The cloud storage hype has been bothering me for some time (see Are there economies of scale in storage?). Even more irritating than the “storage as a service” meme.
The problem with cloud storage is threefold:
- The availability isn’t as good as a disk drive.
- The performance is limited by network latency and bandwidth, i.e. terrible by traditional storage standards.
- The economies of scale for the clusters behind cloud storage remain undefined – but the pay-as-you-go model has a distinct CapEx advantage.
In short, it is a new class of storage with a distinct set of benefits and problems. The problem for product designers: how to best conceptualize cloud storage.
Componentry
That’s where thinking of cloud storage as a component makes sense. It is a storage device with distinct levels of reliability, availability, performance and cost.
Flash, disk drives, DRAM, SRAM, arrays, tape are all storage components. They all get built into storage devices or even – in marketing jargon – “solutions.”
Why not a service?
Dismissing cloud storage as a lousy service – limited bandwidth, high-latency, subject to Internet squalls and provider goofs – isn’t enough. It isn’t a service because it requires a framework of enabling technology around it to be useful – which is why it is a component.
A car wash, haircut or a Google search is a service. You show up in your car, with your hair or a browser and a complete transaction occurs. A job completes.
A component is instrumental in job completion, but it doesn’t do the job. A disk drive is a component. It needs power and a front-end interface – USB, 1394, eSATA – for the simplest use case.
Service, component, who cares?
A key problem in research is asking the right question. A key problem in marketing is giving the right answer to the right question.
In this case, the right question is “how can cloud storage be used to create a compelling value proposition for our target customers?”
The StorageMojo take
The number of similar cloud services on the market suggests the wrong question is being answered. While there is a market for raw cloud storage – as Amazon’s S3 has shown – the real opportunity is incorporating it as a component – in a solution to a business problem.
This is an example of where the common technical meaning of the term “service” – the provision of a discrete function within a systems environment – differs from the common marketing meaning. As a result marketing and engineering are talking at cross-purposes – again! – about a developing market.
As the most successful consumer storage products of the last decade – USB thumb drives and the iPod – show, embedding storage into an attractive package is key. Amazon may be the world’s largest supplier of OEM cloud storage, but the real money will be made by those who build it into a convenient solution.
Courteous comments welcome, of course.
Hi Robin,
I hadn’t thought this way earlier. As you say, the key is posing the correct problem. I guess people will have a total different expectation from cloud if you take it as a component and not as a service.
S.Suresh
You’re on to something here. Cloud storage: slow. Cloud storage+compute: fast.
I have been talking about the cloud as a component, Modular Innovation, for quite some time over at The Product Guy. As companies evolve into whatever the next generation of Internet products becomes they will become more modular and make use of more and other Modular Innovations, from clouds to data portability and many other online products and features. I am recommending this article to my readers…
http://tpgblog.com/2008/12/26/the-product-guys-weekend-reading-december-26-2008/
If you are interested in a good introductory discussion on Modular Innovation and the increasing modularization of online products check out my article…
http://tpgblog.com/2007/12/06/modular-innovation-101/
Jeremy Horn
The Product Guy
http://tpgblog.com
You don’t differentiate between “cloud storage” as a business model (storage offered over an Internet connection) and “cloud storage” as a storage architecture (storage services exported by a collection of networked computers with attached storage devices). Amazon’s S3 is a good example of the former and something like NetApp’s GX with a dozen nodes is an example of the latter.
Your three problems (availability, performance, and economics) don’t really apply to the latter (well maybe economics) the challenge of “Storage as an internet service” are that there are so many variables between you and your storage you can’t ever fix the first two issues without massively impacting the third.
That being said, making existing assets in your already clustered computers available is a win (kind of like free storage) which is what Parascale does and of course if you are delivering web services anyway then having storage in that cloud as well is also a win (disclaimer I work at Google and have an advisory role with Parascale)
I have a picture in my office that I photoshopped (part of a preso I did at FAST) which has a disk drive with 5 head assemblies mounted around the platters. The problem with disks these days is they are acting more and more like DECTape in terms of good sequential performance and poor random i/o. Yes flash is better, to some extent the blended drives have a possibility to be better still, but the use of a thousand drives on a thousand machines works even better. Takes a relatively small amount of CPU away from what ever else they are doing and adds a lot of storage capability to your cloud.
–Chuck
“Service” and “solution” are such relative terms I don’t see that they are being used incorrectly here. Better reason not to use the terms are that they are so prone to misinterpretation.