Rejecting/Avoiding VLAR on an Intel i486 DX4-S/100?


log in

Advanced search

Message boards : Number crunching : Rejecting/Avoiding VLAR on an Intel i486 DX4-S/100?

1 · 2 · Next
Author Message
Profile Grand Admiral Thrawn
Avatar
Send message
Joined: 19 Feb 01
Posts: 53
Credit: 22,263,601
RAC: 0
Austria
Message 1350550 - Posted: 25 Mar 2013, 17:20:19 UTC

Greetings!

I just tried to bring my i486 to run SAH, and to my surprise the project still has generic x86 binaries (386 i guess) for ancient CPUs on Windows. So far so good. Didn't wanna go through the pain of installing Debian 4.0 again anyways.

But I'm getting VLAR WUs, which I fear the machine cannot complete within the deadline, is there any easy way to reject VLAR WU's, like there is for AstroPulse, which I can just deactivate in the profile?

[Here is the machine]

Specs:

ASUS PCI-I486SP3G mainboard with 256kB L2 cache and the large 32k Tag RAM (which means 64MB main memory are cacheable, the maximum!)
Intel SL-enhanced 486 DX4-S/100, 100MHz but unfortunately write-through L1 cache, not write back
128MB (!) EDO-DRAM
AHA-29160 PCI SCSI controller with a 10.000rpm Disk for fast swapping where needed
Win98 SE

It estimates the currently running VLAR at around 1600 hours which is just a bit too long. Or will the ETA shrink over time?!

If not, I'll need a way to drop or reject VLARs (Which I hope can be done without having to make it an "anonymous platform" via app_info.xml).

Thanks!
____________
3dfx Voodoo5 6000 AGP HiNT Rev.A 3700 prototype, dead HiNT bridge

rob smith
Volunteer moderator
Send message
Joined: 7 Mar 03
Posts: 7663
Credit: 44,671,697
RAC: 75,098
United Kingdom
Message 1350566 - Posted: 25 Mar 2013, 17:31:39 UTC

ETA will shrink.
No that I know of stopping your CPU getting VLARs
____________
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Profile William
Volunteer tester
Avatar
Send message
Joined: 14 Feb 13
Posts: 1517
Credit: 6,882,305
RAC: 27,120
Message 1350586 - Posted: 25 Mar 2013, 18:01:05 UTC

There is no way not to get VLAR. Anonymous platform only means you are using your own app, nothing about task choice (within the MB subset).

On CPU VLAR don't take much longer than mid AR.

Estimates for new hosts are overcautious. Let it run for a while and extrapolate runtime from there - that's your upper limit, the app accelerates.

But honestly, the deadline is a function of the AR - if you're tight for deadline with VLAR you will be equally tight at other ARs

I'd let it run for a day or two and see how it gets on.
____________
A person who won't read has no advantage over one who can't read. (Mark Twain)

Richard Haselgrove
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8275
Credit: 44,924,001
RAC: 13,574
United Kingdom
Message 1350602 - Posted: 25 Mar 2013, 18:21:06 UTC
Last modified: 25 Mar 2013, 18:35:51 UTC

We had a user a few years ago who completed a task on a P60 in just under three weeks:

http://setiathome.berkeley.edu/forum_thread.php?id=39157&postid=634780

Since then the workunit running time has been doubled by increasing the sensitivity of the search. Unless your 486 is a DX2 model, I think you may very well over-run the deadline - and that will be for all angle ranges, not just VLAR.

Edit 1: re-reading the thread title, I see you've already said it's a DX4 - but that may not be enough, because...

Edit 2: mention of the P60 led me to Pentium launched 20 years ago (22 March 1993). Apparently the P66 benchmarked at almost twice the speed of the 486DX2/66.

Profile Grand Admiral Thrawn
Avatar
Send message
Joined: 19 Feb 01
Posts: 53
Credit: 22,263,601
RAC: 0
Austria
Message 1350616 - Posted: 25 Mar 2013, 18:31:27 UTC - in response to Message 1350602.
Last modified: 25 Mar 2013, 18:35:41 UTC

It's a DX4 with write through cache and 100MHz. Not the very fastest there is, but definitely faster than any DX2.

