using really old computers

[danuitti]

I have a few really old computers that I can put to use for DF. These are Pentium classics. While DF runs, setting the check in flag to 1000 units means that it checks in very slowly. A power outtage or system crash means that all work is lost since the last checking.

I suggest that the minimum setting for this flag be settable to values as low as 100, which would further minimize the data lost.

For my cranky old rig, a setting of 100 may be 4 hours of work.

Thanks

Dan

[Brian the Fist]

Actually you're lucky it works at all since the minimum requirements are a Pentium II or higher. Anyways, since a $500 machine today can replace about 20 Pentium I's, we encourage users with many old machines (that are their personal property) to go out and buy a new machine as just the savings in power usage will make the computer pay for itself. And as a bonus you'll get a good set of paperweights in the bargain ;)

[danuitti]

Putting the cranky old computer to work is for the FUN of it. I've got powerful computers contributing.

Hey, let's try to save room in the dump for something good. If you want to be an environmentalist about it, let's make these ol' clunkers work again. What do we have those NUC plants for anyway?

I do not believe that the minimum specs were meant to say that these ol' clunkers couldn't do it; they simply aren't worth supporting. AI agree that the newer computers in my FARM are way out producing the old one; but adding this minor feature to make working clunkers useful again would be nice.

Was there any bandwidth consideration in setting the minimum to 1000? I do not think so. I'll bet that it is arbitrary. So setting it arbitrarilly to 100 is a no brainer.

Dan

[runestar]

I guess you should follow the geek crowd of the D.C. scene a little more, Brian. =)

These are perfectly good machines that would otherwise be paperweights. =) We realize that $500 will pay for a machine that could run around in circles the older machines... of course if we all had an extra $500 or so we'd by moving faster up all the D.C. projects' charts. I have to agree with Dan's point though...

Why 1000? What's so special about that number. What's so special about 10,000 also. Assuming a person just uses the default settings, the client will turn in 5000 structures at a time currently. So there's not a problem there. For those with faster machines, its actually to our benefit to turn in larger amounts of structures since frequently turning them in wastes time that could be spent folding.

On the current protein, my T-Bird 850 can turn in over 8000 structures every 2 hours. There are probably people who can probably easily turn in the 10,000 cap in an hour. It seems that a code on the server end to check the turn-in time to make sure its not too slow or too fast would be better. If there is a "violation," so to speak, the server would send out an e-mail letting the person know to either adjust their settings lower or higher dependant on the problem.

Although we can see a problem with results coming in too often, there doesn't seem to be any important need for long-running results to come back... although obviously it'd have to be within the week with the current schedule.

Okays, enough rambling on... inserting two-cents* here... <clink, clink>

RS½
PBB

*plus substitute two of your highest local coin denominations if you don't have any cents. ;)

[danuitti]

Thanks for the ditto! :D

[Brian the Fist]

Ok, here's the short answer.

Maximum 10000 - so uploaded data files dont get too big - the bigger they are, the better the chance of it timing out while uploading/getting cut off.

Minimum 1000 - so malicious users cannot easily setup a bunch of machines to pound on our server.
Remember a structure file is uploaded with every dataset which is fixed size regardless of the size of the data set.

These were not arbitrary but were carefully discussed before changing.

[runestar]

Originally posted by Brian the Fist
Ok, here's the short answer.

Maximum 10000 - so uploaded data files dont get too big - the bigger they are, the better the chance of it timing out while uploading/getting cut off.

Minimum 1000 - so malicious users cannot easily setup a bunch of machines to pound on our server.
Remember a structure file is uploaded with every dataset which is fixed size regardless of the size of the data set.

These were not arbitrary but were carefully discussed before changing.

Okay, gotta admit that's fair. =)

Hey Dan, you gotta be fair in admiiting that they don't have control over the CASP5 proteins. They can be a bit more selective in their own proteins within the confines of the project.

Brian, one question though, when we let the server(?) decide how many structures (default or -s 999) the client will turn in at a time, how is that number (currently 5000) determined? Is it just a middle number between the lowest and highest possible, or is it rounded off from something else that was calculated, or ... ??


Our Thanks,

RuneStar½
PBB

[Brian the Fist]

If you choose to obtain it from the server, the default, then the number is chosen by ME.

[danuitti]

Of course, the malicious user can POUND OUT 10000, just as easily as 100, but I am not issuing a challenge. While waiting for the minimum to be set to 100, I will simply close and re-open my cranky old computer as often as I can to prevent loss.

