using USB flash drives to run farm from...

[rodsfree]

Guys,
Think it would work out to run a farm from a 32Mb USB flash / jump / pen drive?

I know I can boot into win98's version of dos from one. If I could boot into a small version of linux then I could use the flash drive instead of a hard drive or floppy. and not worry about net booting.

The reason I ask is because there are an ass-load of 32Mb pen drives on ebay for like $2.99 + S&H. Floppy drives cost about $7-10 depending. Plus floppies wear out.

If we need more storage... 128 Mb flash drives are going for around $10.00 +/- $2.00

What do you think?

 

[unhappy_mage]

You could go for the best of both worlds, and netboot off the pen drives . There are advantages to this: suppose Stanford releases a new client or something? If it's netbooted, you could just update the server and reboot the farm, rather than moving the pens over one at a time and reimaging them.

Linux would be easier to get going on read-only (practically, the writes'd be slow and kill the disk eventually) media than win98.

Even if you're not interested in this approach, I'd still be interested in seeing if it works. Like you said, cheaper and more durable than floppies.

 

[rodsfree]

There is a third way to go... I think.

Netboot off the pen drives - that way you don't access them to often.

Then use them to store local folding data.

That way if you loose the server then the clients can keep folding, or if you have to restart client or server or entire farm they can pick up where they left off.

The work folder on my little 24/7 folder only has about 11Mb in it and it's been running non-stop for a couple months now. So the netboot floppy is a 1.44 and it's the only thing on a 32Mb pen drive, that leaves about 30Mb for working data. Have a cron script running on the server to delete any files older than, say a month or so. And you'd never run out of local storage, but you wouldn't loose all that work if your PSU's started crapping out on you. Or the server died.

So u_m....
are we onto something?

 

[unhappy_mage]

Yes, we are mostly onto something. However, how fast are these boxes? Are you planning on running big WUs on them? If so, you'll probably need a) bigger pen drives and b) 512mb in each machine. The reason is peak usage. It's only using 11 MB now, but when it gets a new WU, it uncompresses it and disk usage can hit 40 MB for a short while. We hit this wall on Marty's farm, with 64 mb clients and timeless WUs.

The way around this is do network mounts onto the server, and store all the data there (this is coming in 1.1). This can cause problems too (especially scaling-related ones; what happens with 100 clients?), but at least the clients won't crash from out of memory errors as easily.

Oh, and the netboot floppies are about 32k. You could probably set it up so that that's all they take from the pen drive. Size isn't an issue.

It's much easier to use a ramdisk than a pen drive for the temp storage. That's the current structure, and it's a) faster and b) already implemented (Oh, and there's no support for usb storage built into the kernel, so that's no good.) The clients copy their progress to the server every so often, so the process can resume in case of reboots. It's not foolproof, but if Marty can use it, it's pretty close

Jk marty.

 

[Majic]

Ripped from LPerry's great site!

[font=times roman,tahoma,arial,times roman][font=times roman,tahoma,arial,times roman]FAH on a Pen Drive (8 MB or larger)
This version of Linux is for use with a USB Pen Drive and will work with FAH. It is very easy to set up and use, but if you're not at all familiar with Linux, it might be a little difficult. Instructions are in the SPBLinuxFAH zip file. If you are not clear on something or you need a little help, stop by the [H]ardOCP Distrubuted Computing forum and ask for help. One of the [H]ardOCP team members (turmelle) put this little package together and is using it to fold. You will also need another file which I have provided a link to below the SPBLinuxFAH download link. Currently I am not hosting that file, but if the link becomes broken please email me and I will upload a copy I have saved in my archives. Enjoy! If it works for you don't forget to say thanks to turmelle for an awesome piece of work.
SPB Linux for FAH
Make a USB Pen Drive Boot Image[/font][/font]

Not sure on current state of the project but, it has been done!

 

[unhappy_mage]

I tried it a while ago. It works fine, but it's a bit of a pain to get set up right at first.

Maybe I should make a fold-server that fits on a usb pen...