We'll see. I was just thinking that VLAR will take more time because it also took more time on my other CPU hosts that i recently restarted SETI on (I have been inactive for a longer period of time, and am now resuming work). But maybe I was wrong about that.

I will keep an eye on the runtime estimate, if it decreases over time it might work out after all.
____________
3dfx Voodoo5 6000 AGP HiNT Rev.A 3700 prototype, dead HiNT bridge

Josef W. Segur
Volunteer developer
Volunteer tester
Send message
Joined: 30 Oct 99
Posts: 4134
Credit: 1,003,514
RAC: 235
United States
Message 1350786 - Posted: 26 Mar 2013, 4:45:25 UTC - in response to Message 1350616.

It's a DX4 with write through cache and 100MHz. Not the very fastest there is, but definitely faster than any DX2.

We'll see. I was just thinking that VLAR will take more time because it also took more time on my other CPU hosts that i recently restarted SETI on (I have been inactive for a longer period of time, and am now resuming work). But maybe I was wrong about that.

I will keep an eye on the runtime estimate, if it decreases over time it might work out after all.

I ran a 200 MHz. Pentium system until a couple of years ago, and it was able to do work in slightly less than 1/4 the time allowed by the deadline. I think your 486 will probably finish within deadline.

Don't be discouraged by the estimated time to completion coming down slowly at first. It's mostly based on the progress and the original estimate at the beginning, so at 10% progress shown it will also show very nearly 90% of the original estimate.

As progress grows, the weight given to elapsed time is increased so that when the indicated progress is 50% about 3/4 of the remaining time estimate in BOINC 6.6.38 is based on how much time it has used already. Unfortunately, the progress provided to BOINC grows too slowly during the first part so the time to completion is always an overestimate. But both the first and later sections are fairly linear, there's just a change in slope. For a VLAR, the indicated and actual progress are about:

Indicated Actual 5% 8.5% 10% 17% 20% 34% 30% 51% 40% 58% 50% 65% 60% 72% 70% 79% 80% 86% 90% 93%


That's based on data from hosts I've run, I'd expect the 486 to be similar but it might respond somewhat differently.
Joe

Profile Grand Admiral Thrawn
Avatar
Send message
Joined: 19 Feb 01
Posts: 53
Credit: 22,263,601
RAC: 0
Austria
Message 1350811 - Posted: 26 Mar 2013, 8:44:23 UTC - in response to Message 1350786.

Aah, interesting, thanks for providing some insight. Currently it says 1630 hours remaining, which is roughly 68 days. Deadline ends at May 17th, which is why I was a bit concerned, only 52 days left.

But I'll let it run anyway, even if it's past the deadline at the end. Just want to have this box finish at least one WU!
____________
3dfx Voodoo5 6000 AGP HiNT Rev.A 3700 prototype, dead HiNT bridge

rob smith
Volunteer moderator
Send message
Joined: 7 Mar 03
Posts: 7663
Credit: 44,671,697
RAC: 75,098
United Kingdom
Message 1350887 - Posted: 26 Mar 2013, 18:03:13 UTC

Good luck, and good hunting :-)
____________
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Profile j mercer
Avatar
Send message
Joined: 3 Jun 99
Posts: 1405
Credit: 11,608,614
RAC: 66
United States
Message 1350946 - Posted: 26 Mar 2013, 20:43:36 UTC - in response to Message 1350811.

I still run my original/first cruncher; a 1995 200MHz Pentium MMx P55 128MB UW2SCSI rig every so often (1 WU for each new version of BOINC, costs to much to run). The last WU was around 11 days running bare bones XP with everything turned off including Explorer. I have to use Ctrl/Alt/Del to open Task Man to start Explorer again.

I think you will make it. Keep trying for awhile and don't give up on just one failure.

0ld SETI crunchers ROCK!
____________
...

Profile Grand Admiral Thrawn
Avatar
Send message
Joined: 19 Feb 01
Posts: 53
Credit: 22,263,601
RAC: 0
Austria
Message 1351455 - Posted: 28 Mar 2013, 11:59:36 UTC

Actually, the ETA is getting higher and higher each day, instead of lower, lol.

I am not sure, but maybe the OS is also degrading a lot with runtime (it's Win98SE after all).


