Data Center Switch Requirements, Jimmy Ray's Top 5 List

Check out this pic! --------->
It looks like I am trying to pull out this SFP plug by looking in a mirror! Oh yeah..like I am a secret agent...code name: DeepFried...

Hey lets face if you have came to the point of needing a data center then your tolerence for downtime AND maintanace windows are very limited.
Anytime I have engineered a DC design, the requirements have went something along these lines:

1. Upgrades with Zero Service Disruption
And Zero means ZERO!! No fine print no shellgames. Upgrades and bug fixes are part of a day in a life of a data center admin. I can afford not to upgrade for additional features, but not for bug fixes.

2. Lossless Fabric. In the DC (that is the cool way to say data center), all switches are not all alike. A true data center class switch will NOT lose traffic on a passive backplane. This is a huge deal in the world of engineering design. Designing a cost effective high speed digital interconnect is a multi-disciplinary activity where tradeoffs between latency, power dissipation and bit error rate must be properly addressed. Once the system level architectural concepts are in place, one must then concentrate on the lossless factors namely the bit error rate. Major factors that impact bit error rate include conductor skin-depth losses, absorption losses in dielectric materials, and impedance mismatch and stub interconnect structures that have memory. By properly defining routing layers, eliminating stub structures, and applying compensation schemes, one can significantly extend the distance-bandwidth of copper based interconnects with a minimal increase in overall cost and arrive at a truly lossless backplane.

3. Operational Design. Hey, I have to work on this dude so it needs to be easy to service and support cable management for the new stuff. Try bending eight Cat 5e cables in place and now try the same thing with Cat 6a...enough said.

4. Lights Out Managent. A truly effective data center is like the computer room on the Star Trek NG, nobody goes down there because there is not a need. Lights out management is more then just SNMPv3 traps and management. Lights out to me means lights out on the product itself. If the product crashes or is at miniumal power, can I still access the device or do I need to dispatch a tech? When things crash in a DC, folks are running everywhere to fix the problem. Hey let face it, if I can solve an issue at the NOC without drawing attention to myself in that type of crisis, that is lights out management.

5. Forward Investment Protection. Going to the bean counters for money is something I enjoy as much as watching 8mm home movies of my mother in laws trip to the worlds largest ball of yarn. Anything that I can upgrade in place and add more and more value over time is an absolute must. I can afford to treat edge switches like disposible lighters, but certainly not DC switches. See point number one. My DC switches need to grow with my company and have lighting fast support time if I need it in a worse case scenerio. I like to sell this to the bean counters as forward investment protection, but truthfully it is really rear investment protection...as in my big ole tail!!!!