14
$\begingroup$

Sorry if this question is too simple for the community but I really could not find an explanation. A detailed answer would be appreciated.

$\endgroup$
2
  • 2
    $\begingroup$ Where did you find "Mio"? I find it curious that such a modern field would use such an archaic/non-standard abbreviation. $\endgroup$
    – pipe
    Nov 11 '20 at 17:57
  • $\begingroup$ @pipe I believe it's the standard abbreviation in German. Although in German it would be slightly curious for them to use "CPUh" instead of "CPU-Stunden". $\endgroup$
    – hobbs
    Nov 12 '20 at 2:28
6
$\begingroup$

HPCs work by allowing you to run jobs on many computers with many CPUs in parallel. CPUh refers to how many CPUs are being used for how long (the h in CPUh stands for hours).

For instance, you may have a job that needs 64 CPUs and will run for a whole day. That would be 1536 CPU hours. The largest jobs I see on our (relatively small) HPC are 128 CPUs for about four weeks straight. That's about 86,000 CPU hours. Most of our jobs are much smaller, sometimes fractions of a CPUh.

Jobs that need in the millions (Mio) CPU hours are truly massive, and would likely be running on the largest supercomputers around. You might find those jobs in weather forecasting, particle physics, government-sponsored encryption breaking, and similar applications.

$\endgroup$
6
  • 1
    $\begingroup$ @user253751 Thanks for the comment, though I don't quite understand it! Kevin: +1 but I might point out that "million CPU hours" is usually used in the context of CPU allocations (or "grants") to a research group: It can be measured in the number of million CPU hours that the group is given for the following year. Some agencies allocate in "CPU years" instead but ORNL allocated in "millions of hours" with 1 million being the smallest grant. Also I'd say "government-sponsored encryption studies" is more accurate, or do you mean they actually try to find criminals or something? $\endgroup$ Nov 11 '20 at 18:58
  • $\begingroup$ @NikeDattani I was referring to attempting to brute-force or extensive dictionary attacks. Usually, only governments have deep enough pockets to use that with an HPC. $\endgroup$ Nov 12 '20 at 3:25
  • $\begingroup$ Why are they attacking encrypted information? Is it from their enemies, or from North Korea, or to fight cyber-crime? $\endgroup$ Nov 12 '20 at 3:49
  • $\begingroup$ @NikeDattani Obviously, most of this would be highly classified, so I can only speculate. Examples where this might (or might not) have come into play are: The FBI trying to decrypt the iPhones from the Riverside shooters a few years, trying to access data from Edward Snowden, or countries like Iran, Russia, China (or North Korea if they have access to an HPC) trying to break into whatever systems they may be interested in for their enemies. It may also involve pre-computing larger Rainbow tables than are currently available, or many other examples. Again, all of this is speculation, not fact. $\endgroup$ Nov 12 '20 at 4:42
  • $\begingroup$ @NikeDattani Most super computers are actually used for mundane scientific purposes, currently, like modeling. E.g. weather patterns, ballistics calculations, virus/protein folding, etc. IBM's Watson, for example, is currently used to study medical charts and detect things like cancer by looking at large data sets. Some supercomputers are used for cracking encryption, however. In World War II, for example, the British applied a supercomputer to the task of breaking German transmission encryption. The NSA, NRO, CIA, & DHS in the United States in particular probably do similar things regularly. $\endgroup$
    – TylerH
    Nov 12 '20 at 16:49
16
$\begingroup$

"Mio" refers to "Million" and "CPUh" is the abbreviation of CPU hours, also refered to as Core hours.

This is a good read if you need to know more about HPC.

This is an interesting reddit post on the same topic.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy