Re: [ Cloud Computing ] Private clouds and their uses (WAS: What is Cloud)

[Gilad Parann-Nissany]

Jim said: " I'd much rather hear from actual cloud users (a rare bird on this forum), cloud providers, and other cloud technology developers".

Well I'll give it a try. I think we qualify to the definition - take a look at http://g.ho.st/ - we are both a user and a developer. Also I can contribute some thoughts based on my years at SAP.

G.ho.st are a PUBLIC cloud developer. In the public cloud its pretty straightforward.

• As a user, you really do get rid of up-front capital expenditure and replace it with monthly "rent" for hardware (AWS EC2, Azure compute, ...), platform (AWS S3 & SDB, Azure SQL & IIS, G.ho.st file system, G.ho.st user-management system, ...), application (G.ho.st front-end, your favourite SaaS solution, ....).
• As a developer, you take some of these benefits (AWS EC2, Azure Compute, AWS S3 & SDB, Azure SQL & IIS) and package them to put your own solution on top. So you "rent" some capabilities and then "lease" an expanded capability. Hopefully for a consideration...

However this thread of discussion seems more about private clouds. Extrapolating from public cloud to private cloud is tricky. There are several possibilities. In some situations its just hype - "renting" and "leasing" within a single organization is just a hassle. On this pessimistic view, private clouds are a bureaucrats' dream and a line-of-business manager's nightmare.

So is there an optimistic view of private clouds and their uses?

• Long term - it is claimed (I have my doubts) - huge enterprises on the Fortune 500 scale are so big that the extra bureaucracy may be worth the efficiencies of an "internal market".
• For example its been published that the USA Department of Defense is considering some such internal projects.
• The DOD is of course huge, needs to keep things internal (so many secrets), is always interested to be on the cutting edge, and already has some bureaucracy...
• Its worth following, but (in my view) will not be used for the mission critical stuff in Fortune 500 companies - in the next 5 years.
• I was recently at a Gartner conference were some analysts felt so as well.

• Short and Medium term - I believe there will be a market for variations of the private cloud idea, that will be able to make some headway in the departments of medium and largish businesses.
  

What "variations" of private clouds could make it? They need to offer a clear benefit based on cloud economics, namely avoiding capital expenditure.

1. one thing that will happen a lot - people will take internal Web projects and call them "cloud". But that is hype again, so lets skip it.
2. "Virtual Private Clouds" - allowing a company to rent resources OUTSIDE its premises but with a virtual security perimeter that makes it safe to deploy private info on the virtual cloud. The security technology for this is not completely in place, though a number of vendors are trying for this goal.
3. "Managed Private Clouds" - a company rents a complete stack (hardware, platform, software) which is delivered BEHIND its firewall as a complete unit. The seller rents out this complete set (racks, software and all) together with maintenance - maybe even a remote maintenance connection. If the buyer dislikes the solution, they have an opt-out and the seller must take away their entire stack. Of course rental periods in this model cannot be too short - probably years rather than weeks.
4. "Start public, go private" - a company can try out a solution at low risk on a public cloud, once they decide they really like it - they go for #2 or #3. (Or even #1 - just taking in as a capital expenditure the complete solution)
5. ... there will fo course be additional good ideas not on this list
   

Now going back to the discussion of this thread. Private Billing mechanisms are clearly needed for #2 and #3 and by extension #4. They are not needed for #1 (nor for #4 if the "opt in" is to go to a #1).

Note that Jim is right - even if you do need the private billing mechanism, the granularity will be decided by pragmatic considerations. For example in case #3 you will pay perhaps per rack? (just an example, please do not take it too literally...)

Like everything in life, there is no one answer that fits all situations. People will (perversely perhaps) insist on doing old fashioned web projects from time to time (i.e.#1). Their will however be interesting growth in the market for #2, #3, #4. So there is money to be made by the smart vendors who cater well to this.

My bottom line guess: don't bet on the CIOs of Fortune 500 companies - this growth will find customers in the line-of-business departments of the largish companies and in the innovative smaller companies. It will be driven by showing proof that capital expenditure can be avoided and replaced with leasing/renting.

Regards
Gilad
__________________
Gilad Parann-Nissany
VP Products & Engineering
http://g.ho.st/

On Mon, Jul 20, 2009 at 3:08 AM, Charlton Barreto <charl...@acm.org> wrote:

Miha, I would be interested in your view on how fine grained metering should be in your model. Enterprise IT has employed a number of approaches to this, some of which are as fine grained as Jim proposes above.

However, the cost measure of *many* of these are in kW/year, so this is rather partitioning your power consumption metering further.  

Jim, interesting that you mention software 'wear and tear' - care to elaborate on that a little? 

2009/7/19 Jim Starkey <jsta...@nimbusdb.com>

Miha Ahronovitz wrote:
> The question I am asking is how do we use it to make sure that each user pays only for what s/he uses in the cloud. Similar to the light. I switch on, the meter records consumption and I get a bill.  I switch it off, I pay nothing.
>
>

But Miha, electricity one has one dimension.   Computing has many.
Computing esources consumed include:

   * Cycles used
   * Cycles available
   * Cores
   * Memory used
   * Memory allocated
   * Local storage used
   * Local storage allocated
   * Persistent storage used
   * Persistent storage allocated
   * Presence (eg idle but prepared to respond to client)
   * Bandwidth (allocated / used) in the cloud
   * Bandwidth (allocated / used)  to the external world
   * Wear and tear on software

To meet your restrictive model of a cloud, must a provider charge
individually for each of the line items, or a (purported) cloud provide
allowed to use business judgment to develop a simpler model.

And do note that EC2 doesn't charge anything for resources consumed,
only presence.  An instance with four cores churning at full bore has
the same price as an instance sitting idle waiting for work.

I continue to argue that charge back is often appropriate, but not
always.  A cloud dedicated to providing computer services to non-profits
may never charge back but be a cloud never-the-less.

What makes economic sense will persist and what doesn't will go away.
Fads always end (even the XML mania has largely subsided, happily
leaving useful technology).  It isn't necessary to argue the economics
unless you're either a prospective user or a potential investor.  If
you're a user, it makes sense or it doesn't.  If you're an investor,
well, so far it doesn't look very interesting, at least in the sense
that major players are big companies spending internal funds to build
out infrastructure.

Personally, I'm beginning to tire of cloud pundits wannabes and cloud
consultants wannabes.  I'd much rather hear from actual cloud users (a
rare bird on this forum), cloud providers, and other cloud technology
developers.