What does this actually mean?

[EvilAlchemist]

Any Ideas to what this actually means?
"Right now, our SMP client seems to be running fairly well under Linux and OSX -- operating systems based on UNIX, as is found on supercomputers. We use a standard supercomputing library (MPI) to run these WU's and MPI behaves well on Unix-based machine. MPI does not run well on Windows and we've been running into problems there. However, as Windows MPI implementations mature, our SMP/Windows app will behave better. Along the way, we also have a few tricks up our sleeve which may help as well. However, if we can't get it to run as well as we'd like on Windows, we may choose to overhaul the whole code, as we did with the GPU1 client (which was really hard to run)."

Taken From:
http://folding.typepad.com/news/2008/06/what-does-the-smp-core-do.html

 

[alan2308]

There has been talk of dropping WinSMP recently.

But reading that, I'm getting that they're not ruling out rewriting the client without MPI, as a last resort. This doesn't necessarily mean dropping any platforms, that would depend on what is used in place of MPI.

 

[Xilikon]

Basically, if the MPI libs is causing too much headaches and trouble, they will be rewriting to get rid of those libraries but they need to find a different way to handle threads. However, I guess their experience with the GPU2 (which is basically a parallel processor) will help them get the right ideas to code. It's perfectly possible to do it without the MPI libs but you need to understand how threading works and how to communicate between threads within the die instead of resorting to the network stack.

The MPI libs is written for a cluster system made of different systems tied together by a networking layer (often gigabit switch and nics or fiber). They just did the ghetto way and ported it by using the Windows networking stack as means to synchronize but as everyone know, any network hiccup will cause issues or in the case of Syribo, networking issues preventing the client from installing correctly. For them to get out of beta, they need to code it so it's install and run without too much users intervention (and no more password and admin requirement, which is bad security pratice).

 

[Wheresatom]

Does this mean the few of us not already running a virtual machine need to get on the bandwagon with some quickness?

 

[Kendrak]

/shakes fist in air

I like my windows SMP! I run it on 2 boxen and it was going to become 3!

/sigh

This makes Evil exp with the standard client even more important.

 

[SmokeRngs]

Basically, due to how much some things in Windows sucks Stanford is having a difficult time getting the Windows SMP client anything close to stable. Because of this, they are looking at a way to rewrite the client so it doesn't require some of the current client's parts because Windows sucks at running those parts due to the fact that being multithreaded is rather new to Windows compared to some of the other OSes out there and the good implementation on the other OSes doesn't carry over very well to Windows.

It's very unlikely this will mean the end of a Windows SMP client. There aren't near as many people running Linux as a daily OS and folding as there are people running Windows. Because of this, they wouldn't want to relegate the SMP client to Linux only as the production from the client would go down considerably. This is important because as time goes on, it's going to be difficult to find single core machines which will eventually push the SMP client mainstream instead of being just a niche. Of course, Stanford will need to something like Linux and make a single client which can do regular or SMP work units.

Basically, do not count on a new SMP client for Windows anytime soon. There may be some newer Linux SMP clients but I wouldn't count on that either as most of the SMP resources are probably directed at the Windows client.

Bottom line: Stanford hasn't changed in regards to clients. They still don't have a fucking clue how to make or refine a client so it's no longer beta and instead focus resources on new clients that don't fix the problems of the old clients and introduce new problems. Of course, this is while the users of the clients bitch and complain and continue to be ignored.

I wonder if I came off too cynical?

 

[Wheresatom]

I wonder if I came off too cynical?

I am still a bit new, but that sounds about accurate. Cynicism should be about Par around here.

 

[Sunin]

Wow... if the end result is they drop windows they are going to lose what 80-90% of the community? I pray they work this out for windows and soon. I thought they had addressed this with the up and coming A2 Core? Is this in addition to that ?

 

[ICE_9]

Wow... if the end result is they drop windows they are going to lose what 80-90% of the community? I pray they work this out for windows and soon. I thought they had addressed this with the up and coming A2 Core? Is this in addition to that ?

Or you just switch back to the standard client. Either way, work still needs to be done.

 

[Sunin]

LOL i thought 2 x SMP was silly... I'm definitely not going to be all that happy with 4 x Std client, plus the reduction in scientific output!

 

[alan2308]

Basically, due to how much some things in Windows sucks Stanford is having a difficult time getting the Windows SMP client anything close to stable. Because of this, they are looking at a way to rewrite the client so it doesn't require some of the current client's parts because Windows sucks at running those parts due to the fact that being multithreaded is rather new to Windows compared to some of the other OSes out there and the good implementation on the other OSes doesn't carry over very well to Windows.

It's very unlikely this will mean the end of a Windows SMP client. There aren't near as many people running Linux as a daily OS and folding as there are people running Windows. Because of this, they wouldn't want to relegate the SMP client to Linux only as the production from the client would go down considerably. This is important because as time goes on, it's going to be difficult to find single core machines which will eventually push the SMP client mainstream instead of being just a niche. Of course, Stanford will need to something like Linux and make a single client which can do regular or SMP work units.

