Larry,
For the past couple of years I've read innumerable posts from ZDNet and TR about the "coming cloud revolution"...now we're supposed to believe that instead of a cloud we just daisy chain all of our cellphones/smartphones together to create some sort of virtual server???? Just as with the predictions that cloud computing is going to takeover datacenter servers, I don't think too many large enterprises would seriously consider dropping to Snapdragon processors for their in-house servers.
Let's see now...if I use the USB port on my Samsung Blackjack, wire that to my son's LG Slider, then Bluetooth to my wife's Razr...voila!!! Instant home server!!! Not!!!
Discussion on:
View:
Show:
And daisy-chaining cell phones has nothing to do with this. (But that would be some weird cloud, perhaps, given the proper functionality, which does not come in cell phones.)
It is about using a different processor scheme to save energy. Whether the servers are part of a "cloud" or not. Are you sure you read the article?
It is about using a different processor scheme to save energy. Whether the servers are part of a "cloud" or not. Are you sure you read the article?
Direct quote from the article...
"Smooth-Stone is arguing that you can take the low power of mobile phones and apply them to data centers. In a nutshell, Smooth-Stone is talking about stringing together a bunch of Qualcomm Snapdragon chips-or TI processors-from your cell phone and powering a server"
Now, as I said, there have been numerous posts regarding some momentous shift to cloud computing. What chips power all those servers in the datacenters proposed to power our shift to the cloud? Not Snapdragons, not Intel Atoms, maybe a few PowerPC chips, but to have a chip manufacturer come along and say you could scale the same computing power in a bunch of ARM based chips, well, that makes as much sense as daisy chaining a bunch of cellphones together.
Besides, it's Saturday, and I felt a tad sarcastic!
"Smooth-Stone is arguing that you can take the low power of mobile phones and apply them to data centers. In a nutshell, Smooth-Stone is talking about stringing together a bunch of Qualcomm Snapdragon chips-or TI processors-from your cell phone and powering a server"
Now, as I said, there have been numerous posts regarding some momentous shift to cloud computing. What chips power all those servers in the datacenters proposed to power our shift to the cloud? Not Snapdragons, not Intel Atoms, maybe a few PowerPC chips, but to have a chip manufacturer come along and say you could scale the same computing power in a bunch of ARM based chips, well, that makes as much sense as daisy chaining a bunch of cellphones together.
Besides, it's Saturday, and I felt a tad sarcastic!
Made for a marvelous thought experiment. I'd read about this elsewhere, but I didn't think to take it as far as you did!
Of course it could be made to work. Anything can be made to work. NASA should not have been able to put a man on the moon with the pitiful computers of the day, but they did. And the average cell phone probably has more computational power than a dozen of those old mainframes. The questions are:
Will the idea work well?
Will the idea sell well?
I think if you weren't running a full blown Windows Server OS, but instead a Linux kernel with a single purpose, it would probably work very well. Windows server is designed to do everything any server can do, but most enterprise servers are dedicated to a single task. In my opinion, that OS has tons more overhead than it needs to have. (Yes, I know not having a GUI would cripple me first, since I do not know command line work very well at all. I would need to get with the program.)
Looking at it from the hardware side, modern servers have multicore processors, and often multiple processors as well. The servers are often overpowered so it can handle the occasional spike in demand and the requirements of the OS. I don't see any difference (as a concept,) between a cluster of lower MHz processor in a multicore and multiple lower MHz processors outside of a multicore. It is semantics. The difference is probably going to be in the Admin's (or CTO's,) perceptions.
Besides, the power companies transfer current @ 50,000 volts, then step it down for neighborhood distribution. Then it comes into our buildings as 120, 240, or 480 volts. It then enters our servers at 120 volts. Our servers reduce it to 12 volts for the motherboards to use. The motherboards then run the processors at 2 to 3 GHz. Each step of the transfer process has its own efficiency loss. Every step generates massive amounts of heat. The fans that cool the servers use power and generate their own heat. The whole model needs to be reinvented. At least someone is trying!
It is easy to ridicule other people's ideas. I would rather see one of these new servers and evaluate it myself. If it doesn't pan out as a better total solution, at least they tried!
Will the idea work well?
Will the idea sell well?
I think if you weren't running a full blown Windows Server OS, but instead a Linux kernel with a single purpose, it would probably work very well. Windows server is designed to do everything any server can do, but most enterprise servers are dedicated to a single task. In my opinion, that OS has tons more overhead than it needs to have. (Yes, I know not having a GUI would cripple me first, since I do not know command line work very well at all. I would need to get with the program.)
Looking at it from the hardware side, modern servers have multicore processors, and often multiple processors as well. The servers are often overpowered so it can handle the occasional spike in demand and the requirements of the OS. I don't see any difference (as a concept,) between a cluster of lower MHz processor in a multicore and multiple lower MHz processors outside of a multicore. It is semantics. The difference is probably going to be in the Admin's (or CTO's,) perceptions.
Besides, the power companies transfer current @ 50,000 volts, then step it down for neighborhood distribution. Then it comes into our buildings as 120, 240, or 480 volts. It then enters our servers at 120 volts. Our servers reduce it to 12 volts for the motherboards to use. The motherboards then run the processors at 2 to 3 GHz. Each step of the transfer process has its own efficiency loss. Every step generates massive amounts of heat. The fans that cool the servers use power and generate their own heat. The whole model needs to be reinvented. At least someone is trying!
It is easy to ridicule other people's ideas. I would rather see one of these new servers and evaluate it myself. If it doesn't pan out as a better total solution, at least they tried!
and memory and disk don't take as much juice or more than the cpus?
If I have one server drawing 250-600 watts of power but running 8 different virtual servers, that is better than 8 physical servers drawing 100 watts of power each. Granted, I have no idea how much power one of these servers will actually draw, but I highly doubt you will be running VMWare or HyperV on one of these ARM based machines.
IBM offers virtual software that runs on RISC processors. If the processors don't provide the performance per watt that other processors do then they probably won't make it very far.
Just as a note, VM Hypervisors do not have any power management features that a phyical install does so a physical server can scale back it's power usage based on load where a VM can't. This doesn't mean that VM server don't save power, I just think that a little analysis is necessary to really determine what the savings are.
Bill
Just as a note, VM Hypervisors do not have any power management features that a phyical install does so a physical server can scale back it's power usage based on load where a VM can't. This doesn't mean that VM server don't save power, I just think that a little analysis is necessary to really determine what the savings are.
Bill
- Keyboard Shortcuts:
- Prev
- Next
- Toggle
