When will F@H SMP stop using MPI?

[CoW]8(0)]

It's nice that MPI is making use of all the cores on a dual and quad CPU system, but when is F@H going to move on? It's been causing some very annoying internet issues. I find that large flash movies as well as other large file downloads continually fail because of the MPICH2 process.

 

[Carnival Forces]

what exactly is mpi?

 

[Jon855]

what exactly is mpi?

http://en.wikipedia.org/wiki/Message_Passing_Interface

 

[GoldenTiger]

I found the SMP client was killing game performance by not yielding CPU cycles properly when I tried it with a quad-core for awhile, so I had to shelf it for now rather than running it 24/7 as I'd like to. I couldn't find a solution... not sure if there's a way to manually launch it at will and then turn it off in the morning, but I couldn't find one with a cursory set of google searches.

 

[7im]

The SMP client is designed as a high performance client. It will impact system performance some, or a lot depending on your hardware, and usually needs to run almost 24/7 with the short deadlines. However, the PPD is also comensurate with those conditions.

And I doubt they will ever stop using MPI. And BTW, blame Microsoft for the MPI problems when the network settings change or get disconnected. The problem doesn't happen in Vista and 2K3 Server, because MS fixed the problem.

So the SMP client may not be suitable for all folding situations. In those other situations, a CPU or GPU client may be a better match.

 

[BillR]

The SMP client is designed as a high performance client. It will impact system performance some, or a lot depending on your hardware, and usually needs to run almost 24/7 with the short deadlines. However, the PPD is also comensurate with those conditions.

And I doubt they will ever stop using MPI. And BTW, blame Microsoft for the MPI problems when the network settings change or get disconnected. The problem doesn't happen in Vista and 2K3 Server, because MS fixed the problem.

So the SMP client may not be suitable for all folding situations. In those other situations, a CPU or GPU client may be a better match.

I’m not at all sure I’d blame Microsoft for the MPI and SMPD issue and God knows they certainly have their share of things to be blamed for.

The only reason MPI still exists is the combination of Cluster and Super Computing. To make it all work the various machines need to keep in touch so to speak and thus the MPI protocol was established. It also allows SMP programs to communicate with the various cores on a given CPU.

The really unfortunately part of all this is the MPI was written to work with FORTRAN (a programming language for you youngens) and depending on your point of view FORTRAN is split between two camps, those who feel it should be allowed to die a normal death and those who rely on it for scientific oriented processing.

The two big proponents of the continuation of FORTRAN are, at first glance, so far removed from each other one would think neither scientific community shared anything in common. The one community is Astrophysics and the other is our old friend Biophysical Molecular or Molecular Biophysics depending who defines it. The commonality that is shared is the ability to easily track various processes such as the movement of stuff in space and track the processes going on in the human Genome.

It should be known that MPI was a dead issue for almost 10 years until about 1976 when a renewed interest brought about version MPI-1-2 followed by MPI-2.

Long story short, MPI-2 is now fully compatible with not only FORTRAN but Perl, Java and C++ and I believe a few other languages.

The only reason anyone can give me that is slightly valid to continue with FORTRAN is that in this particular use it is simple to use and teach to others. Oh yes, it’s open source so it’s free as well.

If you do any research into FORTRAN virtually all the sites you will encounter spend a substantial proportion of their web space justifying its continued use.

As a result of all the above MS was more or less forced to add MPI support to it’s systems in spite of the fact that MPI is high on the list of legacy applications MS would love to drop.

To those folding I would advise you go into the properties of your NIC and disable its power saving feature, which should help you a lot.

 

[Xilikon]

There is actually a very simple solution : Move to Vista

Vista have a rewritten TCP/IP stack so it would behave much better with the SMP client so even network hiccups wont kill the client.

 

[Pocatello]

There is actually a very simple solution : Move to Vista

Vista have a rewritten TCP/IP stack so it would behave much better with the SMP client so even network hiccups wont kill the client.

Vista works well for me. I like it.

 

[BillR]

There is actually a very simple solution : Move to Vista

Vista have a rewritten TCP/IP stack so it would behave much better with the SMP client so even network hiccups wont kill the client.

To be honest I don’t have issue one with any of the folding home clients not functioning on any supported OS. That doesn’t mean they function well.

Yes, for a fact Vista does behave a bit better then XP, in fact I kind of like Vista 64.

I’ll go way out on a limb here with the same complaint any educated desktop user should be aware of, everything today is being hardware driven, not software driven.

From a developmental perspective this paradigm truly sucks and is holding everything back.

Here I sit at a desktop computer with 14 gigs of processing power and very damn little runs faster then it did a few years back, in fact some things are slower.

14 Gigahertz, think about it, that is beyond the power of a super computer a mere 8 to 10 years ago yet we use it to run office, some chat programs, play some games and do some folding.

We are code strung because there is no true software development going on and I find the same situation in Linux.

We have processors that far exceed the capacity of the peripherals that they need to run to simply function at all and to date nobody has yet addressed the kernel and the horrible scheduling processes that are built in, not to mention the equally bad CPU internal scheduling.

Legacy code has to go, period. RAM that can’t keep up with data requests from the CPU have to go. Internal CPU scheduling is the pits, the scheduler makes an error, backs up the RAM and how many computer cycles are simply gone.

Take a quick look at Dreamscape for a moment. They make CGI video and they broke all the rules. First they tossed all the Sun work stations, then the Intel work stations and last but not least most of the MAC stuff went out the door. Following that they looked at CPU architecture from both Intel, including Conroe and they looked at AMD as it stood.

In the end a decision was made to go with AMD but, they would write their own code because universal code in a word sucks. Now, with off the shelf parts they can out produce Intel based solutions by a major factor.

For something as critical as Cancer research etc would everyone not gain with some new dedicated code? One has to assume yes it would. Will anyone take the time and effort to write the new code? No, as long as the cycles are free and the power is free then the work product is free. Free trumps faster every time.

Sorry for the small rant and no disrespect to Stanford but as it stands results vs. costs could and should be fixed.

 

[aldamon]

Sorry for the small rant and no disrespect to Stanford but as it stands results vs. costs could and should be fixed.

Excellent rant. 9.5/10.

0.5 point deduction for saying you like Vista.

 

[BillR]

Excellent rant. 9.5/10.

0.5 point deduction for saying you like Vista.

But I hate Hillary, doesn't that offset the .5?

 

[sharp]

...ah yes, but there was no blame for Hillary to be found in said rant.

I suspect the 9.5 will stand. Take heart; unlike the Olympics, not everyone gets a 9 or higher here.

sharp