I'd be concerned about "gumming up the works" with the glycol. Try taking a few drops of coolant from your car sometime and rub it between your fingers for a while. They'll eventually be coated with the glycol residue and will be pretty sticky. Water steam under pressure is VERY hot and I'd be concerned that the glycol would gum everything up that same residue.

Bringing a steam turbine online would be a pretty slow process and not well suited for quick trips to the store.

Not to nitpick, but an 80L tank holds 80L of compressed air at its maximum working pressure.

The tank's unpressurised internal volume is much smaller.

The article didn't explicitly state whether it was 14.4 gallons of compressed gas or not, so I have no idea how large it would be.

I just wouldn't want to be anywhere near if it if failed. I saw a photo of a car that had an old steel scuba cylinder fail when it was inside the trunk. It made me take VERY good care of my tank.

http://www.thescubaguide.com/gear/tanks/safety.aspx

We also don't use GCC. We typically use the Intel or PGI compilers and applications like IDL or Matlab.

Using Visual C++ would require windows. The disadvantage of using windows over *nix are many: Licensing, Manageability, ad infinitum.

Our main cluster is CentOS-based using ROCKS cluster software and is running on 208 blades using several different generations of HP BL46[05] G[1-6] blades that get refreshed periodically. We have a mixture of CPUs in production currently, dual dual-core Xeon, dual quad-core Xeon and Opteron and some dual quad-core Nehalem. We're replacing the dual dual-core blades with dual-hex shortly to gain processing power without a power/cool/weight penalty.

We usually disable hyperthreading since we prefer each processor core be used for one job and only one job. Time-slicing between two jobs means that time's wasted flipping back and forth and resources will be used up needlessly. We also size system RAM so there is at least 3GB/core available for each job submitted. Using those strategies, we avoid swap as much as possible and get the maximum CPU utilization. Typically, each core is running at 100% for days on end.

A PS3 cluster would be similar in use, but limited to jobs that didn't need a lot of RAM or disk I/O. Since a PS3 cluster is pretty much a roll-it-yourself endeavor, I don't expect it to get used nearly as much as the production general purpose cluster. As time passes, it begins to look more and more like a toy compared to the newest advances in HPC.

When the PS3 cluster project started, it was one of the "sexiest" ways to get your hands on a lot of power for not a lot of money. Recent developments in GPU-based computing is beginning to attract attention, but not every system is a viable host for a production grade GPU board. System form factor and power requirements are significant issues to overcome. CUDA/Tesla boards show promise, but are only suited for certain applications, they would never be intended to replace general-purpose cluster computing. Typically, we'd want something that can run 24/7/365 with a very high MTBF.

The initial cluster was only 10 systems and was in the proof of concept phase. To do meaningful work, we'd want many more systems and the project seemed to be at a dead-end when they changed the firmware.

Power/Cooling and physical footprint are also considerations one needs to worry about. Price/CPU is very good on a PS3 too. For certain types of tasks, the CEL processor is ideal. We don't need a lot of RAM or hard drive either. We just need to crunch through a lot of numbers quickly.

We do a lot of High Performance computing where I work. I currently run a 1472 core ROCKS cluster and the price/core of a PS3 cluster is pretty good compared with an HP blade.

Because of management issues, I can't see running a huge cluster of PS3's, but it's an angle we're pursuing to see how well it works. Tesla-based clusters are also something being considered, but they type of work you can do on them is a bit more limited than a general purpose cluster based on conventional blades.

Our company was setting up a small cluster of PS3's but the whole project died when Sony locked down the firmware. This should breath new life into the effort.

I was secretly hoping they'd give the "useless" PS3's to the employees though. Sigh.

Yeah, that's pretty much the reason I gave up reading pournelle's column. :)

I subscribed to Byte "back in the day" and was disgusted with its slide into irrelevance. More of the same will only further sully any respect I had for what it was in its heyday.