Are you thinking about buying into one of these ‘storage forever’ programs? You know the ones, buy now with 6 years gold support and get free controllers or credits in 3 years’ time. You could be paying more and getting trapped into upgrades that offer little to no performance improvements.
The performance results in the LinkedIn post from John Martin started me thinking about this and some of the things you really need to look out for.
Before we go any further and as John mentions in his post, all of the following are using the SLOB Oracle IO generation tool from @ClossonAtWork, which is used by many vendors to prove their Oracle credentials, so we’re making a fair comparison here.
I’ll use the table from John’s post to summarise exactly what this is saying
Moving from Pure’s high end controller from 3 years ago to the latest and greatest high end NVMe powered system gives you a 3% reduction in Latency and 6% in IOP’s.
If you had bought into one of these long-term controller refresh programs then that’s it, 3 years in and only single digit improvements.
For the level of support that entitles you to free controllers after 3 years you also pay a substantial premium and I mean substantial, I believe there are other conditions as well. So, you pay a lot of money and get little to no performance improvements over the course of the 6 year contract, doesn’t sound so good does it?
How does NetApp compare over a similar period? We use 75% / 25% Read vs Write compared to Pure who use 70% / 30% so you need to make a small allowance for that.
Here’s NetApp systems that were launched back in 2016, using a version of ONTAP that’s also a few years old now.
And here’s the latest high-end system using FCP and then NVMe
And a table to simplify this
Just with FC in the last couple of years we’ve seen 16% then 34% latency reduction with 60% then 98% improvement in IOP’s.
Now switch on NVMe on the A800, no hardware to replace you just enable it, now it’s a 53% reduction in latency and a 233% performance improvement. Yes, that number is correct! a 233% performance improvement over our AFF8080EX baseline.
Before I joined NetApp the role that I had was quite broad, one part of it was to manage an outsourcing contract that we’d taken on for a client that had decided that IT just wasn’t what they wanted to focus on and they just wanted someone else to take care of it.
I was part of a small team that had to come up with the details of the contract for the outsource, fortunately my role was not the legal side of things but more to do with the customers objectives for the contract as we managed it over the next 3 – 5 years. There were effectively two big areas that we had to agree on, which are fairly typical for these types of long-term arrangement
- That we would reduce the cost of the agreement by 5% every year
- That we would demonstrate a number of enhancements in terms of the speed of delivery of existing services and that we would agree between us on a measure that would show the increase to the speed of their business.
Reducing cost and increasing speed were fundamental, and I believe you should absolutely expect this from storage vendors that are starting to offer these long-term programs designed to get you to make a 6 year or more commitment up front, with controller refreshes or credits thrown in, rather than the more traditional shorter cycles.
- If I’m being offered controller upgrades in 3 years’ time as a part of this higher cost, longer-term commitment, then I expect that the system will have a much higher level of performance than the one I’m currently using. I want to know that it’ll have a continuing positive impact to my business
- If I’m going to pay for something or even commit to it beyond years 1 – 3 then it should be cheaper than if I just go with 3 years up front and then make a decision as to whether I extend later.
As for the Pure benchmark numbers then these are lifted directly from one of their performance bragging posts here https://blog.purestorage.com/flasharray-x-sets-new-bar-for-oracle-performance/
I know some of you are thinking that this is just vendor bashing, even though the numbers speak for themselves, but I’m extremely proud of the incredible achievements that our engineering teams have made and want to highlight them. I also want to set your expectations so that if you’re thinking of going with a 6 year or more storage program then you should absolutely expect significant performance improvements, especially when performance is such an important factor for the workloads these arrays typically support.
I would never have agreed to an outsourcing contract that locked me in for an extended period, cost a lot of money and using historical data, showed me that I’d get little to no improvements as I went beyond the 3rd year.
Make sure that you’re not ‘ever trapped’ with your next storage purchase, be very clear as to exactly what you’re paying for, how much you’re paying for it and what you should expect from your new controllers in 3 years’ time.