As I trudged toward a swanky hotel for a meeting with Dell, the Seattle sky was spitting cold rain like an old man realising the soup in his mouth is way too hot. (Adding more drama to these intros, nice, right?) I expected two things that day: “It’s Seattle in November; it’s going to rain,” and “It’s Dell at Supercomputing; …
Intel part availability
It's all very well Dell saying their hardware models will be available for 5 years, but Intel CPUs don't have that lifespan. You might go to dell in 3 years time and say "I'd like the servers we got before" and they will have to say "We can do the board but the Xeon E5678 parts you paid for then aren't available no more". If the replacements fit into the existing CPU socket and are compatible with the chipset (which implies qualification) then you will end up with a cluster that is only weakly homogenous; the difference in performance will lead to quirks: race conditions etc, and may need new configs to deal with their ability to do more work in less time.
If the CPUs aren't compatible with the old chipsets -and have you tried to get a recent Xeon into a 5 year old box?- then they're going to have to say "sorry, we lied". Or they buy up a roomful of the old parts before they go off the intel price list, which given Dell's attitude to inventory is unlikely.
All x86-based HPC server vendors have this problem; it's why things like ARM designs may have an edge. If the foundry can still fab in the same process, that Core design you bought off ARM will still work.
Not on enterprise machines
My experience over the last seven years is that when you buy the Dell equipment that is targeted at the enterprise they don't randomly change the BOM as the article suggests. Sure if you buy the consumer line of products that does happen, but that is presumably why they are cheaper than the enterprise line.
Sure if Intel stop making the parts you can't buy a server with the same CPU several years down the line, but all vendors have the same problem.
HPC support and tuning
Who told the author that HP, IBM et al have HPC experts working for them. They don't. All of the Tier1s, (excluding Cray), are pretty poor at delivering HPC systems, because they laid off most of their people.
- Vid Hubble 'scope scans 200,000-ton CHUNKY CRUMBLE ENIGMA
- Google offers up its own Googlers in cloud channel chumship trawl
- Bugger the jetpack, where's my 21st-century Psion?
- Interview Global Warming IS REAL, argues sceptic mathematician - it just isn't THERMAGEDDON
- Apple to grieving sons: NO, you cannot have access to your dead mum's iPad