benefit but because too many times machines run out of work... or can't connect or... anyway, I am with the camp that thinks it really doesnt make that much difference if a unit gets turned in 8 hours later.
Anyway, I think what Vijay is trying to say is that it is better for the project to return the WU's ASAP, with 2 instances on HT this is not the case. No it is not slow, but it is not as fast as running one unit. I mostly run 2 instances on HT machines, not so much for the point benefit but because too many times machines run out of work... or can't connect or... anyway, I am with the camp that thinks it really doesnt make that much difference if a unit gets turned in 8 hours later.
- Joined
- Nov 1, 2001
- Location
- New Iberia, LA
David said:
First of all, I just said he was full of it on this subject.
Otherwise he is a very intelligent individual and has put together a very talented team for this project.David, what brought on SETI into this subject? I know nothing at all about SETI as I've never participated in it at all. However, I have been folding for longer than you have and I also know how the projects work.
What is the difference between me running 1 client and returning a big wu in 24 hours or running 2 clients and returning 2 big wu's in 36 hours(just an example). Will that extra 12 hours of processing time make any difference at all on when Stanford can base a new project on the work I just turned in? Realistically no, they won't be able to process the information fast enough for that extra 12 hours to make a bit of difference in the real world. The point I made in my previous post has much more bearing on getting results than worrying about a few extra hours taken due to running 2 clients on a HT capable machine. When their assignment servers send out giant wu's to slower machines (like my example in my previous post), they slow the generation of work results down way more than running 2 clients on HT capable machine.As far as money is concerned, I pay for the boxes and electricity I fold with, not Vijay or Stanford. If I'm generous enough to run multiple machines dedicated to his project, then he should have sense enough (which he does, fortunately) to not insist on running just 1 client/processor.
Running 2 clients on a HT capable machine and getting timely results shouldn't be too much of a problem anyways, since the slowest HT P4 proc made is a 2.4 Northwood. Of more concern should be working on tightening up Stanford's assignment server logic than worrying about this subject and get a better distrubution of work units to slower machines.
Fold on, Bro!
- Joined
- Feb 20, 2001
muddocktor said:First of all, I just said he was full of it on this subject.
David, what brought on SETI into this subject? I know nothing at all about SETI as I've never participated in it at all. However, I have been folding for longer than you have and I also know how the projects work.
As far as money is concerned, I pay for the boxes and electricity I fold with, not Vijay or Stanford. If I'm generous enough to run multiple machines dedicated to his project, then he should have sense enough (which he does, fortunately) to not insist on running just 1 client/processor.
Running 2 clients on a HT capable machine and getting timely results shouldn't be too much of a problem anyways, since the slowest HT P4 proc made is a 2.4 Northwood. Of more concern should be working on tightening up Stanford's assignment server logic than worrying about this subject and get a better distrubution of work units to slower machines.
Fold on, Bro!
I was just using SETI as a comparison - ie it is a project where WUs dont have any real order or relation, compared to the likes of FaH.
There have been long discussions (8 page+ thread) at Folding Community between Pandegroup and the Staff and Pandegroup insist that currently people running two clients on a HTT CPU is hurting the science compared to one client, enough for them to consider what they can do to persuade people not to run two clients. According to them, even a few hours can make a difference. I am not sure exactly how, but thats what we have been told. I am not trying to dispute your knowledge of FaH - I remember when you used to be a Team 32 folder (you fold for someone else now?) and I respect that you are by no means a Fah Newbie
You are right in saying that when running two clients, the deadlines are still met. It was suggested that perhaps making the deadlines sooner could help, but that would also penalise those using older and slower equipment for folding, and could end up eliminating quite a lot of potential FaH power.
As I have said, the best choice IMHO is to make Gromacs SMP aware for FaH, which is possible. This means Pandegroup get their quicker returned results, and HTT enabled CPUs will not be penalised by just running one client.
David
- Joined
- Oct 18, 2003
Well, if getting WU's back super quick is more important than getting more done, why do they let you install FAH on a slow computer that barely makes the deadlines? According to them it would be hurting the science more than helping it.David said:I was just using SETI as a comparison - ie it is a project where WUs dont have any real order or relation, compared to the likes of FaH.
There have been long discussions (8 page+ thread) at Folding Community between Pandegroup and the Staff and Pandegroup insist that currently people running two clients on a HTT CPU is hurting the science compared to one client, enough for them to consider what they can do to persuade people not to run two clients. According to them, even a few hours can make a difference. I am not sure exactly how, but thats what we have been told. I am not trying to dispute your knowledge of FaH - I remember when you used to be a Team 32 folder (you fold for someone else now?) and I respect that you are by no means a Fah Newbie
You are right in saying that when running two clients, the deadlines are still met. It was suggested that perhaps making the deadlines sooner could help, but that would also penalise those using older and slower equipment for folding, and could end up eliminating quite a lot of potential FaH power.
As I have said, the best choice IMHO is to make Gromacs SMP aware for FaH, which is possible. This means Pandegroup get their quicker returned results, and HTT enabled CPUs will not be penalised by just running one client.
David
