Re: How is parallelization implemented using MPI and Geant4?

Help: Forums, postings and threads

Forums, postings and threads

Click on the Forum title, e.g. on the "Forums by Category" page, to read a sequence of postings to the Forum and its threads all in one page. If you are only interested in one thread or the thread following a specific posting, click the thread or the posting, which takes you to a smaller page, which contains only the part you are interested in and may be easier to navigate.

Layout of the posting thread(s)

The posting threads can be shown with/without the text of the postings ("Inline depth"). In addition the depth of the thread can be controlled ("Outline depth"). For both cases several values are provided, which can be toggled. Note that the displayed values are those, which can be chosen, they do not correspond to the actual display status of the thread(s).

Add message

Add a new posting thread to this Forum

Show subscribers

Click "Show subscribers" to get the List of all subscribers, who receive e-mail from postings to this Forum, if they have not explictly unsubscribed from a posting thread. This includes:

Subscribers to the forum
Explicit subscribers to the posting or any of its ancestors
The author of the posting (if he has not unsubscribed)
The authors of any ancestor posting (if they have not unsubscribed)

Some browsers display old pages from the cache

Some browsers (e.g. Firefox) display a page from the cache even though the page has been updated. As a result, one may not get the latest messages in the forum's outline or the logon ID of the previous session appears on the upper right corner. If one suspects that he or she does not get the latest content of a page, try clearing up the personal data from the browser's cache as the following:

Tools -> Clear Private Data -> Clear Private Data (from browser Firefox)

Tools -> Delete Private Data (from browser Opera)

then reflesh the current page to get the latest content.

Details on "Inline depth" and "Outline depth"

Threads and the follow-up messages are numbered. There is one line per posting, normally starting with the posting number. The URLs of a posting and its responses reflect the depth of the follow-up postings. Watch the URLs to see how this is done. Depending on the chosen display options, the thread-display may be incomplete; the number of messages at the next level is, however, displayed, e.g. "…3 message(s)". You may not see all contributors unless you choose "All".

Messages are "chained" if there are only replies at the first level, i.e. 1/1.html, 1/1/1.html etc. In case of "chained" messages the message number is replaced by the icon and there is no indentation.

Inline: Display the subject line only or also the text of the posting(s); for the choice "All" the "Outline" choices are switched off.

Argument Default value Alternative Comment

1 0 1 no text / full text of posting

2 1 All text for level 1 only / text for All postings

Argument	Default value	Alternative		Comment
1	0	1		no text / full text of posting
2	1	All		text for level 1 only / text for All postings

Outline: Choose the depth of the posting thread, successive toggle controls provide increasing detail.

Argument Default value Alternative Comment

1 2 1 2 levels / 1 level (original posting)

2 3 2 3 levels / 2 levels

3 3 All 3 levels / all levels (all postings)

Argument	Default value	Alternative	Comment
1	2	1	2 levels / 1 level (original posting)
2	3	2	3 levels / 2 levels
3	3	All	3 levels / all levels (all postings)

> > In my program, G4beamline, I seed the random number generator with the event
> > number  
> This random generator seeding does sound reasonable. If you don't do
> that, you'll always get the same result running your application. It's
> all pseudo random after all. I however would use the thread number (if
> possible) to name the files.

Yes, one definitely need different seeds. Thread number is not guaranteed to be unique on multiple systems (e.g. a Linux cluster); first event number is (because we arranged it).

> However doing an:
> mpdboot -n 20
> mpiexec -n 20 mympiapp
> Is easier than writing a shell script.

Only after installing MPI on every machine. I use a shell script on my 4-CPU Mac Pro all the time:
    g4beamline input.file first=0 last=999999 >0.out 2>&1 &
    g4beamline input.file first=1000000 last=1999999 >1000000.out 2>&1 &
    g4beamline input.file first=2000000 last=2999999 >2000000.out 2>&1 &
    g4beamline input.file first=3000000 last=3999999 >3000000.out 2>&1 &
The input.file defines the simulation; it uses $first to name the output Root file and processes events from $first to $last ($ means parameter expansion).

> > This works for parallelization up to perhaps 32 jobs. Beyond that it gets
> > rather cumbersome; certainly 10,000 parallel jobs is not feasible this way.
> 
> Well 10.000 sounds hilarious.

There are a number of applications that need it. I expect that in a few years both CMS and ATLAS will scale to that level of parallelization. Perhaps not for simulations, and perhaps not for a single "job", but the need is there, and tool providers must be ready.