Ivan Pepelnjak published an article called Campfire story: Using the wrong tool for the job and it instantly got me thinking about something I've been discussing with my colleagues for the past few weeks, endlessly racking our brains trying to solve a scaling problem in our network.

Among other vendors, we had a few meetings with Arista and I must confess, the techie in me really liked their presentation.

Their switching platform looks solid: they use three types of merchant silicon depending on the feature set that you want (ultra low latency, packet replication, overlay support, programmability etc.), have very good port density in smaller RU formats, they can flash all of the LEDs on the box from the OS (was a recurring theme during the presentation, OK?) and their software is a fully-fledged Linux OS with a very flexible architecture. Wasn't that a mouthful?

Anyway, these things always sound extremely mouth-watering (if only electronics were edible) in slides and stories, but the question at the end of the day is: does it solve my problem?

On the one hand, they're a very specialized switching vendor aimed at DC infrastructures and low-latency environments. Not so much mobile operator/service provider.

On the other hand, there's the temptation to try to fit their solution (in this case a spine-leaf fabric with VXLAN) into your network just because it feels good (fast, scalable, programmable) and you like all the potential-for-amazing-things-TM it has.

To keep the story short, I realized that even if we deployed such a solution now, most of its potential-for-amazing-things-TM would sit there feeling sorry for itself at least 2-3 more years. And let's face it, by that time you'd be looking at something newer anyway.

That said, I will be toying around with vEOS (Arista's operating system), because it's easy to run it in a VM and it really made me curious. Maybe the first thing I'll do is make my switch chat to me via Jabber, because that's apparently something you can actually do.

So try to resist that temptation to get the fancy-tool-that-doesn't-solve-the-problem-very-well-but-has-a-lot-of-potential. Because let's face it, it's something we're all guilty of.

And, as always, thanks for reading.

comments powered by Disqus