Okay, so my search for the truth about software-defined storage leads inevitably to its conceptual/marketing progenitor: software-defined networks (SDNs). SDN, as a concept, has been around for a few years — a spin-off from virtual LANs (vLAN) and virtual SANs (vSANs) that everyone was talking about over a decade ago. As this very enlightening piece by David Strom explains, SDN has more potatoes at present than meat.
SDN is one of those technologies whose need is created by virtual server hypervisors that have managed to bollix up throughput and I/O in network and storage infrastructure. (That suggests to me that SDN is perhaps not the best way to solve the root problem (the hypervisor), than it is a way to cope with a symptom (log jams and inefficiencies in the networks and the inability of the network to modify routes through its interconnections to adapt to workload shift between network clients.
Remarkably, as Strom provides in the results of a survey, consumers aren’t necessarily seeing this as the shiny new thing. In response to the question “What’s Your Take on SDN?” the consumers responded in the majority that it was either “pure hype” (3%) or “too early to tell” (55%).
Apparently, the industry’s warmer view of the prospects of this marketecture have more to do with the big $$$ that funders gleaned from selling Nicira to VMware in July 2012. That deal made somebody rich: $1.3 Billion for a company venture investments valued at a scant $50M. The industry’s interest in SDN (or at least Nicira) seemed validated when eBay and a couple of other big names in the cloud space started leveraging that technology in their own nets.
Shortly thereafter, Oracle tried to capture wind from the VMware-Nicira announcement by purchasing Xsigo, a technology we liked alot when we reviewed it here. But as David Marshall pointed out, Nicira and Xsigo were two different kinds of technology aimed at different goals and objectives despite the fact that each company bandied about the terms SDN and network virtualization like my kids do the packing peanuts that come with the new storage kit. Apparently, the key difference is that Xsigo didn’t use an OpenFlow-based software controller (standards-based technology championed by Open Networking Foundation). OpenFlow is supposed to provide a standard protocol framework that enables consumers to use products from any number of vendors who support the standards in order to build the software-defined nets they want. Xsigo was just seeking to reduce costs and increase the utilization and efficiency of storage I/O – part of SDN, but not the full story.
Another important contrast, from our point of view, is that Xsigo has already proven to be capable of doing what it promises — minimizing cables and componentry by more than 70%, reducing connectivity costs by more than 50% and power consumption by more than 30%, and significantly reducing the need for VLANs, which is always a good thing. I couldn’t find anything stating what goodness accrued in real life to the deployment of Nicira and would like to learn more.
Another point from my preliminary research: support for OpenFlow protocol varies significantly, especially among the carrier nets, where I suspect we need support the most. After all, most of the folks pursuing SDNs, as stated at the outset, are using them in connection with x86 tinkertoy server virtualization, which is in fact creating the problem that SDNs seek to solve. From this paper in 2012, it would appear that the carrier nets are not yet on board with the necessary protocol support, and one, Verizon, has actually done a Gap Analysis that looks pretty damning about the readiness of OpenFlow for industrial strength usage.
I don’t want to get into the politics here. Carriers already offer a bunch of network management, administration, and performance tuning services for which they charge top dollar to (e.g., rob) their customers. SDN/OF might just help reduce those costs, but it depends a lot on the willingness of the carriers to support these technologies.