George:
Nice answer. Do you think cloud services is based on an oversubscription mo= del? Where they hope those who purchase servers don't actually max them out memo= ry/CPU wise?
Do you also believer that cloud services should never have any downtime? To= me=2C cloud services is synonymous with redundancy....
That's an interesting question, and really points more to the fact that "cloud" is rather poorly defined. For example, consider the T-Mobile Sidekick Danger server crash/disaster. This is frequently pointed to as a "failure of the cloud", but in reality, it appears to have been trusting data to a company that wasn't exercising proper care in maintaining its servers. People glommed onto the concept that it was a failure of the "cloud." However, one could argue that quite often, anytime something magically disappears into a part of the Internet we don't have physical control over... I've been toying with defining cloud in a different direction. We have dedicated servers. You get a 10 GHz 24-core CPU with 1TB of RAM. That's pretty clear and familiar to server geeks. We have virtual servers. You get (up to) M GHz and N cores of that same machine. Oversubscription is possible, but not required. In many cases, oversubscription is desirable because that's where the capex and opex savings of less hardware comes in. In both those cases, we get tied up in the specifics of hertz and cores and amount of memory. In the virtual server case, we make some progress towards a model where a VM could be migrated around onto more suitable hardware. This is useful for allowing the proper sizing of a virtual server, for redundancy, upgrades, etc. It seems, though, that ultimately what people seem to be thinking of when they think of the cloud, is the ability to just have stuff "run" without necessarily having to worry so much about the details. In some cases, they're looking for redundancy, or reliability. In many cases, they just want something to be out there without so much effort on their part. They want it to run fast if it gets busy, and don't care if the CPU is oversubscribed ... as long as they can get what they're paying for when they need it. I don't think cloud service purchasers will ultimately be that interested in worrying about whether they "max out" memory/CPU. I think they don't want to have to worry about it too much, though they probably want to be protected from bill shock. That means a model where their server might actually be hosted on a large host with a few hundred other mostly idle VM's, when their VM is idle, and then get migrated onto other hardware if demand spiked. We have technology that can even power on additional host hardware, so there are ways to save on power/cooling during non-peak times. I think you'd find such models are harder to implement if you're too focused on the "evil" of oversubscription. I think what you want to avoid are providers who are unable to maintain sufficient spare capacity to cope with peak demand. ... JG -- Joe Greco - sol.net Network Services - Milwaukee, WI - http://www.sol.net "We call it the 'one bite at the apple' rule. Give me one chance [and] then I won't contact you again." - Direct Marketing Ass'n position on e-mail spam(CNN) With 24 million small businesses in the US alone, that's way too many apples.