Should I set an alarm clock for 4am?

[runestar]

Are you still trying to run two D.C. projects on it?

On the current protein look at my numbers for a P100:

Structures Per Second: 0.10
Structures Per Minute: 6.1
Structures Per Hour : 364
Structures Per Day : 8724

Note that I actually am able to use the extra RAM switch with the current protein, but even without I was still getting like at least 5000 per day with the current protein.

I can't believe your P75 is tremendously worse off than the P100. If you're running two D.C. projects, of course its going to be signficantly slower. Plus its WinDUD 95. Everybody knows you can't run more than two things on it....and Windows Explorer takes up one of those... and even then THAT doesn't run right. ;-)

RS½

[MAD-ness]

Use cron or task scheduler to start and stop the client at whatever interval you desire (time wise).

I don't like task scheduler, but nncronlite seems to work pretty good (I use it with XP, it emulates cron from *nix). To halt the client you simply delete the file "foldtrajlite.lock" This will safely and properly close the client.

Then use another cron command or whatever to relaunch the client.

If you are not good at this type of thing, someone could probably type out the commands to enter, if you provide your OS info, what program you are using (task scheduler, cron, etc.) and tell us how you launch the client normally. If you use the service, you will have to use the commands for halting a service, I don't know them specifically, but the resource kit of NT has stuff like netsvc, etc. that can be made to work.

Or so smart people tell me.

[danuitti]

RS½

Yes, with 32 MB RAM on my cranky old P133; Win98SE and 640X480 monitor, I am running S@H and DF; both Command Line Versions. I have not fiddled with switches to optimize DF, so it cranks out only 100 or so in a few hours.

Setting my alarm clock would be a good move.

We did have an unfortunate power interruption about an hour ago. I probably lost 50 or so units. But S@H picked up right where it left off. So you see my choice of D C projects when it comes to reliability.

MAD-ness - Windows XP... HA

Dan

[runestar]

Well, you can't blame the software entirely for trying to cram two D.C. projects into 32MB of RAM. SETI@Home recommends at LEAST 32MB of RAM to run it.

The checksumming admittingly could be better. I'd venture a guess of 75% of the time it can successfully recover from a crash.

There might be a way to manually recover the files by creating a checklist.txt file manually, but I have to experiment with that some and is contingent on there actually being something to recover from.

Best,

RS½

[tpdooley]

if you're trying to share DF with other DC tasks - and you're complaining about DF not allowing you to upload every 100 structures, then why don't you just split your old computers into two groups. Run DF on one group, and run the other DC task on the second group. DF should produce close to 1000 structures every 4-6 hours on lower end pentiums - as long as it's the only task running.

[danuitti]

tpdooley

The point is that I can! Yes, with a Pentium classic and 32 MB RAM, I CAN run DF and s@h command line clients at the same time.

But even if I run DF by itself, having an auto upload at 100 does really help. To the old dog, a loss of 100 structures is quite significant, a couple of hours of work.

and don't you get ticked off when you lose 5 minutes worth of typing!!!

Can you imagine losing a life saving structure, the fluke that we have been waiting so long to find?

Dan

[tquade]

Some performance data:

Dual Pentium 233, 256meg ram, NT4.0, 750 structures per hour
AMD K5 333, 64meg ram, Win95OSR2, 880 structures per hour
Dual PIII 1ghz, 1gig ram, W2K, 3900 structures per hour
Celeron 950mhz, 1gig ram, Win98SE, 3800 structures per hour

All above running current structure (size 69), structure in ram, client for 95 and 98, service for NT4.0 and W2K.

Ted

[tpdooley]

.. We can transport 100 highschoolers from the local high school to a beach at the end of the road in 2 seater sportscars - taking 100 cars. Or we could pack 'em all in a bus or two - waste less gas, have fewer problems parking, and reduce the number of drivers by... close to 100.. ;) Just because something >can< be done, doesn't mean that that's the way it >should< be done.. :)

A pentium 75Mhz (classic) running win98se, that was upgraded from win98, from win95, from Win 3.1.. currently running NAV in the background (eating up a bit of the 32 megs :) (a clean install should run better.. :) got around 4000 structures a day. (using dfGUI 1.71). so for this protein (69) on this P75, that's 36 minutes of labor.
About every 6 hours it would upload a batch of 1000 structures.
You were talking about 100 structures every 4 hours which would equal 600 structures a day. (sounds like S@H is getting more than 50% of the cpu time.. ) That's an upload of 1000 structures every 1.666 days.
If you run single DC projects on each of the machines - you'll get an upload every 6 hours or less on the ones running DF for this protein. And if you have problems with brownouts, they sell Uninteruptible Power Supplies. (They're almost a neccessity where I live).

