My background over nearly 20 years has been Oracle Database, more recently focusing on the supporting infrastructure of large-scale databases. While many of the IT silos we know have come apart, even with Software-Defined Everything (SDx), knowledge of IT Networking has remained an apparent dark-art, at least to me. So, on the basis that if I feel this way, at least one other person does, here’s a trip around what I am calling “Networking, the Cognitive-Gap in Hyper-Converged Infrastructure”.
Let’s start the conversation by asking – why is it that many articles promoting awesome new 2U+ rack-oriented technology, that can scale to 10-100 nodes, make little mention of networking? I want to know more about what a ToR Switch and 10GbE Fabric are, and if you’re like me, today we want to be able to find a sufficiently technical level of information on our own, not have to approach vendors directly to ask for it means we don’t feel burdened by being sold to.
Given my need to know, I approached a former EMCer, Dennis Smith (now at Brocade), with my problem of understanding and I got this great reply;
Most hyperconverged appliance solutions today are delivered in a single rack. However, if you need multi-rack the most common architecture is leaf and spine with fixed config 40GbE spine switches. For us it would be the VDX 6740 at the leaf, VDX 6940 at the spine. We generally recommend using a flat L2 fabric for hyperconvergence (VCS fabric) since it’s the simplest to deploy, but an IP fabric using BGP or OSPF can also be deployed if desired. And if it’s just a couple of racks, a dense 10GbE VDX 6940 could be used as a “top-of-racks” switch (“racks” in plural).
Dennis also provided a couple of links to papers they prepared in collaboration with VMware and Nutanix, respectively.
These are great papers that I encourage everyone to read, they are very clear and discuss the networking at a level I can understand.
If you need more background on network topology, old and new, I found this post from West Monroe Partners helpful;
There’s also a really fun podcast that helps a lot to deepen understanding of ToR, Spanning Tree, Leaf-Spine, East-West and more;
In summary, this post is a journey to close a gap in understanding that the network is perhaps the most important component in hyperconverged infrastructure, that is the least well understood.
I want to assure everyone, the omission of similar material from Cisco, Simplivity, and anyone else is unintentional, I intend to become familiar with their offerings, and more. I am certain with the recent announcements of 2nd-generation offerings from EMC, HPE and now Cisco, Hyperconverged Infrastructure is about the get even more attention.
Researching this post has helped me better understand networking at the “leaf”-level, in a follow-up post I plan to explore the “spine” of Hyperconvergence. Hopefully with some great feedback from you, comments and links to more supporting material that I did not manage to find are most welcome.
Yours truly, @DBAStorage