Machines on steroids.

[bobajobrob]

Hence the freezer trick. If it's a mechanical problem then this can shrink the parts enough to get things moving again.

[TimO]

If the spindle is accessible, it's a more reliable way of overcoming the stiction, although the freezer trick may also solve it.  To an extent, a good hard whack may also help, but that's really a last resort solution, since it could equally kill other elements as well!

[rae]

To an extent, a good hard whack may also help, but that's really a last resort solution, since it could equally kill other elements as well! 

That used to be standard procedure for Sun engineers on A5000 arrays.  They'd show up, pull the drawer open, lift the offending disk out of its slot, then drop it back in.  They'd then shut the drawer and invariably the disk would spin up, and work fine for months afterwards.

Those servers are split over three racks - we couldn't fill a complete 42U rack with them, for the reasons you say. 

That is one of my pet rants about exotic technology.   Blade servers and the like are terribly clever (and jolly expensive, short lived and complex), but physical space is generally not a problem in a modern data centre.  Heat dissipation is far more of a problem - and a good data centre has a loading of about 2kW/m2.  The average crappy data centre will be lucky to have 1 kW/m2.

[ScumOfTheRoad]

That is one of my pet rants about exotic technology.   Blade servers and the like are terribly clever (and jolly expensive, short lived and complex), but physical space is generally not a problem in a modern data centre.  Heat dissipation is far more of a problem - and a good data centre has a loading of about 2kW/m2.  The average crappy data centre will be lucky to have 1 kW/m2.

Rae, I agree with you.

These servers are not blades, they are standard 1U units with two motherboards per U. They share a power supply, which is 90% efficient and also you share PSU losses between two servers.
We have quite a few clusters using these twins, and I believe they are the cat's pyjamas for high performance computing. Examples are an aerospace customer, this cluster at QMC, particle physics grid at RAL.

Supermicro, Inc. - Products | SuperServers | 1U | 6015TW-TB

Re. heat dissipation we haven't done water cooled racks yet, but we have provided one big system with APC racks with large cooling doors on the back which are just huge fan units.

[Greenbank]

48 x PIII 800MHz, each with 1GB RAM and 2x40GB HDD.

(so 38.4GHz, 48GB RAM, 3.84TB of HDD space.)

6U in total. Under 10A when all 48 cores are fully loaded. Would be less if I removed the HDDs (they all netboot anyway so the drives do nothing but consume power until they spin-down due to inactivity).

Did I say that this was at home not at work?

[mrcharly-YHT]

pah. That's nothing.

I've just loaded Linux onto a 20MHz chip.

[Woofage]

pah. That's nothing.

I've just loaded Linux onto a 20MHz chip.

Details please!

[Greenbank]

20MHz is fast!

I used to have Linux running on an old 16MHz Compaq Proliant SX-16 as a gateway machine back at University. 4 of us in the house could share the single 33.6kbps modem connection we had to the University network.

[TimO]

I keep on thinking I should shove a OpenRISC CPU into one of our FPGAs, and then we can run Linux on one of our instruments in orbit!  (and it would probably be running at that sort of speed!)

Unfortunately, I don't think my boss wants to take the risk with that sort of development ! It's fair enough, since there are all sorts of ways that things could go wrong, and frankly whilst I may have a chance of getting it to work (eventually!) no one else around here would, so it would be rather high risk.   It's a shame, since it would definitely be fun.

[ScumOfTheRoad]

I keep on thinking I should shove a OpenRISC CPU into one of our FPGAs, and then we can run Linux on one of our instruments in orbit!  (and it would probably be running at that sort of speed!)

Out of curiosity, what OS do your orbiting instruments run? And what do they do? Inquiring minds want to know!

[mrcharly-YHT]

pah. That's nothing.

I've just loaded Linux onto a 20MHz chip.

Details please!

2.4.32 kernel with busybox and ulibC + network drivers loaded via JTAG port onto META 4-threaded chip. Power consumption - 0.5W.

[TimO]

I keep on thinking I should shove a OpenRISC CPU into one of our FPGAs, and then we can run Linux on one of our instruments in orbit!  (and it would probably be running at that sort of speed!)

Out of curiosity, what OS do your orbiting instruments run? And what do they do? Inquiring minds want to know!

Our current ones don't run an OS at all, in fact they don't even use interrupts, it's just a damned big loop with lots of polling going on, not very elegant at all but it works.

We are working on a project which uses RTEMS (on LEON), so that's a possibility for future missions.

We've also done a little work using Salvo, but that was going to be used on CubeSat's which have a whole different set of issues and requirements (mostly relating to low everything!)