Reduce autosave time in Prime95

Fraxo0 · 2021-04-19, 14:22

I am new to Mersenne numbers research, and I was reading the "readme.txt" file in the section "Warning and Notes".
It mentions an autosave feature in case of a power outage every 30 minutes, which looks too much.
I was wondering if it's possible to set that value to something like 10 or 5 minutes.
Thanks in advance!

Uncwilly · 2021-04-19, 15:03

Welcome to the forum and the search for huge primes.

If your system is and power is normally stable, every 30 minutes is a reasonable choice. Since stopping to write a save file takes away a little time, that little bit of time is not used to test a number. Over the long run that little bit ads up. If your computer is not stable, that is an issue that you need to fix. If the power where you are is not stable, a battery back up is a really good idea.

If you still want to change that value, you can do so from the menu: Options -> Preferences

axn · 2021-04-19, 17:17

Quote:

Originally Posted by Uncwilly

Since stopping to write a save file takes away a little time, that little bit of time is not used to test a number. Over the long run that little bit ads up.

I don't think P95 actually stops, but rather does the writes on another thread. I might be wrong about this.

Uncwilly · 2021-04-19, 18:13

Quote:

Originally Posted by axn

I don't think P95 actually stops, but rather does the writes on another thread. I might be wrong about this.

Sure, but if you have all cores running a test, and one is diverted to handle the saving, that one has stopped helping with the test.

slandrum · 2021-04-19, 20:48

The environment where it's most useful to save more frequently is when you are running on machines that will suddenly and with no warning have power cut to them - say you are running on a free cloud computing service with arbitrary cutoffs that change every day with no warning where you don't know if it will run for one hour or 20 when you start it up.

LaurV · 2021-04-20, 05:17

Say you do one test (or more) that takes (single or all together) about 10 days. Say you save backups every 30 minutes and you get 4 crashes in those 10 days, after which you need to resume, wasting in average one hour (because the crash can come one minute after saving a checkpoint or 29 minutes after, just before saving a new one, you get in average 15 minutes lost per crash, ignoring the time to restart, because that is always the same, regardless of the saving interval). So, in this example scenario, you will lose about 1 hour of work, every 10 days, plus the 2*24*10*(x%)*t time lost to write the checkpoints, assuming your computer is x% slower during the checkpoint writing (you need to delete the old file, rename the current file to bak, write the new file, this when you only keep a history of two checkpoints, a current one and a backup one**), and you wrote 2 backups every hour for 10 days, and writing every backup takes t time.

Say you save backups every 10 minutes and you get the same 4 crashes in those 10 days, you waste 20 minutes of testing (the average 5 minutes lost per restart, ignoring the restarting time, as said, time 4 crashes). So, in this example scenario, you will lose about 20 minutes of work every 10 days, plus the 6*24*10*(x%)*t time lost to write the checkpoints, because now you save 6 backups every hour. So, it all boils down to the question if the difference of 4*24*10*(x%)*t is larger than 40 minutes or not. You may be surprised...

**) and if you argue about the necessity of keeping two backup files, well... you don't need to, but in that case, saving more often is even more dangerous, because the system spends more time writing backups, so in theory the probability of a crash to happen EXACTLY at the time you write the files (and damage the file, and then you lose EVERYTHING and will need to start from scratch) is higher. P95 is kinda "robust" in this respect, it keeps backup files and save/rename them in a "safe" way, so you won't lose your work if sh!t happens exactly in the moment you write into the file.

We have saving checkpoints every 4 to 6 hours. This works fine and doesn't kill our HDDs either. In the past, before the PRP/GEC era, when I was running two LLs in the same time, to compare the residues, and make sure the GPUs stay sane, I was losing more time rerunning last iterations when they didn't match, than I was losing reverting to checkpoints due to crashes.

Of course, YMMV.

retina · 2021-04-20, 05:28

Quote:

Originally Posted by LaurV

Say ... you get 4 crashes in those 10 days ...

... then fix your system.

It is no fun running wonky a system.

Fraxo0 · 2021-04-20, 06:15

Thank you all.
I found the setting in the "prime.txt" file, there's the variable "DiskWriteTime=x", which by default the x is 30, but, by changing that, it will be possible to change the autosave time parameter.