 

[rodsfree]

It's much easier to use a ramdisk than a pen drive for the temp storage. That's the current structure, and it's a) faster and b) already implemented (Oh, and there's no support for usb storage built into the kernel, so that's no good.) The clients copy their progress to the server every so often, so the process can resume in case of reboots.

So what happens when the server dies, or you loose power for the ramdisk?

I'm not the expert - I'm just thinking out loud, sorta. ( I admit it I'm cheap - $3 is less than $7)
Pen drives are alot cheaper than DDR ram right now, even if we have to go up to the 128Mb pens. They offer non-volitol<sp> storage. and most motherboards can boot from a pen even if they can't netboot. I know that the latest linux kernal includes USB storage support, or all the USB drive vendors would be pissed, so there is a working way of doing it out there.
Maybe mounting the pens after netbooting. Then the OS should see them as just another mouted drive.
I don't know.... most of my little *nix experience is with HP-UX 10.20 - which is defenately not up to date and isn't mainstream.

 

[unhappy_mage]

So what happens when the server dies, or you loose power for the ramdisk?
...
Pen drives are alot cheaper than DDR ram right now, even if we have to go up to the 128Mb pens. They offer non-volitol<sp> storage. and most motherboards can boot from a pen even if they can't netboot. I know that the latest linux kernal includes USB storage support, or all the USB drive vendors would be pissed, so there is a working way of doing it out there.

If the server goes down, then you bring it up again, and no harm done. If the client goes down, you lose between 0 and 20 minutes of work for that client; every 20 min it backs its progress up to the server, with a real hard drive in it. Granted, pen drives are cheaper than ram, but you can use ram for more things. It's "volatile", btw. /grammar_nazi

The latest linux kernel does indeed have support for usb storage, but I didn't include it in order to save on space. I have almost all the network stuff enabled in the client kernel, and it's only 1.6MB which is pretty good, considering all it's got in it. However, it doesn't have support for things like USB storage, or scsi, or IDE, or almost anything. About the only thing turned on in terms of storage are the initrd and the tmpfs drivers. Everything else is out to save space. I could compile a seperate version for you, but that's one more customized installation to deal with. If I'm gonna do something like this, I'd like to either do it with the tools included, or put the tools in fold-server proper and document it so everyone can do it. This is more or less the problem I'm hitting with NAT. Everyone wants to use it, I'm not ready to release quite yet, so I have to tell people individually how to set it up. I've learned my lesson from this; the next big feature set is coming, but I'm not gonna leak any of it until 1.1.

Thanks for bouncing some ideas off me. I need input from the people using it and wanting to use it to know what to add. I've just put up my list of things to come on the website here. I'll update as progress comes. Hopefully I'll actually have some updates to post

 

[gigglebyte]

I don't know.... most of my little *nix experience is with HP-UX 10.20 - which is defenately not up to date and isn't mainstream.

lol...I used to work for HP and it has been a while since I have seen anybody talk about HPUX 10.x

 

[rodsfree]

lol...I used to work for HP and it has been a while since I have seen anybody talk about HPUX 10.x

I'm a Manufacturing Engineer and we have coordinate measuting machines that run off of HP B180L computers. The software was orginally written for 10.20 and everyone is more concerned with it functioning the way it came from the machine builder than speed or staying on the bleeding edge. I've got about 5 of these things I responsible for and I've tried to ensure that they are all identical... down to the brand & size of harddrive that's in them so I can use Ignite to clone them easily.
They work, they are stable, and they are not going to change anytime soon.
So I'm learning HP-UX 10.20

Any good links for 10.20 would be appreciated.

