One of the most common requests that DCIG gets from its readers is to include the actual cost of storage systems in its Buyer’s Guides. The reason DCIG continues to decline that request and only includes starting list prices is that most storage systems may be configured in multiple different ways. This makes it impossible to arrive at a definitive price point. The second part in DCIG’s interview series with iXsystem’s Jordan Hubbard illustrates this point as he discusses how the availability of multiple different storage configurations and services trumps a cookie cutter approach to buying storage every time.
Ben: What is the theoretical limit to deployment and what is your largest deployment?
Jordan: I do not have immediate access to our complete customer database, but I could definitely say it’s in the petabyte (PB) range. I have to be a little bit vague. The limitations, as it were, with the size of such deployments are primarily hardware. There are only so many bytes you can push through a 10GbE interface. Even if you want to link multiple 10 GbE interfaces together, you can go no more than 12-16 ports, and that’s assuming you have the maximum number of slots available to put the appropriate multiport 10 GbE cards in there and don’t mind the fact that PCIe bus bandwidth limitations means you’ll never even approach that theoretical maximum.
Even in the highest end configuration, there is always going to be ingress and egress bottlenecks to contend with. Even with all of the network bandwidth you could possibly want, the number of spindles that one can physically fit into a box or daisy chain across a reasonable number of JBODs without compromising reliability is going to put a certain upper limit on the number of IOPS you can get out of a single head. There are also single point of failure arguments to be made that might discourage organizations from adopting this approach. Once you get to a certain stage of hanging a given number of enclosures off a single head — or even a pair of heads where you have some failover capability — you start to ask yourself, “How many more eggs do I want to put in this one basket?”
At that point, you’ve probably come to the conclusion that you should start spreading your storage out into multiple racks, and once you start doing that and are confronted with physically distinct boxes, you start asking yourself, “How do I want to manage all this data?”
Maybe that’s a better way of putting it. The sheer data management challenges are the bigger bottlenecks. To be honest, no one has really cracked that nut yet.
Ben: Do you provide services to help organizations come up with those configurations?
Jordan: iXsystems obviously does a lot of consulting, and at the front end of its sales process is a lot of Q&A. We ask:
- What are you trying to do?
- What is your workload?
- How much data are we talking about?
Even for just one box, there are a lot of key decisions to be made.
- How much ARC do you need (e.g. How much memory?)
- How much L2ARC (secondary cache) do you need?
- How much SSD secondary cache is appropriate?
If you are doing a lot of NFS operations then a ZIL, a ZFS intent log, is probably going to be important for getting the maximum performance out of just one box. Then, of course, once you start scaling to multiple boxes, other concerns emerge such as:
- How are you distributing the flow of information?
- How are you dealing with replication and backups?
All of these questions are asked during our consulting and pre-sales process. Nobody has truly cracked the storage “blob of mercury” model which everyone would love to have, where you add another box to your network and it just seamlessly disappears into the one big blob of data and no management of it is required whatsoever – just frictionless scaling. Some of the storage players are touting that in their roadmaps, but no one is actually providing it.
Ben: Is configurability a big part of your value proposition?
Jordan: Yes. A lot of our customers come to us because they have kind of fallen out of the templates that the other folks offer. “None of those quite meet my needs.” Or they might, but they’re over-spec’d or under-spec’d, and the customer would have to do some gyrations to be able to fit into one of those configurations.
For the bigger players, their configurations are often more of an exercise in cutting off everything that does not fit, so the bespoke storage market is definitely an area where iXsystems plays well and there is still a lot of room left to play. Obviously there is also the cost competitiveness that comes with being smaller and more agile, with support plans that can be tailored to smaller customers. That’s another area in which iXsystems competes well.
I think the most important attribute is the personal service, however. Being able to call and talk with someone who can understand your storage needs, someone who will work interactively with you to define what is precisely the right storage solution, all of that is a big part of what iXsystems offers. A lot of people definitely come to us without a clear notion of how to address their storage problems, and we help them find that clarity.
This is just not something people get from seeing a couple of fixed product tiers and generic marketing text with little to no clue as to whether or not any of it will actually solve their problems. Customers often start with only the most basic problem statement as to where they want to store a certain kind of data or roughly how much of it. Since they are frequently not storage administrators, they need someone to map their problem to a solution.
iXsystems offers a very compelling proposition. We will actually figure out how to solve your problem and map it to a solution that will work for your application. A cookie cutter approach to buying storage solutions really just doesn’t work as well!