As a veteran I/O expert at IBM, Tom Bradicich is as likely as anybody to know whether PCI Express will succeed in finding a home in next year's servers, as Intel predicts, or whether extensions to the current PCI-X specification will deter its advance. IT Week asked him to explain his vision for the future of I/O technology.
IT Week: There is a lot of debate about the relative merits of PCI-X and PCI Express - how do you see the differences?
Tom Bradicich: PCI-X is an existing technology, which differentiates it from PCI Express, which is a proposed standard. PCI-X is very evolutionary as opposed to revolutionary. Therefore, it is most prudent to stay on the PCI-X roadmap because the incremental improvement to an incumbent technology will always win.
Is compatibility with the base of Raid and network interface cards [NICs] the key advantage for PCI-X?
PCI-X 2.0 maintains backward compatibility with PCI-X adapter cards as well as software. As customers move towards [faster PCI-X speeds] they are easily able to adapt their migration. [Backward compatibility protects investments and this is especially important] in the economic environment that we have today. There is an additional level of manpower and admin needed to employ brand-new I/O technologies.
Despite PCI-X's advantages, some people predict that Intel will be able to drive IT vendors to adopt PCI Express - for example by gaining board-level support for it...
PCI Express is delivered in a core chipset. It's not the microprocessor; it's the economic complement. ServerWorks is number one [in server chipsets today] and is not using PCI Express in the near term. We have our own chipset, HP has its own chipset.
Is your faith in PCI-X based on that technology's strengths or perceived weaknesses in PCI Express?
I think PCI Express is a good technology - IBM co-invented it and co-developed it. But in my opinion its time is in the 2005 timeframe. It has a lot of good features [such as] the ability to serialise. Inter-processor communications is quite a vogue trend and I/O would be one of the last technologies to serialise. The raw technology's time has come [but] as with many standards, we have overlapping capability. You'll be likely to see PCI Express first in the desktop. I believe that in 2005, as the specification improves, we'll see the benefits of a more seamless transfer.
Will that mean systems supporting both PCI-X and PCI Express?
That dual support may become a necessary option. We're currently looking at how to manage that transition. [Comparatively] recently we have been eliminating the PC-AT connector; it would be very difficult to forsake PCI-X cards.
PCI-X 1066 with an 8.5GB/s bandwidth is being developed for next year. Will IBM continue to support PCI-X at that point?
It's too early to say, we're in evaluation.
reader comments