Re: Virtualized Computing

So, I'll start this off with a disclaimer, a big, honkin', 80 point type disclaimer - my company makes a grid/virtualization software for oracle databases. I am one of the founders of that company, and consequently, I obviously believe in grid computing. But, I also have spent a lot of time looking at this space, talking to analysts, customers, competitors, etc. about what they're doing with grid and what it means to them. As another side note, we chose the name GridApp about 5 months before Oracle 10i became Oracle 10g, so nobody better accuse us of riding the Oracle 10g coattails.

All that being said, its important to note that none of the companies you cite (not even my company) are true grid computing offerings as the CS world is currently defining it. The business side of computing is massively disconnected from the academic/research side when it comes to grid computing. There are a couple of grid standards boards that are working out various protocols for building, managing, and interoperating grids. With a "true" grid, there is no single data entry point into the grid and no "master" or "coordinator" node. Every server can act as a data entry point, and every server works in concert to enable resource coordination. In addition, a true grid requires that every node in the cluster can immediately timeshare part of the workload automatically.

As you can imagine, this is a) hideously complex to design and b) more complex to implement, and was not even originally conceived for anything beyond scientific and financial computing. Those worlds have been doing grids for a couple of years now - at a conference I was giving a talk at once, an engineer from a large bank seized me afterwards and grilled me about strategies for enhancing data movement in and out of their 10,000 server financial modeling grid. They had more than enough horsepower to handle data modeling, they just couldn't move it in and out of the servers fast enough.

But, in the last couple of years, companies like Oracle, IBM, Sun, and a bunch of startups saw "grid" as a convenient way of combining a couple of messages they had been trying (with mixed success) to convey to their customers. Basically:

-Utility Computing - computing power isn't a widget you buy, like a server, its a utility, like water - you only pay for what you use -Server Consolidation - save money through centralizing resources -Datacenter Automation - companies spend too much time handling repetitive tasks, and those tasks take too long and are error-prone. -Virtualization - the server is no longer directly tied to the application running on it

Bundle all of these similar, but not identical, features together and you have yourself the business world's notion of "grid computing". I agree 100% that it is the most over-hyped and misused word of the year.   My company certainly benefits to a certain degree from the constant use of the term "grid", but I also spend a fair amount of time when I'm talking to management types clarifying just the concept of grid without even getting into what our software does.

I will say this, though - none of those companies you list are even close to giving the flexibility that is the promise of grid computing. Sun comes closest, but when you look under the hood, the only grid component is a toolkit based on the free and open-source Globus toolkit for grid computing. Oracle's notion of grid is one large database instance that has SLAs defined for different users, schemas, etc. IBM's is consulting services from IBM GS, plus some open-source clustering tech and a lot of marketing message.

Most of the really interesting work in grid for businesses (and especially in the database space) is coming from startups. I think my company has a unique and elegant database grid solution, as well as Topspin, which has a very nice toolkit for resource provisioning in large grids, and Gigaspaces has a neat java-based grid engine. Beyond that, a startup called Ejasent had a very interesting technology - it was swallowed up by Veritas and supposed to be reintroduced Q1 of 2005 (though with the merger, I haven't heard what's up with that). Vmware also has a very elegant product in their ESX server, but its not really designed for allowing distribution of resources across servers in the sense of the other products I listed. I'd be happy to go into the specifics of how any of these companies define their "grid" on request - I'm just giving examples of companies that are offering products that, to me, could add real value to a business.

You're very right - if a company buys a grid solution today they are either early adopters/beta or not really getting a grid. But from a long-term business perspective, enabling flexibility and automating resource allocation in the datacenter is the only way it will remain manageable. It's kind of a tough spot - businesses constantly need more computing power. If they grow by buying bigger and bigger servers, the hardware and maintenance costs threaten to overwhelm them.   If they scale out by buying many small servers, they're faced with having to manage an incredibly complex and difficult infrastructure. Lose-lose situation. Some kind of solution is needed to bridge that gap, and whether you want to call it virtualization, grid computing, capacity on-demand, whatever, its fulfilling a real need for a lot of companies.

It is just very disappointing to me that companies like Sun, IBM, and Oracle, who certainly know how to innovate and can build real products, resort to inventing marketing messages to jump on a bandwagon.

Thanks,
Matt

--
Matthew Zito
GridApp Systems
Email: mzito_at_gridapp.com
Cell: 646-220-3551
Phone: 212-358-8211 x 359
http://www.gridapp.com


On Dec 19, 2004, at 11:51 PM, Jared Still wrote:



> My apologies for this this somewhat general subject, but I

> could not think of a better way to state it.

>

> The over hyped buzzword of the year is 'grid'.

>

> Oracle has a grid, Sun has a grid, and now IBM has a grid.

>

> http://www.sun.com/software/n1gridsystem/

> http://www.oracle.com/technology/products/oracle9i/grid_computing/ 

> index.html

> http://www-1.ibm.com/services/ondemand/

>

> It looks like large proprietary systems are in the works.  It's deja vu

> all over again.  Kind of seems like the 1970's and 80's, only on

> a larger scale.

>

> My question is this:  Does anyone here have real world experience

> with grid technology?  By real world, I mean in production.

>

> If so, which one?

>

> Is anyone here involved in plans to implement grid computing?

>

> I'm personally a bit of a skeptic at this point, and see it as bleeding

> edge, with early implementers serving as quality control testers

> and troubleshooters.

>

> But then again, I'm in a small shop. I don't get out much.  :)

>

>

> -- 

> Jared Still

> Certifiable Oracle DBA and Part Time Perl Evangelist

> --

> http://www.freelists.org/webpage/oracle-l


--
http://www.freelists.org/webpage/oracle-l