Grid services/utility computing is going to take off... in 10 years.

*NOTE*I originally wrote this Sunday - May 15, 2005

Some thoughts on grid and utility computing....

I think grid and utility computing is going to take off. In ten years. The pieces are starting to come out. They just don't knit tightly together yet. Also, people have to be convinced that this will work, it's secure and it'll save them money.

For example, in the solaris world today, if someone had a solaris 10 N1/grid/container/zone/what-ever-name-the-marketing-folks-dream-up-next world, you'd take a relatively low end machine, install solaris 10 on it, create a zone with zoneadm, install your software into the zone, test and validate the configuration and move that zone onto the grid. On the grid the end users would access what ever web server, app sever or database server you had running in it. Getting resource bound? Pay the grid owner to add additional resources of what ever you need. Maybe your already consuming the entire V890 you're running on, so you take a maintenance window to shut down your zone, deport the diskgroup (veritas lingo) and import the DG on a bigger piece of iron. Presto, no re-validation on the new hardware, no missing conf file or cron job that people forgot to tell you about. Because it's all in the container/zone it won't get lost. Because the hardware is abstracted, it's guaranteed to work. No risk! This can be done today, but no one, that I'm aware of, is selling "grid" services. I'm sure some one would do it for you, but today you'd probably be their only customer and everything would be a custom job just for you. So they'd bake in the growth costs into their fees and it probably wouldn't be any cheaper than doing it yourself, today.

Now, here's where the future starts the get real interesting. In the short term, I seep companies making their own production grids. Out of a cluster of mid-to-high end severs, say 2900's or what not. They use Sun or Veritas clustering to manage who runs where. This should happen in a year or two. This is where someone will see the cash to be made by running one gigantic grid for everyone to outsource to. But before that can happen, the security of grids needs to be improved and rigorously proven. The next piece is being able to move a container from one box to another with out shutting it down. Veritas is going to have this "real soon" I believe. Once their "UpScale" product, which they have working, is released you will be able to do just that. The next piece comes from Sun. Sun needs to make the servers meld into one seemless machine no matter how many actual boxes you have. Some of this is close already, but there's a long way to go. I'd say Solaris 11 at best. That feels like 2008 to 2010 to me right now. There is a big question if Sun can hold on that long.

Do I see this happening in the Linux space? Yeah, but only after Sun or someone else does it first. It' be foolish for the linux community do invest any significant effort into this idea until it starts to prove a viable model. Otherwise you tie up the kernel with lots of code that may serve no purpose.

Just a thought,

Rich