In the software industry, “The Cloud” is used as a metaphor to represent the Internet. Any computing resource, software, or service that can be shared over the Internet is considered to be in the Cloud. This sharing of resources and anytime/anywhere accessibility makes Internet-based software more efficient and cost effective than traditional on-premise software.

http://www.surgeforward.com/InternetCloud.aspx

some good points you made there on lack of standards, I made the point in the post that they are all being done in a proprietary way now, for the distributed runtime/services layer idea to work it needs that standardisation; otherwise we just keep building “pots” of clouds that can’t talk to eachother easily.

Blades/rack mounts/10Gb/1Gb hardware; it’s all abstracted in my reference architecture throught the heavy use of a hypervisor – the underlying hardware and interconnects can be designed and upgraded/scaled back based on load without affecting service (using tech like HA and DRS)

To me what would be ideal is if Amazon, MS-Azure and Google let you download the VM’s they use to run bits of their clouds as “appliances” and run them on-premise; (we know for a fact that Amazon and MS use hypervisors to run the underlying services) the fact that you can run them on a standard/reference architecture stops it becoming pots of implementation/service – you can still provide them on one infrastructure if you need to.

This would instantly get round the whole standards bit for me (no slow and painful design by comittee, the best or most commerically astute tech wins) – even if you only rent/licence the instances from Amazon/MS/Google to cover their IPR/dev cost you don’t use *their* compute power/inf to execute/store you use your own.

This would be a good way for them to increase adoption and break down those data ownership/risk barriers for the corporate market – let’s face it are the banks going to put their crown jewels into Amazon’s hands? but would they leverage Amazon/MS/Google cloud API tech to build big scale distributed applications if they could run it in their own walled garden?, I would say a resounding yes!

This is why I particularly like VMWare’s vCloud concept – all you really need to standardise is the hypervisor layer (or use products like PlateSpin PowerConvert to convert VMs from one to another), the rest is config – the “clever” bit is the tech that runs inside the runtime/services layer code {VM appliance} that you lease/buy from Amazon etc. – look how sucessful Google have been with their very expensive (re-badged Dell ) search appliances that they sell for “on-premise” use – it’s the same principal for me.

Microsoft have said they are “looking” at this type of on-premise running for exactly this reason, although they have told me that Azure would be very complicated for a customer to run on-premise – I would think this is more due to it being so new, and needing a lot more work to make it run this way – essentially packaging it and makign it simple for the end IT user/implementer; this is MS bread & butter business (Windows Server, Live Services etc.)

In reality this is no different from a VMTN virtual appliance that you can deploy internally and build apps to talk to; like a LAMP VM with SugarCRM for example, but with some back-end cleverness to deal with distributing load/execution etc.

Anyway, just my 2p

http://rodos.haywood.org/2009/01/what-is-cloud-conversation.html

I think much (not all) of what you are describing in the text is utility computing which is the basis for delivery of most cloud infrastructure, certainly the IaaS part.

Great conversation you have kept rolling here and look forward to some good interaction and keeping it going.

Cloud standards will happen, but it won’t be via processes we’re used to. Amazon’s APIs for example are relatively clean and a good example to follow (as Eucalyptus have done) and others have (relatively) independently come up with similar results. Once we have a large enough pool of users standards should naturally evolve and there is little point in forcing the issue. There is after all good reason for saying “the devil’s in the details” – just take a look at Grid and WS-* – premature standards at this early stage could well throw a wet blanket over the rampant innovation taking place.

Your reference architecture looks good at a glance, but again it’s too much detail for users (even if relevant for providers and enablers) and it lacks cloudy concepts like local node storage and fast, generic interconnects (eg [10]GigE). The blade formfactor may well see a resurgence now we’ve found a good application for them but there’s somewhere some more standardisation might be useful (eg for an enclosure which turns a handful of horizontal rack units into a bunch of smaller vertical ones, such that I can install generic blades from various vendors). Google appear to be leading the way on next generation hardware anyway.

What is sure is that there are interesting times ahead!

Sam