mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   PrimeNet (https://www.mersenneforum.org/forumdisplay.php?f=11)
-   -   Welcome Ben Delo (https://www.mersenneforum.org/showthread.php?t=24819)

Viliam Furik 2021-10-11 07:29

[QUOTE=tuckerkao;590145]Looks like 2 of Ben Delo's engines got assigned on the same exponent -> M[M]106097143[/M]. At least the format of the numerical arrangement looks like something I'd guess on earlier in the M103M range.

Since most of his engines are automatic, could the duplicates be visible on several other exponents as well?[/QUOTE]

I think it was Ben "poaching" himself, in the sense that the first machine that got assigned the exponent didn't work as supposed, so the assignment expired, but stayed in the worktodo file. The second machine, which got it assigned, completed the assignment, and then the first one completed it about a week later.

I think this is the most likely scenario.

Chuck 2021-10-23 19:57

Ben Delo AWS instances
 
I have never been able to understand exactly how AWS works. The explanations here on the forum are too complex for me to follow.

When an instance completes its PRP work, does that same instance name continue on to another exponent, or does a new instance start up? Do the instances run each job to completion, or can they expire in the middle of a job?

ATH 2021-10-23 20:04

Yes, the instance continues on the next exponent. You decide how long you rent the instance for.

If you use "spot" instances which is much cheaper, then the instance can shut down at any time when the prices goes over the limit you chose. But then you can set it so another instance starts automatically in another cheaper zone or later when the price is lower again and continue automatically.

You store the files on another volume that is separate from the instances going up and down, and you pay a little bit for that volume as well per GB per month.

chalsall 2021-10-23 23:08

[QUOTE=ATH;591455]Yes, the instance continues on the next exponent. You decide how long you rent the instance for.[/QUOTE]

Correct.

AWS (along with GCE and M$ Azure) offers some amazing inexpensive compute. But they aren't exactly easy to "make friends with" unless you are an experienced SysAdmin.

Even just getting past the ephemeral nature of the lowest-cost offerings takes some knowledge of mounting filesystems. Automatic (AKA programmatic) restarting of instances, by definition, requires some programming.

Chuck 2021-10-24 00:59

Do we know how many instances Ben runs? I looked at the "recent results" list, sorted by member name "Ben Delo" and computer name within member name, and I counted 64 distinct instance names.

chalsall 2021-10-24 01:26

[QUOTE=Chuck;591466]Do we know how many instances Ben runs?[/QUOTE]

We do not. If he's doing what I suspect he is doing, it will be a function of the "spot price" at any particular moment in time.

[QUOTE=Chuck;591466]I looked at the "recent results" list, sorted by member name "Ben Delo" and computer name within member name, and I counted 64 distinct instance names.[/QUOTE]

That seems logical. As in, he has a maximum of 64 instances, which he has code which "spins up" as many as makes sense (economic curves, and all of that).

AWS is quite amazing. And, when you get down into the weeds, quite complex.

P.S. Please don't even get me started on regional VPCs, 10/8 vs 172.16/12, dual-NICs, subnetting, and internal routing... :wink:

ATH 2021-10-24 04:47

See post #11 in this thread. Back then in late 2019 he said he was running 50 instances of c5.metal, which is 48 cores Xeons, so 2400 cores total. Maybe its 64 x 48 cores now, or 64 other instances.

c5.metal or c5d.metal cost in US-east-2 right now $0.92/hour for a spot instance, or $4.08 / hour for on-demand.

Ben Delo 2022-02-02 16:08

I am sorry to say that after 93,997,250 GHz-Days and 252,205 first primality attempts my AWS Credits have run out. I have had to scale down to 10% of my original power until I get another deal with a cloud computing provider. Watch this space!

firejuggler 2022-02-02 16:16

Thank you for all the work you done.

mrh 2022-02-02 17:55

[QUOTE=Ben Delo;599273]I am sorry to say that after 93,997,250 GHz-Days and 252,205 first primality attempts my AWS Credits have run out. I have had to scale down to 10% of my original power until I get another deal with a cloud computing provider. Watch this space![/QUOTE]

Wow, that is impressive!

kriesel 2022-02-02 20:48

[QUOTE=mrh;599278]impressive![/QUOTE]Roughly comparable in magnitude to DC of everything that's now waiting for DC. In the course of that a lot of PRP certs got run also.


All times are UTC. The time now is 07:43.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2022, Jelsoft Enterprises Ltd.