maxFAH - windows multi client installer/manager

[codysluder]

It may be obvous to you, but you didn't ever say that this is intended to be for Windows-only. This forum already has other installers for Linux and for the Macintosh OS so that's not a big deal, but it would be nice if you had said so in or near your title. (For the PS3 people are stuck with Sony's installer.)

[MtM]

It may be obvous to you, but you didn't ever say that this is intended to be for Windows-only. This forum already has other installers for Linux and for the Macintosh OS so that's not a big deal, but it would be nice if you had said so in or near your title. (For the PS3 people are stuck with Sony's installer.)

No I didn't, basicly because I got mostly the same code in Realbasic which can compile to both linux and mac os ( not sure about mac but I think it does ). It's just realbasic is much harder to program then .net and I am stuck with certain pieces of code I couldn't port from .net to realbasic. I do plan on making it OS independant, but I'm not sure if it's even possible the way I'm approaching it atm

When it's done, it will be possible to port the .net code to asp.net maybe, and make the installer browser based. Btw, afaik Windows does have the larger userbase.

I got some help from someone willing to test the deino service part, if everything works out I will release the installer today or tomorrow.

[MtM]

Project suspended, asked for permission to make it more automated and less dependant on user action ( deino and mpich install ).

[smartcat99s]

FYI, I've got VS2003, 2005, and 2008 and denio SMP running on a XP SP2 machine.

PM me when this gets rolling again if you want the help.

[MtM]

FYI, I've got VS2003, 2005, and 2008 and denio SMP running on a XP SP2 machine.

PM me when this gets rolling again if you want the help.

Much appriciated, I'll keep it in mind for sure

[MtM]

Bumpity, who got some brave shoes and wants to help me check if everything works as it should on more systems then I got

[MtM]

Not everyone at once, jeez

[toTOW]

If I have some time, I could try it tomorrow ... but you didn't post any link to the binaries

[MtM]

I pulled the link since I was changing the setup method. Mpich already requires no more user interaction to set up, now working on Deino to work the same way. Thought it would be confusing to get people to report on a binarry which will be drasticly changed a day or so later. I think I can put them up today ( this evening, gmt+1 zone ),

Sorry toTOW I thought I had said it in the 1st post as well.

[MtM]

23:28

Link up in first post, Mpich and Deino installed and configured without user action needed. Will put up some new screenshots, as the main form now includes a password field which is used to install the smp client libraries without interaction.

Tech notes: I wanted to do Deino in the background as most of Mpich is done, but the developers have a> not documented the command line options for create_credential_store.exe ( the guide has a flag /true or something, but this flag does not change the behaviour of the console if you leave it out ) and b> do not allow me to attach to the console ( all tools have anti debugging/dissasembly options in them ). So I couldn't attach my process to their message loop, and I had to call them with shellexecute, and use windowsapi calls to set active window, send keyb events to the window and check if things went ok by creating wait statements and checking if the process has excited after the 'normal' amount of events. There been times ( I tested this about 20 times in a row, and some during debugging sessions with the whole installer ) where the first try did not go as planned so there is a built in loop which treis to configure the credential store 6 times before 'calling it quits'. In those loops it kills the still active process, waits a short time, starts over. This proofed to work 100%, but I'm not sure about 'in the wild' cases.

There next release of DeinoMPI I read on their site will include the sourcecode for the command line tools, and I hope also they will include the right command line parameters for the cl tools ( sourcecode for me is not even important ).

Going to add the basic start/stop options to maxTray now, think it will take a few minutes so people who try this will also have a limited gui to use to control the installed clients.

[bruce]

Have you tried contacting the developers?

[toTOW]

What should we try first ?

[MtM]

I haven't emailed them yet but I think I can do that indeed. Their documentation is abit scarce, they mention 'if you start without parameters you will have to configure them... ' so they hint to there being command line options

You can try the installer from the link in post 1, that will install clients on your system based on your hardware, and configure them based on the guides and standard console settings. I just updated maxTray to be able to start and stop clients ( both console and service, though I haven't tested service ).

Ow and I forgot to turn of fw, so I had 2 clients just dl wu's which I can not finish ( or I would have to put my vm's offline which will mean I can't finish those in time I guess.. ) Sorry for that, I know that's the worst thing one can do for the project! If you want I can post pcrg for them so they can be 'reset' maybe?

