Nutanix had a bit of a rough start to their presentation. They were asked to provide a bit of background on the company, which is fine, but the main issue I had was with the length.
My impression was that Greg Smith didn’t understand his audience well enough, or maybe he was aiming at a different one than the Tech Field Day audience. He gave a pretty generic marketing type presentation designed to impress CIOs. It was full of social proof indicators – impressive names of their financiers, big tech names like Google as being similar technology – all designed to convey safety to people who might be thinking of buying a relatively new product. There were waay too many buzzwords.
Greg was true to his word: he said he’d take about 20 minutes, and he did. I reckon he could have boiled it all down to less than 10 minutes, so having plenty of time actually worked against him here. I’d expect the messaging to get much crisper in future. There was good content there, but it was surrounded by a bit too much fluffy language and guff that really didn’t add anything to the presentation. People rarely complain if you finish early.
Sadly, Binny Gill suffered from a similar “too much time” problem, and I guess felt the need to fill his slot with something. A lot of it was spent re-defining the problem to an audience that is very well aware of the problem. *sigh*. Eventually we stepped in to move him along, because we’d spent about an hour hearing a pretty standard sales pitch from Greg and Binny. Unfortunately, Binny kept going with what he’d prepared.
This was very much a problem of not respecting the audience. I can give a pass to a brief intro for a non-technical audience because the session was being recorded, and some folks would need a quick primer, or might not have heard of Nutanix. Ok. Greg did that, so I have no idea why Binny felt he needed to give the presentation he did. A kind interpretation would be that he wanted to make sure everyone had the right background to understand the discussion. To be harsh, I’d say he didn’t spend time preparing and phoned in a presentation that had been done before, possibly by someone else.
It’s been over a month now since TFD9, and the YouTube videos for Greg and Binny’s presentations have about 280 views total. There are over 535 views for the detailed tech demo from Steven, so I think that explains a bit about the relative interest in the material from the Tech Field Day Audience. If only the time allocated by the team had reflected where the audience’s interest lay.
NoSQL and PAXOS
Nutanix is essentially a shared-nothing architecture using commodity hardware and some custom software. It’s one of the (scarily) numerous storage companies that have taken the technologies developed at Google, Facebook and others and turned them into a commercial product suitable for enterprise IT.
The hardware is all x86 based, as most stuff is, so it’s relatively cheap. It scales well, because you just add more nodes, and the smart software handles data replication. Lots of nodes makes read-access scale well through striping, and writes can be sped up using some other clever software techniques (such as Infinio!)
I particularly like the way Nutanix use the PAXOS group of protocols to manage the metadata. It’s part of the trend towards acknowledging that hardware failures happen. Rather than trying to get software to pretend that hardware is perfect and spending huge amounts of time and money trying to make it look like the hardware never fails, we just write software that handles the failures. At scale, failures are assured, so being fault-tolerant is a vastly superior, and more cost-effective, solution than trying to make hardware perfect.
It doesn’t suit everything, but it’s fine for well over 80% of what most businesses do. In fact, it’s probably fine for over 95% once you start to make realistic, empirically verified cost/benefit tradeoffs.
There are a couple of things that concern me, such as the way the hashing function allocates data to nodes (do watch this presentation on database scaling by Jonathan Ellis of the Cassandra database project, from PyCon 2010), but I’m sure the boffins developing the software are orders of magnitude smarter than me, so it’s probably fine.
Steven Poitras did a great job showing off the tech of Nutanix. If you want to see what setting up and using a Nutanix cluster looks like, go watch the video. It’s well worth your time.
The browser based console is quite pretty, but the thing that excites me the most is that they’re implementing a REST API. The API browser, to help you write your scripts, is such an eminently sensible idea I boggle at why other companies haven’t got one.
Having a REST API makes it much, much easier to integrate the Nutanix gear into an orchestrated, automated environment. It’s where the industry is headed, and it’s great to see Nutanix on the leading edge of this trend. Expect to see a lot more of it in coming months and years.
I like the look of Nutanix. For distributed type workloads, and highly virtualised environments, it looks like a sensible choice. The REST API, and the live API browser, is one of the features that would tip me over the edge into choosing Nutanix over a different brand of storage, all other things being equal.
I’m a little concerned that some of the marketing people are still trying to be all things to all people, when that’s not what Nutanix is good at. I hope they focus on the markets that the technology suits best, which is not vertically scaling databases.
Again, most company stuff isn’t that, so don’t get drawn into discussions about where Nutanix isn’t suited. If they play to their strengths, they’ll do well.