____________
3dfx Voodoo5 6000 AGP HiNT Rev.A 3700 prototype, dead HiNT bridge

Profile Grand Admiral Thrawn
Avatar
Send message
Joined: 19 Feb 01
Posts: 53
Credit: 22,263,601
RAC: 0
Austria
Message 1352703 - Posted: 1 Apr 2013, 9:03:14 UTC

Looking really bad.

It's now 159 hours into the VLAR WU, 3.47% done. ETA has risen and risen, now sitting at almost 1800 hours. This is looking bad even when considering Josef W. Segurs information.

I think there is no way it can finish before the deadline.
____________
3dfx Voodoo5 6000 AGP HiNT Rev.A 3700 prototype, dead HiNT bridge

rob smith
Volunteer moderator
Send message
Joined: 7 Mar 03
Posts: 7663
Credit: 44,671,697
RAC: 75,098
United Kingdom
Message 1352714 - Posted: 1 Apr 2013, 9:29:40 UTC

Grand Admiral Thrawn
Sir, we salute your efforts
____________
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Profile Grand Admiral Thrawn
Avatar
Send message
Joined: 19 Feb 01
Posts: 53
Credit: 22,263,601
RAC: 0
Austria
Message 1354616 - Posted: 8 Apr 2013, 9:47:46 UTC
Last modified: 8 Apr 2013, 9:48:16 UTC

ETA close to 1900 hours now, lol.

There is no way this is ever going to work, unless I get lucky with some super-short WU. Maybe rebooting Win98 would also help, cleaning up the memory, but I dunno..

Even if I would re-install Debian 4 on that box (which was an immeasurable pain to get to work), I doubt it would be any faster by custom-compiling the worker app. I mean, custom-compile for what SIMD extensions? There just are none! Not even MMX.

I'll let it continue, but it looks grim. ;)
____________
3dfx Voodoo5 6000 AGP HiNT Rev.A 3700 prototype, dead HiNT bridge

Glenn savill
Send message
Joined: 20 Aug 99
Posts: 1194
Credit: 780,865
RAC: 0
Australia
Message 1354622 - Posted: 8 Apr 2013, 10:33:23 UTC - in response to Message 1354616.

Hello you haven't said if you have turned the screen saver off or not .If it is active and you haven't set it to blank the screen that will slow it down heaps . Maybe this is why the time is going up instead of down.

Mind you it is a slow system good on you for trying but a 486 @66 MHz
N.A.S.A don't even need them chips now the shuttle is sadly no more
____________

Glenn savill
Send message
Joined: 20 Aug 99
Posts: 1194
Credit: 780,865
RAC: 0
Australia
Message 1354623 - Posted: 8 Apr 2013, 10:52:13 UTC

I know you said you turned everything off but i'm shore in win98se when bionic loads it sets screensaver to seti by default and doesn't set it to blank the screen

good luck
____________

Cheopis
Send message
Joined: 17 Sep 00
Posts: 139
Credit: 9,565,846
RAC: 8,510
United States
Message 1354766 - Posted: 8 Apr 2013, 19:27:02 UTC
Last modified: 8 Apr 2013, 19:28:26 UTC

The processing power jump going from the 486dx4 to the Pentium 1 (after they fixed the Pentium 1's math errors - oops) was pretty extreme if I remember right.

Something that you might want to try if the machine has a beefy amount of RAM for it's architecture is to set up a RAM drive. I'm not sure how useful that will be, honestly.

Also, on Windows 98, if I recall correctly, there were still effective memory manager programs designed to clean up the way that Windows allocated memory in order to allow high RAM usage applications to run faster.

Memturbo was a RAM defragmenter iirc and Memmaker was a startup application to build a memory allocation with as little waste as possible.

Also, check and see how often you are writing partial data to disk. If your machine is liberally using the drive for swap space, and you are writing to it every 60 seconds, you might be thrashing the drive causing slowdowns.