I'm abit to tired to put up new screenshots

Most questions/room for improvement is if I had a list of supported ati cards which I could compare with the found hw, now it's just trying to see if it's hd2xxxx or up. And driverversions I can check, but I don't atm because I don't have an comprehensive list of which one's are good or not. If someone would create a list for those, both ati and nvidia I can compare installed drivers. Ow if someone feels like doing so, provide the driver version as reported by wmi please, ( for example 7.15.11.7784 for 177.84, or maybe should I just disregard the other sections? I'm more inclined to think Ati is more dependant on the right drivers anyway as with Nvidia they include CUDA in all release drivers? Though, I still usually tell people to get them from CUDA_GET )

Then, I was thinking since I already start/stop the clients and got their associated processes, I can also set affinities myself ( thought about using 3rd party lib since my first treis at this a long time back didn't work because I didn't have the right permission set. But I think I can get it working now, and if not I can ask someone to help I'm sure ).

Like I said I can also offer kind of an 'auto update' since the tray control also loads the webxml and can compare running/installed clients with the one's there.

I'm putting in the log parsing code, so I can track eue's and all their info, and I'm writing a vb class for qeueu.dat parsing so I can track some things through that as well.

edit:

Ow and what the best tests would be btw, is you who know how this works already treis to use it and break it! Try to find the weak spots and where things should be diffrent. I already changed some major flaws in the past day ( eg single core ht enabeld cpu's who got the option to install smp clients! ). And I need to probably take a core from the cpu's for each 2 gpu's or so on xp, but not sure? Those things would be nice to have answerd.

IT now only checks for the SMP Affinity changer software from nick4eva, and if installed offer a choiche to use one client per 2 or per 4 cores. I know that's probably not the intended PG way, so maybe I need to remove that. I think I will be using afc as external tool to set affinities for multiple clients ( smp+gpu combinations) and give a gui config option for the afc program.

Further, it is possible to connect to networkd windows boxes and remotly install folding on them so I might add that to maxTray.. I will add the parsing ect to it, and probably a deamon style remote control option at some time. And email notifications on client outages/eue rate to high ect.

I was writing some web services to collect eue data, but my host doesn't support asp.net so I put those aside for awhile

[toTOW]

I'm getting this error at WMI CPU detection :



The setup is a bit exotic : Windows 7 Build 6801 in VMWare Workstation 6.

I tested it on the host (XP Pro SP3) and it's running fine : it successfully detected my CPU and my GPU.

[MtM]

Yeah vm's don't expose wmi in the right way

I just noticed a ss from a member of xs who had a dual core and his Smpcores combobox was enabeld, it's only an appearance flaw as even if he set it to two or four he would not be able to set up more then one smp client.

How did deino setup go? Did create_credential_store work in one pass?

I send the developers an email about the command line options, and not being able to attach the console. Hope they will respond and offer a solution so I can do it more cleanly.

Also maxTray does start and stop clients, but with smp it sometimes opens a second client when there already is one open. Which is weird because on startup the process is added to the client class, and when the startstop procedure is ran it checks for the process, if it's valid then it has been ran once, if not it's not running at all, and if it's valid and is still running it should switch to close client.

It's just for now anyway, the client processes will eventually be tracked with events but I wanted to show abit of functionality. Also I want to give the option to create seperate tray icons for each client, and was hoping PG would not object if I used the icons from the tray client? I'm not a graphics guy so I need someone else to give me some nice icons to represent their state ( eg client stopped, client running, client error ( as in EUE/config error/failed upload over a certain limit/can't get work ect ).

Thanks for trying!

edit: during debugging, I did get into a strange situation. I added a check to startup if the client folder already exists and if so run the uninstaller first to prevent problems. But the uninstaller fired up while the setup program continued execution as well and that's a bad thing as both at stages lock user input to use api calls to uninstall/install smp hosts (deino/mpich) and both try to put a window in the foreground and send messages to it. Both of those together do not work well ( lock user input can be canceld with ctrl+alt+delete if someone get's stuck somewhere btw, though it should not be needed ). I treid to replicate the event but haven't been able to, which is to be expected as the installers uses process.waitforexit and I think the situation occured because of the debugging environment.