Building smarter networks [Day Three]

In the third and last part of Ian Yates's interview with Dimension Data's Paul Wilkinson, the pair consider how to implement gigabit and ten gigabit ethernet.

Ian Yates: You've mentioned obvious places that shift massive data, I guess they'll always be saying I can use more bandwidth, but do you think perhaps that before we get 1GB to every desktop it's going to require applications which drive the need?

Paul Wilkinson: You see 1GB Ethernet ports on every machine you buy today. The price difference between a 100MB chipset and a 1GB chipset has probably approached the point where it doesn't matter.

Ian Yates: Well certainly I know lower volume manufacturers don't bother; they just use 1GB chipsets because even if it costs slightly more, it's still cheaper than having two chipsets on the production lines.

Paul Wilkinson: That's right. Some of their customers are going to want 1GB, so we may as well give all of our customers 1GB and solve the problem. Yes I think particularly something like that which is really just some software and coding and would not be a physically different connector or anything like that really, so it's pretty straight forward. You'll probably see it take a little bit longer in the floor switch, as people come out to do a new building fit-out or revamp their hardware, they're going to look at it and say I can now get one 1GB blades for my switch that cost the same as the 100MB.

I was tossing this up at dinner with a colleague who works at Cisco and he said one of the problems is they insist on seven years of availability of all their parts, which in the IT industry is pretty hard because a lot of chip manufacturers are only looking for a three to four year life. He's saying that Cisco have got some models that they have shipped consistently for more than seven years. It's going to come to the point where Cisco's going to say we actually don't want to sell 100MB switches anymore because we can't get parts at a reasonable price. So 1GB versions will be cheaper.

The same thing will eventually happen with 10GB as well, but I think today if you don't need 10GB you probably won't pay the premium.

Ian Yates: Yes, I've heard people say that unless you absolutely need 10GB, it's probably still cheaper to run several cables and run 1GB on each.

Paul Wilkinson: Yes, whereas today if you try and buy an old 10MB Ethernet hub, you just can't. The $50 thing you get at the computer market is already a 100MB switch, possibly with a couple of gig ports thrown on the end.

Ian Yates: And of course, you can only go as fast as your cable allows.

Paul Wilkinson: A lot of people have been putting in Cat 6 cable for a while in new buildings so that would have been up to six years ago. Is it still going to work? I haven't had any experience with that, but I can imagine a situation where if the cable plant isn't up to scratch, then you may have to actually manually go through and configure everything to get reliability.

Ian Yates: The standard actually says that unless you've got Cat 6E you won't get 10GB at 100 metres from the switch. The accepted wisdom at the moment is that 10GB is going to be primarily for risers and data centres, and possibly there may be some move to get fibre attached storage using 10GB.

But then again you've also got the problem, if you stick a 10GB NIC in a standard server today, you may be bumping into a backplane that hasn't got a hope of keeping up.

Paul Wilkinson: Exactly. I remember when there were 100MB ISA cards. ISA cards couldn't keep up with that. But then there's one thing with network performance - it doesn't necessarily have to make sense to the purchaser. Quite often it's 'Ooh, it's 100MB, we'll have that'.

Ian Yates: I suppose it's a bit like buying a Ferrari when the speed limit's 110km/h.

Paul Wilkinson: Yes, or you live in Sydney and you can only go 20km/h anyway because of the traffic. But in theory, you can go really fast.

Ian Yates: But these days CIOs can no longer get away with the toy shop mentality. They're now a bit more constrained by business reality and they've got to prove to the boss that they're going to get some benefit somewhere out of this.

Paul Wilkinson: Yes, I think so. That's why I think it really boils down to the people who really need it will be the ones who purchase it. If I'm Channel Seven and I need to move a high definition movie from one side of the building to the other, the faster I can do it before it goes to air, then cost isn't really the problem in that situation.

Ian Yates: Yes, time becomes money so if it takes all day to move it across the floor and I'm sick of dumping it onto a disk and carrying it, then I'll pay the price.

Paul Wilkinson: Yes, and the same thing for hospitals. If I can get the radiology or imaging up to the operating theatre in less seconds then it saves lives and all that sort of stuff. But then I think in server farms, in the server room itself, they'll use the bandwidth. Maybe you've got a database server with fast storage, or you've got a web front-end for some application. It becomes a sort of escalating thing that if my clients are going to get 100MB throughput then my back-end has to be correspondingly faster to avoid the bottleneck. So within the server room, sure.

Ian Yates: Yes, but in other areas we'll wait until we have applications that demand it, and as you say, unless you just have to have it.

Paul Wilkinson: There's always someone who can justify more bandwidth.

< Back to Day Two

Read more on Data centre networking