[danuitti]

Yes, your figures are about right. The cranky ol' rig just plain locks up randomly, regardless of what is running on it. So I lose completed units every time.

So I have it set to 1000, and manually quit and restart every 4 to 8 hours. When it is locked up, I can see how many structures were lost. It has rarely accumulated more then about 800 when it is dead. Usually around 300. So I may be wasting some 500 structures per day on average. This is about 10 percent.

Was one of the lost structures the cure for cancer?

Dan

[tpdooley]

No.. none of your lost structures were the cure to cancer. None of the billions of structures we've uploaded have anything to do with cancer. We're given the building blocks of a protein structure, and we fold them. Then they're graded as to how close to the actual protein they came.
We're helping with a process that is allowing Howard to test out his team's ideas, and to produce close-to-properly folded proteins. We've been getting a few improvements to the folding algorhythm, which help get to the best structures faster. Hopefully a few more ideas will come to light, and the processing required will become even less - and the testing will be more accurate. (so the best scoring structures are the best fitting structures).
Once we've helped provide the proof of concept to this approach, then it can be used for the search for cures.
And when companies get to that stage, (and if they opt for a DC approach to help, vs running it in-house on a supercomputer or clustered servers..) one hopes that those running the project at least use modern systems that are fairly reliable.

If it will help, PM me with your address, and as soon as the Athlon 2400+ comes out, I'll send you 1.3Ghz Duron that's only been used for DF.. (as long as you promise to only run DF on the Duron.. :)

[runestar]

So, can we have you too for the team? ;) We promise you lots of geek points. =)

2400+? What no, 2600+?

BTW, how have has the Duron compared in performance the Thunderbirds and XP+?

Cheers,

RuneStar½

[tpdooley]

*snicker* I'm not teamless - I joined the team of the folks whose link to this project convinced me to join. (OCWorkbench with the handle "BennyRop").
And it's because of them that I'm running as many machines at work as possible, rather than just mine at home.

As for performance: a Duron 1.3Ghz cpu, with 256Megs 133mhz sdram (cl2), on an all in one motherboard from msi (lan, video, audio) with DF running with extra ram, and priority of 20, with nothing else running on the system - is getting around 66k structures a day from the 111AA structure. (based on 15 minutes of testing..) - it got played with last night, so I had to restart it.. :)

the same setup, with an Athlon XP 1700+ is getting 104k structures a day. (based on 24+ hours running).

And I'm actually looking for a cooler running cpu than my Athlon XP 1800+ from last year.. so not jumping up to a 2600+.

[runestar]

Hey, you offered the Duron... thought maybe we could get you too in the deal. ;)

Have you considered the Swiftech MCXC370 for the CPU? It's gotten pretty decent reviews on it... monster of a thing, but short of water cooling, that's going to keep that chip cooled down.

RS½

[tpdooley]

I offered the Duron to help Danutti replace the collection of old pentium 1 systems that he keeps mentioning are unreliable. (Whether it's unreliable hardware, corrupted windows, or problems running multiple DC projects - on systems that only have enough ram to run one or the other). Granted, he'll still use all the P1s to run the other project, unless this shows him how much easier it is to maintain a single computer that outperforms a farm of P1s, and he replaces a second P1 with a similar setup.
He'll still need an atx case, a motherboard, and ram. (floppy, cdrom, and hard drive can be stolen from an old machine). Reformat the hard drive, install whatever copy of windows he's using, and everything will run fine (and 10-20 times faster than the pentium 1 system he replaced). An Antec case can be picked up at www.insight.com (make sure the case has slots for case fans - 1 in the bottom front of the case, 1 below the power supply in the rear) They now run around $60? Find an all in one motherboard (to help eliminate problems with the old video card & nic) from a reliable name brand motherboard maker. I've had problems with Asus, so have moved on to MSI. The one I got was about $80 but it uses sdram, not DDR. (look for them on places like www.pricewatch.com and www.insight.com). And then grab CL2 memory that the motherboard supports. (133SDRam or 266 DDRam. I opted for www.crucial.com - as I've been tired of doing hardware testing for the unknown memory firms that we've dealt with since '88. Crucial/Kingston/Viking are all reliable, easy to find memory board companies that I'd be comfortable with. A 256M pc133 SDRam cost about $50.
------
While Cost Is No Issue with my home computer, I'm a bit frugal with replacing machines at work.. *snicker* When we start running even larger protein structures again, I'll be forced to get my work systems 2 more sticks of 256Meg. :)