Basically, do not count on a new SMP client for Windows anytime soon. There may be some newer Linux SMP clients but I wouldn't count on that either as most of the SMP resources are probably directed at the Windows client.

Bottom line: Stanford hasn't changed in regards to clients. They still don't have a fucking clue how to make or refine a client so it's no longer beta and instead focus resources on new clients that don't fix the problems of the old clients and introduce new problems. Of course, this is while the users of the clients bitch and complain and continue to be ignored.

I wonder if I came off too cynical?

Too cynical? I wonder if you're cynical enough.

You forgot the part where they talk down at you and point the finger back your way every time anything goes wrong. But they don't say it directly themselves, they hide behind the forum and its mods. Because there could never be anything wrong with the client, its flawless. It HAS to be an instability with your setup. And how dare you hold back science with your faulty hardware causing EUE's?

[/RANT]

 

[BassTek]

Everyone should just switch to Vista, it seems to handle network interruptions quite well. Probably due to the newer network code that it contains.

 

[jws2346]

Now I know I've "overheated" one of the 2x brain cells I have left. I've had experience with the Linux SMP client and I'm currently using it on 4x boxen. As of yet I've had no experience using any kind of GPU folding client. What I call my "main boxen" is a Q6600 runing WinXP w/SP3 as it's OS and it will hopefully be VISTA Ultimate eventually.

I'm no programmer and I'm not much up on computer languages. I don't know or even care what the MPI thing is except in how it may effect my folding by not being able (shutter) to use the WinSMP client. Remember in real life I am not and don't plan to be a computer technician. I only have desktop boxen and I'm only a computer hobbyist. To even think about doing away with the WinSMP clients IMHO is right up there with being an incompetent computer programmer or administrator. IMHO it would be a very foolish idea because, as someone stated above, a great majority of folders use the windose WinSMP clients on their work or home boxens and love the "poop" (PC ) out of it. I can't imagine any business converting their computers over from windose to Linux because, if they're not using MAC's, they'll have to if the WinSMP client is done away with and they don't want to do use the nonSMP windose clients for some reason or "heaven forbid", stop folding.

I know I won't convert my "main boxen" over to 100% Linux any time soon (it's already a "dual booter") . I did try it one time usuccessfully and the main reason it failed is, other than folding, the rest of the computer world seems to be the windose or OSX . (games, utilitys, etc) ATM my windose boxen folds 24/7 using WinXP w/SP3. I don't say I won't convert 100% Linux because I'm not impressed with Linux (either flavor) because I am. I just don't think any Linux versions are "up to speed" yet to compete with the windose desktop world. (but it may be in the near future)

Just my $0.02 cents worth.

 

[russnuck]

So far the best setup I found for me atleast.

Q6600
6-8GB Ram
XP/Vista64
2xCentOS VM

I can run 2 VMs + 2 Windowed instances of EQ + 2 Windows Terminal Sessions + iTunes and still get frames in the ~14 min range

 

[SmokeRngs]

Wow... if the end result is they drop windows they are going to lose what 80-90% of the community? I pray they work this out for windows and soon. I thought they had addressed this with the up and coming A2 Core? Is this in addition to that ?

The core really has nothing to do with the functionality of the client in a manner of speaking. The MPI stuff is obviously disconnected from the core because you can run different cores with the same client.

From what I know, the A2 core is supposed to scale better when going to 4+ cores and that's it's main function.

LOL i thought 2 x SMP was silly... I'm definitely not going to be all that happy with 4 x Std client, plus the reduction in scientific output!

Heh, my rarely used Windows install on my main machine is setup that way. I'll go weeks between the times I use it so the SMP client is not going to be of any use in that situation. The regular console client is a hell of a lot more stable than the SMP clients so it's usually a set it and forget it type deal. Besides, it installs properly as a service so you won't even see it running if you don't want to.

Too cynical? I wonder if you're cynical enough.

You forgot the part where they talk down at you and point the finger back your way every time anything goes wrong. But they don't say it directly themselves, they hide behind the forum and its mods. Because there could never be anything wrong with the client, its flawless. It HAS to be an instability with your setup. And how dare you hold back science with your faulty hardware causing EUE's?

[/RANT]

I have a feeling that's why I normally stay away from those forums. I'd just end up getting banned really fast anyway.

Everyone should just switch to Vista, it seems to handle network interruptions quite well. Probably due to the newer network code that it contains.

/me shudders. I'll stick with Linux thank you very much. I have Vista installed on one machine right now as I wanted to see how it would do as a multimedia server. Well, I wasn't impressed with it at all. Add to that problems with being stable (at least one reboot for no reason each day) and it's something I don't care for at all. Once openSUSE 11.0 final is released in a couple days, Linux is going back on it. Yes, it's Vista with SP1 and all patches and updates.

