If ever I doubted that MPI was good for the world, I think that all I would need to do is remind myself of this commit that I made into the Open MPI source code repository today. It was a single-character change — changing a 0 to a 1. But the commit log message was Tolstoyian in length:
- 87 lines of text
- 736 words
- 4225 characters
Go ahead — read the commit message. I double-dog dare you.
That tome of a commit message both represents several months of on-and-off work on a single bug, and details the hard-won knowledge that was required to understand why changing a 0 to a 1 fixed a bug.
Read More »
Tags: HPC, mpi, why-mpi-is-good-for-you
This has been posted elsewhere, but it’s worth mentioning here to both because iWarp and InfiniBand are popular HPC interconnects, and to get as wide an audience as possible: the OpenFabrics Association is hosting its annual workshop in Sonoma on March 14-17, 2010.
The theme of this year’s Sonoma Workshop is “Exascale to Enterprise.” A preliminary agenda is available for your browsing pleasure.
Yes, I’ll be there. Will you?
Read More »
What to do about the implicit Fortran MPI interfaces (i.e., mpif.h) in MPI-3? This is something that I’ve been thinking about a lot recently.
Sidenote: Some people refer to mpif.h as “the Fortran 77 MPI interfaces.” That isn’t quite correct; there’s actually stuff in mpif.h that didn’t exist until well beyond the Fortran 77 specification, such as KIND attributes and whatnot. So if someone calls mpif.h “the Fortran 77 MPI interfaces”, you have my permission to give them a slugbug punch. Ditto if they call an “MPI process” a “rank.”
As I’ve mentioned in prior entries, we’re going to have much-updated explicit Fortran interfaces in MPI-3 (the so-called “Fortran ’03 interfaces”, but just like “the Fortran 77 interfaces”, that name isn’t quite accurate, either). As I swear I heard Snoop Dog say once, “These new Fortran explicit MPI interfaces are da fa-schizzle”. They offer a bunch of language features that MPI ignored before, and also fix some long-standing problems — most importantly with regards to asynchronous buffer control.
So the question isn’t so much about what to do for the future; it’s more a question about what to do with the past. Should we deprecate the old, ancient, decrepit, harmful, stanky, nasty Fortran implicit interfaces? The answer is not quite as obvious as I would hope.
Read More »
We recorded an RCE podcast earlier today talking to Tiki L. Suarez-Brown, Ph.D, and Hai Ah Nam, Ph.D, the two co-chairs of the SC10 student cluster competition, and Doug Smith, the faculty sponsor of the Colorado cluster competition team from the past few years. Scheduling to get everyone together for the recording was a bit dicey; the recording will likely be available a little later than usual (Brock usually releases recordings on Saturdays — this one will likely be out early next week).
The competition is no cake walk: teams of students get a very specific power budget (26 amps) to run a whole schlew of real-world HPC applications within a limited time frame. The teams are graded on several metrics, to include the highest Linpak number, most computational work processed in the time allotted, a question-and-answer interview, etc.
Fun fact: 26 amps is about how much you need to run 3 coffee makers.
Read More »
I just ran across a great blog entry about SGE debuting topology-aware scheduling. Dan Templeton does a great job of describing the need for processor topology-aware job scheduling within a server. Many MPI jobs fit exactly within his description of applications that have “serious resource needs” — they typically require lots of CPU and/or network (or other I/O). Hence, scheduling an MPI job intelligently across not only the network, but also across the network and resources inside the server, is pretty darn important. It’s all about location, location, location!
Particularly as core counts in individual server are going up.
Particularly as networks get more complicated inside individual servers.
Particularly if heterogeneous computing inside a single server becomes popular.
Particularly as resources are now pretty much guaranteed to be non-uniform within an individual server.
These are exactly the reasons that, even though I’m a network middleware developer, I spend time with server-specific projects like hwloc — you really have to take a holistic approach in order to maximize performance.
Read More »
Tags: HPC, hwloc, mpi, NUMA, NUNA