ATCA Build vs. Buy

We had a great “online tradeshow” today, the Light Reading xTCA COTS Virtual Event. (Check out the recording after May 11th at this link.)

A lot of the discussion focused around the usual topic in ATCA: Build vs Buy. My position on this topic is nuanced: if you ignore R&D costs, then you can almost always build a blade that’s cheaper for your application than you can buy it on the open market BUT – and it’s a critical “but” – you can’t afford to update your blade as new silicon comes out, so you’re quickly one or two technology generations behind and soon paying a huge premium compared to your competition.

If you build blades internally, you’ll be lucky if you can get funding approval to refresh them every three years, and it’s much more likely that you’ll be using that blade for four or even five years; it takes that long for institutional memory to fade and people to stop asking, “Why do we need a new CPU blade? Didn’t we just finish building one?”

I looked back, and we at Continuous Computing have now released six generations of ATCA processor blades and we’re working on our seventh:

  1. XE20 (2004) – Dual Xeon
  2. XE30 (2006) – Dual Dual-Core Xeon
  3. XE40 (2007) – Dual Dual-Core Xeon with more I/O
  4. XE50 (2008) – Dual Quad-Core Xeon
  5. XE60 (2009) – Dual Quad-Core Xeon with Hyperthreading and better memory
  6. XE80 (2010) – Dual 6-Core Xeon

In the space of six years, we’ve averaged almost a CPU blade a year and seen roughly a 16X increase in performance. If you’re a network equipment provider building your own blades and still using 2007 technology and waiting for the next product cycle in 2011 to refresh your design, you’re paying 3X the cost per compute cycle of your competitors who are using the XE80.

It sure doesn’t make that internal blade feel so cheap now, does it?

There’s clearly a place for internal design, but in my opinion it should be focused on an area where you have competitive differentiation and the COTS market is under-serving the space – or the number of players in the market is so small that innovation and cost competitiveness isn’t lacking.

This surely isn’t the case in ATCA for compute, switching or packet processing, but is arguably the case in media processing – but that’s a subject for another post….

What do YOU think? Please submit your comments below.


Twitter icon