 

[rodsfree]

If the server goes down, then you bring it up again, and no harm done. If the client goes down, you lose between 0 and 20 minutes of work for that client; every 20 min it backs its progress up to the server, with a real hard drive in it. Granted, pen drives are cheaper than ram, but you can use ram for more things. It's "volatile", btw. /grammar_nazi

The latest linux kernel does indeed have support for usb storage, but I didn't include it in order to save on space. I have almost all the network stuff enabled in the client kernel, and it's only 1.6MB which is pretty good, considering all it's got in it. However, it doesn't have support for things like USB storage, or scsi, or IDE, or almost anything. About the only thing turned on in terms of storage are the initrd and the tmpfs drivers. Everything else is out to save space. I could compile a seperate version for you, but that's one more customized installation to deal with. If I'm gonna do something like this, I'd like to either do it with the tools included, or put the tools in fold-server proper and document it so everyone can do it. This is more or less the problem I'm hitting with NAT. Everyone wants to use it, I'm not ready to release quite yet, so I have to tell people individually how to set it up. I've learned my lesson from this; the next big feature set is coming, but I'm not gonna leak any of it until 1.1.

Thanks for bouncing some ideas off me. I need input from the people using it and wanting to use it to know what to add. I've just put up my list of things to come on the website here. I'll update as progress comes. Hopefully I'll actually have some updates to post

I completely agree with everthing you said. I definately don't have to or want to run a different setup from everybody else. I'm bouncing ideas around trying to come up with the cheapest and most stable farm plan I can.

Could you make the 20 minute "back up to the server time period" variable? Depending on the protein you're working on, 20 minutes could be several frames. If not, no biggie.

Thanks for all the hard work.

 

[unhappy_mage]

Could you make the 20 minute "back up to the server time period" variable? Depending on the protein you're working on, 20 minutes could be several frames. If not, no biggie.

It's already a variable, I just hadn't documented that yet. I'll put up a section on that. Thanks for the suggestion.

 

[Fenris_Ulf]

If you want to run Win98 (less learning curve for Linux newbs, but that's OK), then I recommend getting a copy of 98Lite. You can strip 98 down to the bare essentials that way to fit on a USB drive. Run either a ramdrive for storage or map a network drive to your server if you don't feel secure.

The number of boxen you'll have may determine your threshold for pain in the various approaches. Cloning a few USB drives every time the FAH client changes wouldn't be too bad, but 100 could be tedious. Server side you could have a different directory for each client and a script to change client packages relatively simply for a large number of directories. This would be my approach, plus EM3 or a monitoring script would be pretty simple to configure this way, and data wouldn't be volitile.

 

[Hito Bahadur]

If you want to run Win98 (less learning curve for Linux newbs, but that's OK), then I recommend getting a copy of 98Lite. You can strip 98 down to the bare essentials that way to fit on a USB drive. Run either a ramdrive for storage or map a network drive to your server if you don't feel secure.

The number of boxen you'll have may determine your threshold for pain in the various approaches. Cloning a few USB drives every time the FAH client changes wouldn't be too bad, but 100 could be tedious. Server side you could have a different directory for each client and a script to change client packages relatively simply for a large number of directories. This would be my approach, plus EM3 or a monitoring script would be pretty simple to configure this way, and data wouldn't be volitile.

nice to know. I think windows FAH might run a bit faster than Linux. I'm just basing this on the fact that the Pentium II 400 mhz runs the same speed as the 233 I had folding and the only difference was the OS.

 

[KodiakStar]

what os' were you running at the time Hito? ie which linux which windows?

 

[Hito Bahadur]

Well, like I stated, 2 different systems. The one was a Pentium II 233 with 96MB RAM, 6 GB HDD and operating Windows 98 Second Edition (this was the system I was going to send to you before it gave up the ghost).

This new system is a Pentium II 400, 196 MB RAM, 3 GB HDD with Fedora Linux but running in terminal mode.

Timeless Tinkers took just over an hour/frame with the 233 (say 61 minutes). They take 59 minutes with the 400. I would have expected a little more performance boost given the difference in MB.

 

[KodiakStar]

Ahh ok... yea im installing win 98se on my 150mhz laptop 24mb ram... i doubt it will get through them that fast.. but.. it will still fold.. (if it dont die on me)

and to think... 4 years ago.. err. wait.. umm. 7 years ago *cough* it played UltimaOnline
just fine...