Simon’s Blog

Tim Bray is talking about Grids today. His article, On Grids, is worth a read if you’re interested in Grids. I have a couple of comments.

Let’s put that another way: memory space and compute cycles are pretty cheap. Disk space is effectively free. Moving data around in large quantities is expensive. In Jim’s case, he’s been working with big astronautical data sets. When he wants to send one to a colleague on the other side of the continent, he loads it onto a desk-side computer stuffed with 250G disks, puts that in a FedEx box and off it goes.

CPU cycles are cheap? Disk space is even cheaper, effectively free? I take the point that’s being made here, but it’s not entirely true. You have to look at running costs - at scale, running costs are a huge issue. If CPU cycles were so cheap, there would be no need for technologies like CoolThreads. However, it is true that hardware companies are addressing power consumption for servers. But as I was saying in my CoolSpin post a few days ago, power consumption on hard drives is now becoming a big issues for people working with large data sets - we need some new technology to address this. Think electricity bills in the millions of dollars per year, just to keep disks spinning. So, at scale, spinning disk is far from free.

The “moving data in large quantities is expensive” comment is interesting. This is absolutely true right now. For most people, moving large data sets around the network in the wide area is prohibitively expensive. However, bandwidth costs are falling fast, possibly faster than disk and CPU costs. So, in the future, we may not have to worry so much about bandwidth.

If you want to know what the future of grids might look like, take a look at the Worldwide LHC Computing Grid being built in the particle physics comminity. I was talking with the guys building this recently, and one thing that I found particularly interesting is that they get their bandwidth for free - which skews the economics compared to those of private-sector grids. What this means is that they can send data around the world at 1Gbps - just like a GigE LAN, except the data can go all over the world. And it does - you can see what’s happing on this grid in real time via a Java applet monitoring system (start the applet - NB it takes quite a while to start up [applet frame can take a while to pop up, and after that it can take a while to connect to the grid] so you need to be patient). If you’re running the applet, keep watching because every so often (perhaps every few minutes), you will see visualisations of the live 1Gbps data transfers happing on the grid (lines will appear connecting the nodes). The computing guys told me that the zero-cost nature of their network has significantly influenced the architecture of the grid.