Resources (Q&A and Errata)
Q&A
- Q: Why the PS3?
A: In short, the Cell Processor ‘packs a punch’. One of the authors (Khanna) estimates that his MPI computations run much faster than on desktop workstation chipsets, and that his original 8 PS3 (i.e. 64 core) Cell cluster had comparable if not better performance to a 200 Node IBM Blue Gene system.
- Q: Is there another way to install Linux, MPI, and the IBM SDK to build a Cell based cluster?
A: Yes, our friends at Fixstars (formally Terrasoft) have Yellow Dog Linux (YDL). YDL is an RHEL/CentOS Linux derivative, available on DVD or for free from the public mirrors. Meanwhile, Y-HPC is a commercial cluster construction suite allowing one to build an entire PS3 cluster. You can find more information about it here. Samir Gartner kindly donated his step-by-step guide to build a cluster using PS3s running YDL 6.1. You can download it here.
- Q: Only seeing 2 cores/penguins not the 8 that you show?
A: Penguins show up two times in boot sequence -- the first time you see 2 penguins; in the second round, you should see 2 + 6 total. Do you? Another way is to type -- ls /sys/devices/system/spu/ ...this will show you a separate directory for each physical SPU
- Q: Everything seems to be installed and working locally, but I can't seem to get the cluster to return from all nodes.
A: The number of processes fed to mpirun (say a default of 16) is the number of slots used by the cluster at runtime. This number seems to have to be equal to or greater than the number of allocated slots in the openmpi-default-hostfile (or things seem to get out of sync). For the PS3/Cell this includes a process for each of the cores in the processor.
- Q: How do I boot into my PS3 gaming environment and or how do I uninstall Linux from my PS3?
A: Go to the PS3 game system by typing "boot-game-os" (Fedora 6) or "ps3-boot-game-os" (Fedora 8). When you are back to the default PS3 environment, in "Settings/System-Settings", go to "Format utility". Format the disk, giving the entire drive space back to the PS3 environment. Meanwhile, a hard reset is performed by holding the PS3 power button down until it beeps 3 times then turning the rear power switch off after normal shutdown. This resets the system to the game OS.
- Q: Is it really that expensive to run simulations on supercomputers?
A: Commercially and privately yes. But if you are an academic, it should be noted that academic researchers in the U.S. can apply for free resource allocations on TeraGrid (www.teragrid.org). Sponsored by the National Science Foundation’s Office of Cyberinfrastructure, TeraGrid features more than a petaflop of computing capability and 30 petabytes of data storage that are accessible via high-performance networks. Extensive documentation, consulting, and training in support of research and education in all fields of study are available.
- Q: If you're a free site and research group, why do you guys have/had Google Ads?
A: Though this is a free guide, obviously HPC research is an expensive effort in time and hardware. All income from the site will go exclusively to our SuperComputing /Clustering efforts at the University of Massachusetts: Dartmouth. That said, the ads are currently inactive (due to internal changes).
Errata (Links)
PS3 Information: http://en.wikipedia.org/wiki/PlayStation_3Cell Information: http://en.wikipedia.org/wiki/Cell_(microprocessor)
Cell Kernel: http://www.kernel.org/pub/linux/kernel/people/geoff/cell/ps3-linux-docs/ps3-linux-docs-08.06.09/
Cell SDK/Docs http://www.ibm.com/developerworks/power/cell/
Cell video courseware: http://www.cc.gatech.edu/~bader/CellProgramming.html
PS3/Linux Memory Management (Thanks to Michael Farrar): http://www.ibm.com/developerworks/linux/library/l-linux-ps3-2/index.html?S_TACT=105AGX03&S_CMP=EDU