Get two free computers for a year! - Page 3

chalsall · 2013-02-07, 21:10

Quote:

Originally Posted by Batalov

Would you write a short and sweet step-by-step tutorial for the rest of us?

OK. Log into your EC2 interface. The default environments are WinBlows or Ubuntu Linux, but CentOS et al are also available.

Make sure your firewall settings allow port 22 to be open. It doesn't appear to be by default. Then create a key pair, and launch the instance.

Next, log in, and run the OS's update.

Then create a user for yourself. (Careful if you're running Linux -- regular user logins might not be enable.) And then load any software you might regularly use.

Then create a new AMI based on this instance. Be aware the virtual machine being imaged will be shut down during this process unless explicitly being told not to.

For the advanced users, a EBS volume might be created which can be mounted under an EC2 instance. Perhaps under /home/.

Batalov · 2013-02-07, 21:35

Both in win and linux instances, I've observed over a day of running that though DC processes can (and occasionally run) at, say, 0.029 s/iter, 99% of the time they run ~10x slower (wobbling between ~0.2 and 0.4 s/iter). At this rate even the smallest DC jobs at ~29M will take three months per job.

I am not sure why the program (or the host's OS) does that. I have thrown the niceness parameter into "10" (sadly, there wasn't an "11" setting). We'll need a non-apologetic version of P95 compiled - i.e. the version that doesn't renice and doesn't throttle on other processes (from other invisible users on the same physical node).

chalsall · 2013-02-07, 21:41

Quote:

Originally Posted by Batalov

I am not sure why the program (or the host's OS) does that.

What part of "virtual machine" isn't clear?

Batalov · 2013-02-07, 21:54

Well, I would imagine that they oversubscribe the physical nodes like the airlines do. In 2015, we will be lucky to get an upper ticket in a doubled-up seat on any airline. :-) Or lower position on the seat, if you don't mind the weight or rather, prefer a little weight vs something pressing against your lower back. There will be a surcharge for the non-default preference.

Anyway, that's what I meant by non-apologetic. mprime/Prime95 tells every process "Oh! Terribly sorry, Sir, I will wait. No, no, after you! I insist! This virtual node has no real load; please let everything else run first."

dleclair · 2013-02-08, 00:20

Amazon's t1.micro instances allow for short bursts of high CPU usage but anything longer is severely throttled.

They work passably well for hosting basic web sites and similar tasks but even simple operations like launching a decent text editor can trigger throttling.

For Windows instances, running Windows Update always triggers throttling turning it into an even more painfully long effort than usual.

I used a couple of t1.small instances for my business and they are much more usable but of course they come at a cost.

Don Leclair

chalsall · 2013-02-08, 00:28

Quote:

Originally Posted by dleclair

Amazon's t1.micro instances allow for short bursts of high CPU usage but anything longer is severely throttled.

I'm finding them very handy for mounting EBS volumes into which I rsync several thousand images for processing in larger virtual machines....

ET_ · 2013-02-08, 08:40

Quote:

Originally Posted by chalsall

Such as?

Higher costs and lower CPU usages.

I read the other posts, and I decided not to try: the virtual environment doesn't match fine with high requests.

I've been through this when ISPs offered free (but painfully slow) connections.

Luigi

Batalov · 2013-02-10, 01:50

Well, this idea works, even if the ETAs are wrong.
Let's see how good these t1.micros are after 3 months.

mbn451 · 2013-03-30, 03:03

So I tried this out to get some experience with EC2, and I have not been impressed with the results. A first time LL on a 62m exponent was not on track to finish within the year, with iteration times of ~.400 seconds each. So I switched to P-1 and I got a "broken pipe" error when trying to enter stage 2. Trial factoring is possible, but is essentially useless considering it is about 100x slower than a GPU.

I set it to try a Double-Check, maybe that will complete within a year...

BTW this was on a Linux micro, the windows one was much much slower and still hasn't finished stage 1 of P-1.

So essentially I feel as if logging into the instances to check them costs more ghz/hr to my desktop than I gain by having the instances.

Batalov · 2013-03-30, 03:37

Meanwhile my two instances didn't die yet. And I didn't visit them for two months, no sirrie.

mbn451 · 2013-03-30, 18:03

Quote:

Originally Posted by Batalov

Meanwhile my two instances didn't die yet. And I didn't visit them for two months, no sirrie.

Yeah I am thinking double checks might be the best use for these. I guess you could always transfer a partially completed first time LL...

2013-02-10, 01:50	#30
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 10010010101110₂ Posts	Well, this idea works, even if the ETAs are wrong. Let's see how good these t1.micros are after 3 months. Attached Thumbnails

2013-03-30, 03:37	#32
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 22256₈ Posts	Meanwhile my two instances didn't die yet. And I didn't visit them for two months, no sirrie. Attached Thumbnails

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Leave it to the computers - will you?	storflyt32	storflyt32	16	2013-04-03 16:39
Year Over Year TF Progress	petrw1	Factoring	3	2013-03-20 19:34
v4 computers	hrdubwd	Information & Answers	0	2013-03-17 14:08
V4 Computers	MurrayInfoSys	Information & Answers	3	2009-05-17 13:52
Getting more computers	Citrix	Prime Sierpinski Project	2	2005-09-07 13:04

2013-02-07, 21:35	#24
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 2×3×5×313 Posts	Both in win and linux instances, I've observed over a day of running that though DC processes can (and occasionally run) at, say, 0.029 s/iter, 99% of the time they run ~10x slower (wobbling between ~0.2 and 0.4 s/iter). At this rate even the smallest DC jobs at ~29M will take three months per job. I am not sure why the program (or the host's OS) does that. I have thrown the niceness parameter into "10" (sadly, there wasn't an "11" setting). We'll need a non-apologetic version of P95 compiled - i.e. the version that doesn't renice and doesn't throttle on other processes (from other invisible users on the same physical node).

2013-02-07, 21:54	#26
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 2·3·5·313 Posts	Well, I would imagine that they oversubscribe the physical nodes like the airlines do. In 2015, we will be lucky to get an upper ticket in a doubled-up seat on any airline. :-) Or lower position on the seat, if you don't mind the weight or rather, prefer a little weight vs something pressing against your lower back. There will be a surcharge for the non-default preference. Anyway, that's what I meant by non-apologetic. mprime/Prime95 tells every process "Oh! Terribly sorry, Sir, I will wait. No, no, after you! I insist! This virtual node has no real load; please let everything else run first."

2013-02-08, 00:20	#27
dleclair Mar 2003 115₈ Posts	Amazon's t1.micro instances allow for short bursts of high CPU usage but anything longer is severely throttled. They work passably well for hosting basic web sites and similar tasks but even simple operations like launching a decent text editor can trigger throttling. For Windows instances, running Windows Update always triggers throttling turning it into an even more painfully long effort than usual. I used a couple of t1.small instances for my business and they are much more usable but of course they come at a cost. Don Leclair

2013-03-30, 03:03	#31
mbn451 Mar 2013 2²·3 Posts	So I tried this out to get some experience with EC2, and I have not been impressed with the results. A first time LL on a 62m exponent was not on track to finish within the year, with iteration times of ~.400 seconds each. So I switched to P-1 and I got a "broken pipe" error when trying to enter stage 2. Trial factoring is possible, but is essentially useless considering it is about 100x slower than a GPU. I set it to try a Double-Check, maybe that will complete within a year... BTW this was on a Linux micro, the windows one was much much slower and still hasn't finished stage 1 of P-1. So essentially I feel as if logging into the instances to check them costs more ghz/hr to my desktop than I gain by having the instances.