 

[aldamon]

With the amount of work GPU2 and PS3 are and will be doing, do they really need Windows SMP? Windows could be covered 100% by the standard client or a better GUI that opens multiple work units for multiple cores automatically (like WCG, <cough>).

PS3
GPU2
Linux SMP
Windows Uniprocessor

Probably same production in the end. Points for Windows standard client could be adjusted like they're already doing.

 

[Kendrak]

/me shudders. I'll stick with Linux thank you very much. I have Vista installed on one machine right now as I wanted to see how it would do as a multimedia server. Well, I wasn't impressed with it at all. Add to that problems with being stable (at least one reboot for no reason each day) and it's something I don't care for at all. Once openSUSE 11.0 final is released in a couple days, Linux is going back on it. Yes, it's Vista with SP1 and all patches and updates.

I can't understand why you have such problems

Years ago everyone hated win 98, but I would have uptimes of well over 3 months before I got the upgrade itch, and have to shut down for install (installing RAM while system is on is bad)

I have uptimes over well over 3 months with my main boxen on XP, running 2X of the SMP's and running all kinds of games and well... everything.

I just don't know, maybe the hardware fairy likes me

 

[SmokeRngs]

I can't understand why you have such problems

Years ago everyone hated win 98, but I would have uptimes of well over 3 months before I got the upgrade itch, and have to shut down for install (installing RAM while system is on is bad)

I have uptimes over well over 3 months with my main boxen on XP, running 2X of the SMP's and running all kinds of games and well... everything.

I just don't know, maybe the hardware fairy likes me

I've had very few problems with XP that weren't my fault and the same with 2k. I've also run Vista on my main system with few problems. However, I just don't care for the OS as it does things I don't want or need it to do and it doesn't do some things I do want or need it to do. I just flat out don't like Vista and I don't think I ever will.

Also, I had no trouble with 98. I ran it in tandem with 2k for a long time. I only used it for stuff that wouldn't run or run properly under 2k of course which was mostly games.

 

[Kendrak]

ok... if it's you just don't like it

 

[SmokeRngs]

ok... if it's you just don't like it

I've used every consumer OS from Microsoft since Windows 3.1 (even though 3.1 is actually just a graphical shell around DOS). I've had trouble here and there with all the OSes and the only one I refused to use was ME. I'm sure everyone can agree that was the bastard child which never should have been conceived. /me shudders.

As it is, I don't care for Vista and I've had some trouble with it on most machines I've run it on starting back with the 5270 beta. It's just not an OS that works for me.

 

[APOLLO]

As it is, I don't care for Vista and I've had some trouble with it on most machines I've run it on starting back with the 5270 beta. It's just not an OS that works for me.

I totally agree. I just can't see the need to install Vista on anything custom built that's worth the potential hassles. Heck, if I could get rid of it from my laptop without losing some functionality, I would do so.

 

[Sunin]

I'm loving Vista Ultimate 64 on my HTPC not a single issue thus far!

 

[relic]

I have to agree, Vista is better than XP for folding.
Stability is better than XP , but not as good as Linux.
I run all three. I'd stick with Linux if the support was there, but so far Vista is the best of the three.

 

[BillR]

Basically, due to how much some things in Windows sucks Stanford is having a difficult time getting the Windows SMP client anything close to stable. Because of this, they are looking at a way to rewrite the client so it doesn't require some of the current client's parts because Windows sucks at running those parts due to the fact that being multithreaded is rather new to Windows compared to some of the other OSes out there and the good implementation on the other OSes doesn't carry over very well to Windows.

It's very unlikely this will mean the end of a Windows SMP client. There aren't near as many people running Linux as a daily OS and folding as there are people running Windows. Because of this, they wouldn't want to relegate the SMP client to Linux only as the production from the client would go down considerably. This is important because as time goes on, it's going to be difficult to find single core machines which will eventually push the SMP client mainstream instead of being just a niche. Of course, Stanford will need to something like Linux and make a single client which can do regular or SMP work units.

Basically, do not count on a new SMP client for Windows anytime soon. There may be some newer Linux SMP clients but I wouldn't count on that either as most of the SMP resources are probably directed at the Windows client.

Bottom line: Stanford hasn't changed in regards to clients. They still don't have a fucking clue how to make or refine a client so it's no longer beta and instead focus resources on new clients that don't fix the problems of the old clients and introduce new problems. Of course, this is while the users of the clients bitch and complain and continue to be ignored.

I wonder if I came off too cynical?

Too cynical with regards to the crap for code we have been subjected to of late? Too cynical because the points issue fix is a pen and paper away?

No buddy, you are right on the mark:

http://www.hardforum.com/showpost.php?p=1032625980&postcount=39