You can parse this down to clock speed, GB Memory per core, bandwidth & latency - but at the end of the day, what really matters is your application's end-to-end performance. When someone hits your application, how fast does it respond? Does that response rate hold true under 10000x load? And in a federated cloud where you've got your 10gen App server running on a compute system at Peer1 using storage from S3 - whose throat do you choke if that magic number isn't met?
That kind of accoutability/SLA/performance is going to be a cost driver for many organizations. Add to that the complexity and cost of things like audits and corporate compliance, and $/cpu hr doesn't quite cover things.
If youv'e set up a compute grid between organizations (even on a campus or within a company), you know that the politics and SLAs involved are just as (if not more!) challenging than the technical side of just getting it to work.
It's going to be interesting to see how the market adapts to these challenges
Shane
--
http://www.10gen.com/
http://twitter.com/shanebrauner
On Tue, Dec 16, 2008 at 5:38 PM, Tim Freeman <tfreeman@mcs.anl.gov> wrote:
MAybe from a sales perspective, but if we are truly talking about getting 'IaaS
On Tue, 16 Dec 2008 16:52:17 -0500
"JL Valente" <jlvalente@cittio.com> wrote:
> As usual on this group we have very different camps and opinions which makes
> the whole debate quite useful. But again I don't think that this an either/or
> proposition. It depends who the "end user" is really. Some of the ones we are
> dealing with go deep and require serious metrics and understanding of what
> they are really getting from their cloud provider.
units' traded as a commodity in some future, it probably better be just one,
agreed-upon set of metrics.
I think we have a general consensus that this involves much more than hardware
stats, including external-to-the-host factors like latency and throughput
in/out of the facility and internal choices like VMM versions and
configurations (with both versions and configurations, there are a LOT of
precarious things that will affect performance).
I don't see how it could be done without a set of comprehensive benchmarks
routinely run on the providers (hopefully by auditors we can trust).
Tim
>
> ________________________________
>
> From: cloud-computing@googlegroups.com
> To: cloud-computing@googlegroups.com
> Sent: Tue Dec 16 15:02:25 2008
> Subject: [ Cloud Computing ] Re: Compute surface as a traded commodity?
>
>
> From Paul Moxon:
> [if you keep it simple, then they will get it]
>
> I agree with Paul. From a cloud-computing, sales engineering perspective,
> price is based on what you can offer customers, not what it costs the company
> to provide such services. The latter has to be quantified to ensure
> profitability, but no matter how much the 'processing-markup' if you have
> what customers don't want, or do have what they want but make the process to
> complex, you will loose customers and slow down the migration to 'the cloud'
> overall. That doesn't help anyone's bottom line.
>
> What the company I work for does is create a formula, with multiple
> components that averages what a single 'seat' or 'user' will consume with the
> hosted application they've purchased. There will be statistical outlyers (in
> the case of AV, these may be part of the formula), but overall it needs to be
> reasonable and reaslistic. Then they charge the customer in 'seat' tiers
> (1-1000, 1000-3,000), annualized (can be monthly) and so on. This has so far
> proven to be both manageable and profitable, and includes lots of SMB and
> Enterprise sized customers...
>
> So as an SE, I would work it backwards - from the end-customer to the
> hardware - Price the applications, support model and so on. For example: I
> just bought a home. I have one utility provider, but am charged by the
> 'application' (DirectTV, Internet, Phone, etc.). In this example, the utility
> provider is charging Megawatt hours, which is obviously where this thread was
> starting, but I think it's folly to separate the two, when the same company
> can Provide both the Applications and the Power to run them. All the cloud
> provider have to do is package OEM'ed solutions. Both Sprint and AT&T have
> dived in head first with this model for the last year or so.
>
> Standardizing on GHz/hour or other hardware based metric is wrong, in MHO,
> because the applications that are hosted in the cloud will have different
> efficiency ratings. This may matter to the cloud provider, but the
> end-customer can care less - and the sales person selling the
> application/services can care less, and so on. Infrastructure costs will not
> be a factor, at least in the short term.
>
> Customers also don't care how you provide the service - except for the part
> they interface with. In Paul's example, the only part he would care about is
> the price of fuel, the location of the gas station and so on. Nobody cares
> about oil refinement, oil discovery costs, or the trucking of refined fuel to
> the gas station. The same should be true of services provided in the cloud.