Final thing *I can think of* to try, if the machine has a USB connection, is a plugged in flash drive configured to act as a SSD swap drive? I do not know how well this might work on a 486, it's been way too many years :P

Profile Grand Admiral Thrawn
Avatar
Send message
Joined: 19 Feb 01
Posts: 53
Credit: 22,263,601
RAC: 0
Austria
Message 1354815 - Posted: 8 Apr 2013, 21:27:35 UTC
Last modified: 8 Apr 2013, 21:35:18 UTC

Machine's 100MHz and 128MB of EDO-DRAM with 64MB being cached by L2 (largest possible L2 cache address space with maximum size tag SRAM). The CPU's L1 cache is write-through though, not the slightly faster write-back variant.

There is no screensaver active, not the Win98 one and not the SETI one!

RAM drive won't do anything? I/O is not limiting the performance, so moving data from disk to RAM would just be a waste of RAM as I see it. From what I can tell it's just the CPU itself slowing everything down. RAM is still ok (quite full, but it's *not* swapping).

Also, when the drive is being used, it's not going to be a problem (U160 SCSI with 10.000rpm drive on modern PCI bus using an Adaptec AHA-29160N with Win32 drivers, not Dos16 ones). Of course, for swap it still sucks, as access time and IOPS/s limit stuff, but well, it does not seem to be swapping.

I guess the CPU just SUCKS. lol.

Edit: What's crap is that BOINC itself already consumes 10-30% of CPU on the system, leaving whatever rest there still is to SETI@Home itself. I mean, what is BOINC even doing there? pffh. ;)
____________
3dfx Voodoo5 6000 AGP HiNT Rev.A 3700 prototype, dead HiNT bridge

Profile Link
Avatar
Send message
Joined: 18 Sep 03
Posts: 813
Credit: 1,500,415
RAC: 437
Germany
Message 1354821 - Posted: 8 Apr 2013, 22:09:59 UTC - in response to Message 1354766.

Also, on Windows 98, if I recall correctly, there were still effective memory manager programs designed to clean up the way that Windows allocated memory in order to allow high RAM usage applications to run faster.

Memturbo was a RAM defragmenter iirc and Memmaker was a startup application to build a memory allocation with as little waste as possible.

All those (completely useless btw.) programs did, was requesting large amount of RAM, so that other programs were swaped out to the pagefile and had to be put back to RAM as soon as you needed them again. Yeah, that sounds very useful and effective...
____________
.

Josef W. Segur
Volunteer developer
Volunteer tester
Send message
Joined: 30 Oct 99
Posts: 4134
Credit: 1,003,514
RAC: 235
United States
Message 1354868 - Posted: 9 Apr 2013, 1:41:55 UTC - in response to Message 1354815.

Machine's 100MHz and 128MB of EDO-DRAM with 64MB being cached by L2 (largest possible L2 cache address space with maximum size tag SRAM). The CPU's L1 cache is write-through though, not the slightly faster write-back variant.
...

My 200 MHz. Pentium system had that same L2 cache limitation. A few times after a reboot I had run diagnostics or other applications before starting SETI processing, and the memory allocated to SETI was in the non-cached upper half causing very poor performance.

Edit: What's crap is that BOINC itself already consumes 10-30% of CPU on the system, leaving whatever rest there still is to SETI@Home itself. I mean, what is BOINC even doing there? pffh. ;)

When I transitioned that 200 MHz. system from Classic, I also noted that BOINC 4.45 was using more CPU than I had hoped. IIRC it was about 3 percent, and BOINC Manager about the same when it was on screen though dropped to about 1% when minimized to the tray. So although boinc.exe is a necessary burden, I only use BOINC Manager temporarily as needed, exit it completely when not.
Joe

Terror Australis
Volunteer tester
Send message
Joined: 14 Feb 04
Posts: 1626
Credit: 200,747,476
RAC: 69,671
Australia
Message 1354927 - Posted: 9 Apr 2013, 6:50:04 UTC - in response to Message 1354868.

When running BOINC on a P4 machine I built some years ago, I found that BOINC Manager and Windows Explorer were big CPU hogs.

I got around this by only having BOINC manager running when I needed it and going into Task Manager and closing Explorer. this mean't I lost the desktop but when it's needed you can call up Task Manager to restart it.

T.A.

1 · 2 · Next

Message boards : Number crunching : Rejecting/Avoiding VLAR on an Intel i486 DX4-S/100?

Copyright © 2014 University of California