Best design information of the week: Chris Marget write a series on practical data center network designs (10GE servers connected to Nexus 5K) – Part 1 describes ToR Nexus 5596, Part 2 ToR Nexus 5548 and Part 3 a pair of Nexus 5548 with ToR FEX.
Kill the Telcos – Save the Internet by Todd Hoff is a fantastic summary of an hour-long ranting between Jan Žorž, Greg Ferro and myself. I’m always amazed at how much some people can explain in just a few paragraphs.
How buggy can you get?
As if Greg’s and Ethan’s rants wouldn’t be enough (they’re both a must-read), here’s another bug: when you aggregate BGP routes and suppress the more specific ones, the suppression kicks in after a while. In the meantime, you have a nice route flap. Obviously the guys speeding up the BGP path selection code forgot an ancient code path that’s still triggered only every 30 seconds or so ... and of course some had to add another knob you can twist.
STP is not the problem, but FabricPath will fix it. A great overview article.
Virtual Subnet (VS) – another attempt to do L2 DCI, this time involving MPLS/VPN and BGP
Why I’ve finally had it with my Linux server. A fine Linux-distros-stink rant. Unfortunately most of the article is true.
Hadoop network design challenge – Brad Hedlund explains Clos architecture using real-life data center designs. This article inspired me to add the Clos architectures to the Data Center Fabric Architectures webinar. Thanks again, Brad!
IBM launched a 10GE switch with OpenFlow support. The specs look almost exactly the same as QFX3500 or Nexus 3048 with one important difference: IBM is the only one of the three companies mentioned in this paragraph that cares about IPv6 in the Data Center.
Infrastructure scalability pattern: partition by function or type. Another great scalability 101 article with the usual “this is how you do it with F5” twist.
In what ways is AWS better than its competitors? Part of the $1B answer: “they aren’t trying to recreate legacy DC environment.” So obvious, and so obviously ignored by everyone else.
Why will OpenStack falter asks the right question: what makes an open-source project successful. Read it, think about it, and apply the same reasoning to OpenFlow.
Why we moved off the cloud – the intractable problem is the performance variability.
Stack Overflow, a true cloudless bare-metal shop, is considering the merits of virtualization. In the end, the great answer to life and everything is “it depends” (or 42, if you prefer the HGU encoding).
From Monopoly to Marketplace – interesting perspective on the importance of pricing in private clouds.