> If the customer has to think to much about it, what is the advantage of
> moving from CPE ot the cloud anyway??
>
> Daniel Scafuto
>
>
>
>
>
>
> On Tue, Dec 16, 2008 at 11:26 AM, Paul Moxon <paul@moxonsonline.com> wrote:
>
>
> Do end users really need to see this level of complexity? Internally,
> the cloud provider might want to base the charge rates on various
> measurements, such as CPU, memory, bus speed, etc. However, the end user will
> probably want something much simpler e.g. I want to run on a (virtual) single
> host or on a cluster or, even, I want 24x7 availability. Elastica uses a
> fairly simple system like this when you build an application to be deployed
> in their cloud. Anything more complex can make it overwhelming for anyone but
> the most sophisticated user.
>
>
>
> To continue Ray Nugent's comparison of petroleum and West Texas sweet
> light, when I buy gasoline for my car, I get the choice of Regular or Premium
> at the pump. Now, I don't know what goes into the regular gasoline blend___how
> much is West Texas sweet light, or Brent light or Saudi heavy oil___and, guess
> what, I don't care. The regular gasoline is good enough for my 8-year old
> Jeep and that's all that I need to know. Similarly, end users don't want to
> be confused by a huge menu of CPU speeds, memory allocations, bus speeds,
> etc. ___ if you keep it simple, then they will get it (as long as they can
> upgrade to a different configuration if needed).
>
>
>
> Paul.
>
>
>
>
> ________________________________
>
>
> From: cloud-computing@googlegroups.com
> [mailto:cloud-computing@googlegroups.com] On Behalf Of William Louth
> (JINSPIRED.COM) Sent: Tuesday, December 16, 2008 12:59 PM
>
> To: cloud-computing@googlegroups.com
> Subject: [ Cloud Computing ] Re: Compute surface as a traded
> commodity?
>
>
>
> I am not sure why we are looking for one single resource to meter. I
> would expect that various resource meters will be used as cost drivers in
> determining appropriate charges that will be passed up to the next layer on
> the cloud computing stack with each layer in the stack introducing its own
> meters derived partially from lower level meters - partially because there
> must be value added somewhere. Once we get above the bare metal platform I
> expect to see more diversity in costing and billing approaches. Currently we
> seem to have carried over a large amount of baggage tied to current (legacy
> in this context) enterprise system/network management approaches that provide
> very coarse grain resource metering at the process level or data traffic
> pattern levels. I am confident this will change to more (user/software)
> activity based costing with the metering correlated to actual software
> execution performed on behalf of the user or cloud service. Unlike our opaque
> OS based process containers threads of execution in the cloud will operate as
> lawyers do today - billing the client context for every activity perform
> using various meters (wall clock time, number of photocopied sheets, number
> of letters dispatches with postage,........). Threads will not touch a
> resource unless they have a client billing code. This will never be possible
> with ESM/NSM because one cannot see the computing above it and the other
> below it. http://www.jinspired.com/products/jxinsight/meteringthecloud.html
> Kind regards, William Christopher Drumgoole wrote:
>
> Given the variances in CPU clock speeds, Gigahertz Hour is easier to
> compare.
>
> ---
> Chris Drumgoole
>
>
>
> -----Original Message-----
> From: cloud-computing@googlegroups.com
> [mailto:cloud-computing@googlegroups.com] On Behalf Of Pittard, Rick
> Sent: Tuesday, December 16, 2008 11:50 AM
> To: cloud-computing@googlegroups.com
> Subject: [ Cloud Computing ] Re: Compute surface as a traded
> commodity?
>
> Actually, the price of a barrel of oil is for a very specific grade
> at a specific location. The real prices vary depending on quality and
> location - maybe just like a CPU-hour should.
>
> Rick
>
>
> -----Original Message-----
> From: cloud-computing@googlegroups.com
> [mailto:cloud-computing@googlegroups.com] On Behalf Of Jim Houghton
> Sent: Tuesday, December 16, 2008 9:16 AM
> To: cloud-computing@googlegroups.com
> Subject: FW: [ Cloud Computing ] Compute surface as a traded
> commodity?
>
> Interesting thread ... I had discussions with executives at a large
> investment bank (one of the few still around today!) as far back as
> 2002 when we were implementing large grids for risk & portfolio analysis that
> leveraged 'scavenged' resources for some of the compute footprint. I
> agree
> this will happen, but interoperability is not the only obstacle.
