The hype cycle for grid computing started many years ago, and today it is mostly relegated to uses in bioinformatics, financial services, and 3d modeling (think crash testing, oil discovery, etc.). With yesterday’s announcement, Sun is making this grid infrastructure available to anyone, anytime, and on demand at Network.com. As Jonathan Schwartz points out, consumer plays have been driving the surge behind massively scalable web services. Think Google, Yahoo, eBay. On the enterprise side think about any dozen of SaaS vendors like Salesforce, LivePerson and Rightnow. Rather than building your own infrastructure, imagine being able to create your app or service and deploy it with no wires to pull, no datacenter or storage infrastructure to manage, and all with the frictionless use of a credit card or PayPal? Pretty interesting thought? I haven’t done the ROI analysis of $1 per CPU/hr (would love to see someone’s rough cut at this including operating overhead) but this certainly levels the playing field for scalable backends. Imagine if you are a startup and can’t get VC funding but have a killer app to deploy. Without any upfront capital expenditure, why not throw your service on the grid, pay per use, and build from there. That is a big concept, if it works. I believe this is the beginning, the very early beginning, of on-demand utility computing. With Microsoft moving behind software as a service, I see them deploying their own grid on their own stack on a rent per use basis. IBM will too. Competition will breed even better pricing and more opportunities for startups to focus on their apps and product and less about the back-end. I don’t see startups rushing to deploy their whole infrastructure on the Sun grid right away, but it certainly is worth looking at and monitoring over time. What has changed from 10 years ago is that the focus is not on corporate computing (those guys still want their own grids) but as Jonathan so aptly points out, the long tail-the renegade departments who don’t want to wait, the many startups that have new web services, etc. Ironically it was Sun that printed money from the VCs and startup community during the bubble-our checks went to a startup and they bought a Sun, Oracle, EMC backend. Well today the open source wave has killed that business and maybe this is another take for Sun to get at this capital. So how about that ROI analysis?
Published by Ed Sim
7 comments on “Grid 2.0”
Comments are closed.
In my opinion, the SUN Grid Services are really not that great for companies developing web-based software. Grid programming has been around for a bit with frameworks like Globus. The problem is that with a huge number of applications, computation is not the biggest bottleneck. Many SMBs developing web-based software have a far greater demand for top-quality, managed hosting solutions. If I were SUN, I would create a simple hosting service using their Niagara machines and offer the utility services as a premium add-on package. The Niagara architecture assumes a web-app environment. Accordingly, it sacrifices floating point performance and focuses on highly optimized threading. Unfortunately, Niagara machines are way to expensive right now for a smaller firm that is sensitive to capital investment. I know that SUN would argue that overall the initial investment makes sense from an ROI perspective. It probably does. Unfortunately, cash is king. Try getting away with the explanation, “It will pay off in the long run – I promise” to your board when they review your budgets. My company, as well as a ton of other companies would love this type of service. Moreover, I think that they could effectively sell more “utility” services” as premium add-ons to customers that have already purchased a hosting solution and need to expand their compute capability. Hopefully SUN will wake up and help us out because we all have money to spend.
Ed,
What Sun is not telling you is that their grid is only good for computational applications such as business intelligence, oil and gas exploration, simulations and special effectsfor movies. Sun’s grid is a great PR move designed to downplay the fact that Sun has fallen hopelessly behind in that business.
Jonathan Schwartz recently mentioned this in print as “winning Wall Street back”. What he really meant was that every large financial institution already has a grid. Those grids are built from commodity servers (think Dell and HP) and run Linux and other open source software. For example, JP Morgan is running a 10,000 server production grid, while Morgan Stanley is reportedly at 35,000 servers and counting. By comparison, Sun’s 5,000-server farm strikes me as a classic “too little, too late” response.
I agree with you on the impact utility computing will have on Web 2.0 startups, SaaS companies and many others. Keep in mid, though, that 90% of business and consumer applications today are transactional Web applications. You can’t run those on Sun’s grid. To address this kind of applications requires a new kind of grid architecture.
I have been talking with real-world customers on this issue for a couple of years now. Here is what a typical customer has to say, copied from Nicolas Carr’s blog at http://www.roughtype.com/archives/2006/03/grid_unlocked.php
“As the head of an IT department, I’d love to use the Sun Grid. I want to buy the minimum amount of cpu time I need to run my applications. I want to take 200 individual points of failure and move them into a grid where apps can be moved around to meet an SLA. I want to stop caring about the OS, the network, and the hardware. I want to only care about the application as the entity. I want to have an infintely expandable logical server to handle varying loads. I have been asking for this since 1997.
Utility computing companies don’t understand the question, or don’t have the capabilities to do this. The Sun Grid is a great idea. Try and run a database, an app server, or webserver on the Grid. You can’t. I could hire a managed service company to do all of this. However, they are going to do the same thing I am now, just with different people at a monthly cost. This doesn’t help me.
Finally, I want predictable variability. I don’t want to have to pay for $1/cpu/hr for a JVM to go haywire and consume an ever increasing amount of CPU time. I want to be able to predict costs monthly for budgeting reasons and to keep shareholders happy. I have this now based upon hardware constraints. On an infinitely flexible machine, aka Grid, I can have an infinitely variable monthly cost.
Also, I want a pony.
Posted by: Andrew at March 23, 2006 01:19 PM”
At my company, 3TERA, we have been building what customers like Andrew are growing tired to ask for: a transactional grid on which you can run web servers, application servers and databases. Without having to care about the OS, the network and the hardware. By working with the application as the entity. Without a single point of failure, and with predictable resource budgeting. The system is in Beta, with seven customers who have been using it for the last couple of months. In the last week, two of them called me and said they are ready to talk business.
Your vision of Grid 2.0 is right on the money. If you want to see it in action, send me an e-mail, and I will be happy to set up a demo for you.
As far as I know, Sun has offered this service in various flavors for at least a few years. Perhaps this is some kind of “relaunch”.
Hooman and Vladimir you are spot on. This is not for every web-based app and particularly suited for more algorithmic challenges, which as data volumes grow play a bigger and bigger role in this world. Think of all of the companies that gather massive amounts of data and run proprietary algorithms to deliver results (vertical search, zillow, etc.). That being said that is not the majority. Callidus Software is outsourcing its on-demand back-end to Sun and was upsold the grid to offload its monthly transactions. Vladimir, your back-end sounds more akin to what I wish for.
…from 451 group’s grid analysis and from what I hear from Sun insiders most big companies paid over 7$ bucks/CPU hour for grids. Most of this isn’t pure operating and capital costs, its design and architecture overhead. Down in Houston where lots of Oil companies buy this type of standard offer prices can get down to as low as .50/hr from IBM–but there are lots of minimum upfront charges wrapped around that.
The real cost savings comes from eliminating all of the high paid design/deploy guys from inside the enterprise and standardizing. Those are the same people who do vendor selection. Thats the big tension right now in enterprise IT.