--------
As for a cooler CPU - I had a T-Bird 900Mhz cpu on an Asus A7V series motherboard. The T-bird came bundled with an "orb" hsf. The CPU ran hot. I wouldn't have cared much, but it was really keeping my room hot. I replaced the fan with a MasterCooler HSF; but the cpu didn't cool down as much as I expected.
I found out that the motherboard was feeding the cpu 1.95v instead of the 1.75v it needed. (I'd been foolish enough to leave it in jumperfree mode.. and had to jumper it to around 1.575 volts to get it to actually produce 1.75v for the cpu.. ;)
(and replaced the motherboard soonafter, although they both went to heaven after a few months.. poor used and abused thing :)

After reading the following reviews, I went with the Noisecontrol Silverado. I like it as quiet as possible.. :)
http://www6.tomshardware.com/cpu/01q2/010521/index.html
http://www6.tomshardware.com/cpu/01q1/010306/index.html
http://www6.tomshardware.com/cpu/00q4/001211/index.html

and visiting AMD's site showed that the case I was using wasn't setup properly for an Athlon. (great for k6-3s, though. ;)
http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/23794.pdf (note page 27.. although all cases today probably have a case fan slot/hole setup beside the cpu now).

.. and after all that work to keep my cpu cool, and my room cool, I'll stick to replacement cpus that produce the same or lesser amounts of heat than my 1 year old Athlon XP 1800+. :)

[runestar]

I think his problem may mainly stem from Windows 95, although it could be in part the system too.

I liked the Silverado, in fact I saw those articles too, but I can't seem to find them. The local shop where I am planning to get several parts for my new system has the Thermaltakes which as you found although it looks really cool and impressive, leaves quite a bit to be desired... They also have the Swiftech MCX(C) 370 was one of the highest rated, if not highest...I don't remember off-hand.

If I could afford, I'd go for one of their water-cooling solutions, but I'm not a hard OC'er and the money could be spent for bettering my new system. =)

If Dan doesn't want the Duron, may I have it? =) I can use it for a spare machine and turn off the Pentiums. I've meaning to try out Linux again. I can make a nice little Linux box out of that.

As for older Athlons, I have a K7V motehrboard which originally had a 750 (was supposed to be 700 but the order was late so they upgraded me fo r free) with a custom HS/Fan solution. Unfortunately the company went out of business it seems when that died, I got a replacement T-Bird 850 though. Supposedly the K7M support up to 1Ghz T-bird, and since the K7V was the immediate successor to that board I figured it should be able to handle it just fine which it did without a hitch.

Well, that's all for now...TTFN!

RuneStar½

[danuitti]

The long awaited description of my gumpy ol' computer.

Yes, it is an Intel Motherboard of '96ish vintage with a real Pentium 133 chip. Due to capacitance leakage and such, the motherboard is reaching the end of life.

It sports a fairly old, but functional 2GB hard drive. i believe that it is a quantum. I am sure that it will outlive the other components.

The battery circuit on the MB has been dead of about a year now. Replacing the battery does not help.

Yes, I have only 32MB ram in the rig. I had put the other 32MB to good use about a year ago. My nephew needed it in his sound box; where he creates sound.

The Nic card is a PCI vintage thing, and will find a new home when the cranky ol' rig finally gives up. The video card is also an old vingage, but might provide further life somewhere.

About 18 months ago, I installed Windows 98SE from scratch. I only installed the minimum, since I did not want to see any further unstable activity. With the new and minimum configuration, it is a hardware problem of sorts that causes it to LOCK UP. Memory and Video memory - who knows. Certainly not worth troubleshooting. hey, it runs between 10 and 40 hours between crashes.

The ol' AT case is also aging, but is still functioning. It is a '94 vingate, starting life with a blazing 486 DX 100.

So when "end of life comes", it will be to the BONE yard for most of it. Spending even a penny is not what I have in mind.

Dan