Slashdot Mirror
Dell Exec Calls HP's New 'Machine' Architecture 'Laughable'
jfruh (300774) writes HP's revelation that it's working on a radical new computing architecture that it's dubbed "The Machine" was met with excitement among tech observers this week, but one of HP's biggest competitors remains extremely unimpressed. John Swanson, the head of Dell's software business, said that "The notion that you can reach some magical state by rearchitecting an OS is laughable on the face of it." And Jai Memnon, Dell's research head, said that phase-change memory is the memory type in the pipeline mostly like to change the computing scene soon, not the memristors that HP is working on.
No-one's interested in his shitty computers anymore
If you gave me a choice between a printer and a giraffe with explosive diarrhoea, i'll get my ladder and my raincoat
HP has a long history of OS and CPU design, including their own computers with a proprietary architecture. Not all of their designs were successful, since they were co-designers of the Itanium with Intel. So HP has the exactly opposite corporate background the Dell.
Why would anyone pay attention to what a Dell talking head has to say?
Why is Snark Required?
HP is their competitor. HP just announced that they're working on something that even if the entire thing doesn't come to fruition, likely some part will and it will change the computing landscape. Understand, this announcement is pointed directly at Dell's share holders.
Best case scenario HP actually pulls it off and they've got some radically fast system running something that looks like Linux.
Mid case scenario, they figure out how to make memsistors at scale and then sell licences for everybody to make blisteringly fast SSD's, etc. Then others come along and figure out how to put the pieces together. HP makes out like a bandit in royalties, etc.
Worse case, nothing comes out of this. HP shrugs, files a whole pile of patent applications. Someone else takes bits and pieces of it (like IBM) and does cool things with it. In all three cases HP is going to be enhance their IP portfolio and possibly make their stock worth more.
All of those scenarios are bad for Dell. Dell doesn't do fundamental science. They design motherboards that use components supplied by everybody else and crank out cheap computers. If scenario #1 comes true... HP is NOT going to sell any of this to Dell, cutting them out of the market. If scenario #2 comes true, HP is going to get these components at a price that Dell can't compete with. If the last scenario comes true, Dell still ends up being a VAR like everybody else and HP racks in royalties.
The CEO of Dell is almost obligated to thrown cold water all over this, otherwise Dell shareholders are naturally going to ask if this announcement is going to make Dells stock worth less and/or worthless.
Yes Francis, the world has gone crazy.
"With persistent memory, the machine state gets messed up, you are so screwed."
Uh, have you looked into your computer recently? I believe you'll find either this little device called "an HDD" or this other little device called "an SSD". And people with those seldom get screwed.
Ezekiel 23:20
The notion that you can reach some magical state by rearchitecting an OS is laughable on the face of it
Why, thank you, Captain Obvious! It's not about rearchitecting an OS, it's about matching SW to the HW. For ages, we've had the distinction between block-addressed devices with streamed access and byte-addressed devices (mostly DRAMs) for low-latency. Virtually all our software is impedance-matched to that idea! I believe the only thing remotely close to how a machine with huge persistent RAM should (would?) work are those nice Azul boxes, with zero-pause automated memory management even on 500GB+ heaps. Those machines still use RAM and have disk I/O for ordinary data manipulation, but I'm convinced that had the Azul people had non-volatile RAMs at that time, they would have gone for persistent objects. It's such an obvious idea! No more serializing and deserializing for disk I/O (except for backups, of course), performance on the order of millions of transactions per second. Obviously the price is that you absolutely have to rewrite the software bottom-up, otherwise all that extra performance potential gets lost.
Ezekiel 23:20
"The notion that you can reach some magical state by rearchitecting an OS is laughable on the face of it," John Swainson, head of Dell's software business, told reporters in San Francisco Thursday when asked to comment on the work.
Well, sure, you also have to rearchitect the hardware, which is what HP is talking about. John Swainson is an idiot. Sadly, the richest idiots with the best-connected families fail upwards rather than downwards. This is why we can't have nice things.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
"With persistent memory, the machine state gets messed up, you are so screwed."
Uh, have you looked into your computer recently? I believe you'll find either this little device called "an HDD" or this other little device called "an SSD". And people with those seldom get screwed.
If you read the article from the previous slashdot story about HP's "The Machine", you will find that they are not simply trying to use memsistors to replace main memory, but that they are also trying to consolidate the storage memory and working memory into a single piece of memory, this is why it is considered to be substantially different memory architecture which also requires the OS to work a little differently too... if you are old enough think "Ram Disk"
The difference being that usually any stored data to be used by the processor has to first be loaded into working memory from the large slow storage memory... as i'm sure you are aware, which is why SSDs are so popular... but even NAND is many times slower than SDRAM, so the separation remains.
The idea is that if a sufficiently fast, dense, persistent and cheap type of memory can be found then the best of both can be consolidated into one. The concern of the OP is that issues affecting running state could affect the traditionally less dynamic stored state... Working memory is usually treated as volatile and disposable, and your block device is not, but the line is now blurred.
I think it's a reasonable concern, but one that is likely to be addressed by the OS, a less physical separation between what is running state and what is not would need to be implemented, but at the same time the advantages of not "loading" data need to be retained... making everything that goes into the running state duplicate would bring back the "loading" problem slightly.
Actually the financial math is all about amortizing those costs over the life of the product. So if Apple sold 4 iPhones they would have to allocate 500,000 man hours to each phone. The same with all those developments over time. Modern PCB technology is actually quite cool and no doubt took some serious development, but it has been amortized over a zillion PCBs. Apple would actually be paying those amortized costs as well in that any recent developments would still be including those costs when some company uses a recent development to supply them with a part.
But the key to amortizing a cost is that it eventually effectively hits zero. So the costs from Industrial Revolution developments were long ago reduced to zero. Although many times the amortization is a curve that is asymptotically zero; thus to be pedantic it is possible that some impossibly small portion of an iPhone is still paying off the development time spent 100's of years ago. From an economics point of view this is not actually impossible. There could be an area that specialized in say, fine machining, 300 years ago to a point where the same companies are in the same area still leaders in that field. Thus apple would have bought some of their manufacturing equipment from that company. Examples of this abound in Germany where there are plenty of companies that are from the Prussian Empire or before that are world leaders in their area of expertise; so good they survived Napoleon, WWI, and WWII. Krupp I believe is around 400 years old.