> Placing
> security off to the side - let's assume for the discussion someone has
> already overcome their technology or compliance hang-ups - there is a
> major
> business challenge to overcome.
>
> We all know what an ounce of gold, or bushel of corn, or a barrel of
> oil is
> around the globe. So what is the equivalent unit of trade for
> computing cycles?
>
>
> Think before you answer ... 'CPU hour' just wants to jump off your
> tongue,
> but as we all know not all CPU's are created equal (even by the same
> manufacturer). Then of course there's memory, bus speed, network
> bandwidth,
> network throughput, operating system, latency to/from your origination
> point, disk read/write speed ... I could go on and so can you. I've
> been living this for 6+ years working with clients who want to build internal
> utilities (clouds), and even there it's difficult to get agreement as
> this
> forms the basis for what they are going to get charged for the
> resources they consume. It's not much of a 'utility' if users got a flat
> annual allocation charge, is it? Yet that's by far the most common situation
> in
> large enterprises today.
>
> There's the closet economist in me who feels (hopes) someone will just
> start
> such a market and soon thereafter the laws of supply and demand will
> set the
> appropriate prices. Those with high quality service will be sold out
> and
> can increase their prices, with the reverse also true. However,
> especially
> with the current state of global economic affairs, I am doubtful it
> will happen anytime soon. Nor do I think we can count on any standards forum
> to
> tackle such an issue, and the major vendors will undoubtedly look at
> normalization (translate: commoditization) of their technologies as a
> bad
> thing.
>
> Anyway, hopefully this provokes some thoughts - look forward to your
> responses.
>
>
> Jim
> _________________
> Jim Houghton
> CTO and Founder
> Adaptivity, Inc.
> (845) 494-9419
>
> www.adaptivity.com
>
> -----Original Message-----
> From: cloud-computing@googlegroups.com [mailto:] On Behalf Of Simon
> Plant
> Sent: Tuesday, December 16, 2008 7:03 AM
> To: cloud-computing@googlegroups.com
> Subject: [ Cloud Computing ] Compute surface as a traded commodity?
>
> Bruce wrote:
>
>
> Will the "Cloud" ever become a pool of hosting providers who
> pitch
>
> their
> prices, SLA's and storage cost so customers will come to their "cloud"
> for
> services?
>
>
> I foresee a time into the future where the compute surface is
> virtualized
> and standardized enough that hosting contracts can be traded as a
> commodity
> on a market, rather than the RFP type process we have today.
>
> Such agreement would allow business to place a deal on an exchange
> much like
> FX today and get bids to run based on some parameters. IT hosters
> would price the deal with a spread in the same way as a currency trade today,
> the
> deal done in a matter of seconds and hosted for the duration of a
> contract
> window.
>
>
> If virtualization vendors deliver on their hybrid end-vision, this
> could be
> a reality of packaging workloads with SLA manifests and using internet
> vMotion-type tools to migrate. It would fundamentally change the way
> we write software frameworks and applications themselves to be more self
> contained and highly standardized to achieve the best 'tradability'.
>
> Interoperability via standards between VM platforms, portability of
> data,
> code business logic and processes are all key to how we build out the
> Cloud.
>
>
> Such openness may be a far extreme view, but would you want the
> opposite view of the world where switching costs and lock-in are extremely
> constraining and we are forever stuck in a platform cycle of
> distribute-and-consolidate?
>
>
> Simon Plant
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
> >
--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google
Groups "Cloud Computing" group.
To post to this group, send email to cloud-computing@googlegroups.com
To unsubscribe from this group, send email to
cloud-computing-unsubscribe@googlegroups.com
To post job listing, send email to jobs@cloudjobs.net (position title, employer and location in subject, description in message body) or visit http://www.cloudjobs.net
To submit your resume for cloud computing job bank, send it to resume@cloudjobs.net.
For more options, visit this group at
http://groups.google.ca/group/cloud-computing?hl=en?hl=en
Posting guidelines:
http://groups.google.ca/group/cloud-computing/web/frequently-asked-qu...
This group posts are licensed under a Creative Commons Attribution-Share Alike 3.0 United States License http://creativecommons.org/licenses/by-sa/3.0/us/
Group Members Meet up Calendar - http://groups.google.ca/group/cloud-computing/web/meet-up-calendar
-~----------~----~----~----~------~----~------~--~---
Posts
No comments:
Post a Comment