Do you really have to slash and burn to upgrade your storage?

Whether it’s new storage architectures, software defined networking (SDN) or cloud computing, the assumption is you start with bucket-loads of cash and either a slash 'n' burn approach to your existing set-up or develop a green field site into which you can install the latest all-singing, all-dancing technology. But what if …

[Crazy Operations Guy]

Power and rack-space savings of virtualizing network gear

According to my tests in my datacenter there is no saving and in fact maybe require more rack-space and power. This was with two rows of 12 server racks and 1 network rack each; one set up with traditional networking kit (two datacenter grade top-of-rack 10-Gig and 2 1-gig switches in each rack and 2 many-blade core switch with routing and firewall modules with separate physical IDS/IPS and Load-balancing appliances) and the other row with basic layer-2 switches as specified by the SDN provider and the network rack replaced with a server rack.

Per-watt the traditional switches were far more power efficient per packet end-to-end between VMs and the exterior of the test network than the software-defined network and virtualized appliances. As far as testing for space, I only ended up saving a paltry 10 RU of space in a row of 13 server racks.

Overall, the added management overhead, power, and server usage of SDN cancelled out any benefit when using very basic layer-2 switches and virtualizing everything. I will be sticking with traditional networking and trunking all VLANs to each of the VM hosts and letting the network gear handle the packets.

[Trevor_Pott]

Re: Power and rack-space savings of virtualizing network gear

In opposition: I just compressed 12 physical hosts (each of which was already running ESXi 4.0 and several VMs) into two new servers. The new servers use half the power per node of the old cluster. Bonus: network visualization means I got to pitch almost all the network gear (was 6 switches and 3 routers.) That site is now 2 switches + 1 router + two compute nodes and finito!

And VMs move from that site to others without fiddling with the network! Doesn't that just beat all?

It's all about how you use it...