Quite often when people see the the unbelievable storage performance that can be extracted out of commodity x86 servers with SSD’s when combined with PowerFlex software — they literally laugh out loud.
Why do they laugh? Because usually it is way more performance than they could ever imagine using. However, we have a saying in the PowerFlex world and that is that performance is never wasted…
- Business results for your applications
Although your applications today may only be using ‘x’ amount of IOPS, there is very much such a thing as the “quality” of those IOPS. What do I mean by that? IOPS and latency have a relationship that is not always widely understood. Many vendors will say their systems that can deliver millions of IOPS* (Usually with an asterisk, always look for the asterisk), but unless you read the fine print you’ll see that these numbers are usually fairly dubious hero numbers. They will use unrealistic high queue depths and forget to mention the latency, or they will be random read hits from cache. In their defense, almost every vendor does this as they attempt to beat each other.
With PowerFlex on the other hand, the numbers that you see are true 100% random IOPS with no caching layer or any other tricks involved. With every Volume being spread out evenly in very small granular chunks over every single SSD in the storage-pool, you get this massive parallelism which means consistent predictable performance all the time.
Therefore the “quality” of these IOPS means that typically they are very low-latency / high bandwidth. This is what applications truly benefit from more than just high IOPS alone. This is what responsiveness looks like, and is something that your application users will soon get used to!
What does this mean for your business?
– Running more queries / analytical workloads without concern.
– Happier end-users, instantaneous like results.
– User retention for online use-cases. How often have you left a page that simply took too long to load?
– Running more frequent snapshots and backups to protect your data.
- New use cases.
Most storage owners are very conservative of what applications they will allow onto their shared block storage arrays. Why? They know too well what happens when a noisy neighbour starts running an intensive query that sucks all of the performance from that array. In order to avoid this, they either apply QoS limits which will delay those queries from being run quickly, or find a different array to run that particular workload on. Both ways, this leads to inefficiency across the business and a lot of manual balancing / tuning.
PowerFlex on the other hand being able to deliver ~250K IOPS per node (All-Flash with a high clock speed CPU), we simply say “do your worst”. While it has advanced QoS features, it’s rarely used in an enterprise setting as there is so much performance to go around that it simply doesn’t make sense to cripple those applications that can take advantage of it. There is no special tuning / balancing that is required, this is simply just brute force performance. And if in the extremely rare situation that you actually did need more performance – no worries, just add another node or two to your cluster — scaling up to 512 controller nodes is hard to beat.
With all of this performance, even the wildest ideas can now be entertained without fear of crippling the storage system.
- Ability to cater for unexpected spikes.
During COVID-19 we saw a huge shift to working from home – and many companies had to scramble into this new reality in a very short amount of time. Often this meant having to procure new servers and storage in order to accommodate for this sudden spike in demand. Except for Mr David Lee:
“PowerFlex supports the entire infrastructure our school needs in order to operate – from virtual networking appliances, surveillance servers, security systems and digital signage to our learning platforms, management information systems, administrative databases and even our own suite of Minecraft servers. PowerFlex’s massive performance and availability allowed us to seamlessly transition to ‘work from home’ and ‘off campus learning’ during the recent lockdown period. IT provisioning becomes quick and uncomplicated with PowerFlex; invisible to the end user and a dream for IT admins who need to provide robust and flexible solutions in an environment where the only constant is change. It just works.“
-David Lee, Director of IT Infrastructure and Development, Nexus International School (Singapore)
- Performance = Availability.
PowerFlex being an enterprise grade software defined block storage product, manages to deliver 6 nines of availability with just 2 copies of data. How does it do this? Simple – it does this with a many-to-many massively parallel mesh-mirror architecture. Rebuilds occur in minutes for individual SSD failures, and at worst an hour or so for a full node. This is very different from other storage technologies which can take hours and hours, if not days in some cases. It is during those rebuilds that data is potentially at risk and by minimising this window, PowerFlex is able to deliver this 6 nines solution. It is also why others need other RAID capabilities with dual-parity or three copies of data in order to achieve that same level of availability – and both of those have trade-offs. Parity protection involves a ton more backend writes reducing the life-span of your SSD’s. Having 3 copies of data is simply expensive and a waste of good hardware. And both of course, reduce your overall performance capability.
- Performance = Efficiency = Cost savings
PowerFlex gives a best in industry IOPS/Mhz — This means that for high performance workloads, it is going to have an unbeatable cost to serve ratio. It also means that in an HCI design you can spare the most expensive part of your environment (CPU and RAM) for what matters most (your applications). PowerFlex only uses the CPU that it needs, so if you have a period of low I/O, it will only consume low CPU.
Most other products end up requiring a lot more hardware to reach the same level of performance as PowerFlex, which means more rack space, more power and cooling, and more hypervisor/OS licenses. Often they might look less expensive up front, but once you have calculated the true costs this is definitely a case of you get what you pay for. Just by the simple fact that your average latency is going to be greatly reduced with PowerFlex, means that your CPU’s are also working much more efficiently and not sitting around all day waiting for data to process.
- Saving your precious time
This was only meant to be a Top 5 benefits, but I had to throw this one in too as it is simply too important to ignore (thank you Dan Aharoni for the reminder).
The most scarce resource of all – time. How does massive performance help you to save time? Most often it saves you vast amounts of effort in troubleshooting storage performance issues in the first place, which can be a huge drain of human resources. Application is running slow? With just a quick glance at the dashboard you’ll be certain it’s not storage related. And if in the very rare event that it was – more performance can always be added quickly and easily with just a few extra nodes, or maybe things could be re-configured slightly to help give additional performance where it’s needed most. This is the power of PowerFlex.
I often try and think of car analogies and what would PowerFlex be — how about if Ferrari made a tank, which could also do 100 mpg while cruising at 300 mph, AND never had to stop for maintenance! If you’d like a test drive, let us know — we’d be more than happy to put you in the drivers seat.
Matt Hobbs – APJ PowerFlex Presales – Dell Technologies – July 2020