2021-04-19, 14:22	#1
Fraxo0 "Francesco Vincenzi" Apr 2021 Torino, Italy 2 Posts	Reduce autosave time in Prime95 I am new to Mersenne numbers research, and I was reading the "readme.txt" file in the section "Warning and Notes". It mentions an autosave feature in case of a power outage every 30 minutes, which looks too much. I was wondering if it's possible to set that value to something like 10 or 5 minutes. Thanks in advance!

2021-04-20, 05:17	#6
LaurV Romulan Interpreter "name field" Jun 2011 Thailand 41×251 Posts	Say you do one test (or more) that takes (single or all together) about 10 days. Say you save backups every 30 minutes and you get 4 crashes in those 10 days, after which you need to resume, wasting in average one hour (because the crash can come one minute after saving a checkpoint or 29 minutes after, just before saving a new one, you get in average 15 minutes lost per crash, ignoring the time to restart, because that is always the same, regardless of the saving interval). So, in this example scenario, you will lose about 1 hour of work, every 10 days, plus the 22410(x%)t time lost to write the checkpoints, assuming your computer is x% slower during the checkpoint writing (you need to delete the old file, rename the current file to bak, write the new file, this when you only keep a history of two checkpoints, a current one and a backup one*), and you wrote 2 backups every hour for 10 days, and writing every backup takes t time. Say you save backups every 10 minutes and you get the same 4 crashes in those 10 days, you waste 20 minutes of testing (the average 5 minutes lost per restart, ignoring the restarting time, as said, time 4 crashes). So, in this example scenario, you will lose about 20 minutes of work every 10 days, plus the 62410(x%)t time lost to write the checkpoints, because now you save 6 backups every hour. So, it all boils down to the question if the difference of 42410(x%)t is larger than 40 minutes or not. You may be surprised... ) and if you argue about the necessity of keeping two backup files, well... you don't need to, but in that case, saving more often is even more dangerous, because the system spends more time writing backups, so in theory the probability of a crash to happen EXACTLY at the time you write the files (and damage the file, and then you lose EVERYTHING and will need to start from scratch) is higher. P95 is kinda "robust" in this respect, it keeps backup files and save/rename them in a "safe" way, so you won't lose your work if sh!t happens exactly in the moment you write into the file. We have saving checkpoints every 4 to 6 hours. This works fine and doesn't kill our HDDs either. In the past, before the PRP/GEC era, when I was running two LLs in the same time, to compare the residues, and make sure the GPUs stay sane, I was losing more time rerunning last iterations when they didn't match, than I was losing reverting to checkpoints due to crashes. Of course, YMMV. Last fiddled with by LaurV on 2021-04-20 at 05:24*

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
First time using Prime95, looking for advice for setup	dka71	Software	10	2016-04-26 07:07
More efficient to reduce worker count?	CuriousKit	Hardware	21	2015-10-24 03:40
How to reduce number of worker windows?	Chuck	PrimeNet	7	2011-07-03 19:17
Reduce your debt!! ... I'm curious.	petrw1	Lounge	59	2009-01-21 12:48
Any way to reduce CPU usage?	Jarl	Hardware	5	2007-03-30 19:13

2021-04-19, 15:03	#2
Uncwilly 6809 > 6502 """"""""""""""""""" Aug 2003 101×103 Posts 10101100111011₂ Posts	Welcome to the forum and the search for huge primes. If your system is and power is normally stable, every 30 minutes is a reasonable choice. Since stopping to write a save file takes away a little time, that little bit of time is not used to test a number. Over the long run that little bit ads up. If your computer is not stable, that is an issue that you need to fix. If the power where you are is not stable, a battery back up is a really good idea. If you still want to change that value, you can do so from the menu: Options -> Preferences

2021-04-19, 20:48	#5
slandrum Jan 2021 California 1000110000₂ Posts	The environment where it's most useful to save more frequently is when you are running on machines that will suddenly and with no warning have power cut to them - say you are running on a free cloud computing service with arbitrary cutoffs that change every day with no warning where you don't know if it will run for one hour or 20 when you start it up.

2021-04-20, 06:15	#8
Fraxo0 "Francesco Vincenzi" Apr 2021 Torino, Italy 2 Posts	Thank you all. I found the setting in the "prime.txt" file, there's the variable "DiskWriteTime=x", which by default the x is 30, but, by changing that, it will be possible to change the autosave time parameter.