Tuesday, September 05, 2006

PCI-X has legs


But how long are they? Dear blog posters, I’d love to hear what you think.

LSI Logic released six new storage adapters today. The most surprising thing about the news is that in this age of surging support for 2.5Gbit/s PCI Express half of the new cards were for the aging PCI-X slot.

The PC industry loves technologies that have all the costs wrung out of them. But development stopped some time ago on the PCI-X bus with little effort even going into the 2.0 version that hits data rates up to 533MHz. See the PCI SIG’s page on PCI-X at http://www.pcisig.com/specifications/pcix_20/

For it’s part, LSI said it will release no more PCI-X cards for the Serial Attached SCSI 1.0 spec, and it doubts PCI-X can handle the next generation SAS and Serial ATA specs that will deliver up to 6 Gbits/second. The SCSI Trade Association says 6Gbit SAS products should hit in the middle of 2007. See their SAS road map at http://www.scsita.org/aboutsta/Roadmap1192005v3.pdf

So the clock is ticking. How much time do you think PCI-X has? --rbm

3 comments:

FishyMcDonk said...

Let's see. The ISA interface is just now going away. Folks are STILL asking for 5V compliant PCI interfaces. In embedded applications PCI-X is just now hitting peak volume. At full bandwidth, PCI-X is a good 1 GByte/s - enough for nearly all non-high-end-storage apps. I'd say PCI-X has 10+ years left. Even in servers if 5V PCI support is any indication of the need for legacy support.

Mike said...

It is not surprising that I/O and system vendors continue to develop and deliver new products based on PCI-X. It simply reflects the reality of customer requirements and adoption curves. Certainly, PCIe adoption is increasing and some I/O vendors only deliver their products on PCIe but one has to examine the size of these I/O vendors and their target market segments. Some of the higher volume vendors are able to invest successfully in both technologies and continue to do so because they continue to make good money.

Taking a step back here are the questions to ask or areas to consider:

- Local I/O technology transition isn't as simple as the marketing implies. Keep in mind customers have 10+ years of largely blissful ignorance when it came to PCI and now they have to incorporate the significance a transition to PCIe represents. Fundamental business issues such as hardware / solution qualification, inventory management, troubleshooting tools, administrator education, etc. all directly impact the transition cost and time to execute. When one considers many customers have 100's to 1000's of servers, you can appreciate just how daunting the task can be. What is really being provided within the industry to help customers? It takes more than a pile of new technology to migrate a customer.

- PCI-X 266 is the top of the PCI-X offering. In spite of the marketing hype, that is enough to drive nearly all volume and a large majority of the niche I/O offerings. Proven technology is hard to overcome so the value being provided has to be more compelling than the new technology provides values that are largely focused on vendor problems than customers to date.

- Is there a technology inflection or set of inflections coming up that make the transition compelling? You point out 6 Gbps SAS / SATA but that only represents one of the fundamental I/O types required to deliver a server and may require PCIe. Projecting out, the industry will have 8 Gbps FC, multi-port FC / Ethernet, 10 GbE RNIC, etc. all coming to market. Add in PCIe gen2 signaling, new chipset advances for virtualization, native I/O virtualization, perhaps a new volume OS that exposes PCIe for the first time, etc. and you can start to construct a story with enough ROI that can motivate customers to upgrade their platforms.

Is 2008 the year when the industry starts to deliver pure PCIe for the server space? It is quite possible but don't expect to see it happen over night. Transitions take time and in an ever conservative market place with ever demanding customers, it may take until the end of the decade to see the majority of servers be pure PCIe. Even then, some fundamental problems persist. For example, server chipsets simply have too few Root Ports to enable platforms to be constructed without having to incorporate expensive PCIe switch chips which may be required for fan-out - something that is a significant drain on profit margins in an already thin margin business.

So, does PCI-X still have legs? Yes. Large install base combined with conservative often long time to implement upgrade process guarantee sales. However, as with any technology, it will become too expensive to continue the investment and people will transition to PCIe. How long PCIe lives on post this transition is perhaps the topic for another day as come perhaps 2012, I/O as it is known today, may be fundamentally different in how it is delivered and that can dramatically impact the control points and working relationships of the major and minor players within the industry - both hardware and software.

Rick Merritt said...

Thanks for chiming in gentlemen. Your combined wisdom tells me PCI-X has another 10+ years for new products in embedded and at least 2+ years in mainstream PC servers...great inputs!

Thursday, after some embargoed news breaks, I'll have a new wrinkle on this question, so stay tuned.

 
interconnects