This week, Dell EMC have announced the availability of their latest and most powerful storage array to date, the PowerMax. The announcement took place at Dell Technologies World in Las Vegas, where I was present as an industry analyst and blogger / influencer. We learned that the PowerMax will be offered alongside with the VMAX series, so what is this new product announcement about, what market it is targeting and how to make sense of the high-end product line at Dell EMC?
Introducing the PowerMax platform
The PowerMax is an end-to-end NVMe all-flash array (AFA) that marks Dell EMC’s return to the Tier-0 storage market after the short-lived EMC DSSD platform was retired.
If you wonder what the Tier-0 market is all about, think of it as the Pebble Beach Concours d’Elegance, but for storage arrays – with one twist: beyond the elegance of designs, Tier-0 arrays serve a very specific purpose as the industry workhorses that must process the most intensive & demanding workloads. In the Tier-0 market, arrays must support systems & applications that ingest and process massive amounts of data at the lowest latencies, almost in real-time. Low latencies, high resiliency and sustained throughput are characteristics that are expected to be found in Tier-0 storage arrays.
The PowerMax is not a successor to the VMAX. It is a new portfolio addition and the VMAX platform will continue to live on as Dell EMC’s Tier-1 storage platform for mission critical workloads. The PowerMax was partially built on the VMAX platform achievements, but is very different for several reasons:
First of all, because the PowerMax leverages NVMe (and is also ready for SCM – Storage Class Memory), hardware adjustments were needed. The PowerMax sports dual-ported NVMe drives, meaning each drive is connected to two PCIe controllers for better resiliency. These drives use a different interface than SSD/SAS drives, which has an impact on redesigning those interconnect elements on the chassis. The extremely fast operating speeds of NVMe also require the use of an InfiniBand backplate to sustain the data throughput. To achieve greater performance, the inline deduplication and compression features were also offloaded to a dedicated processing module within the array and are no longer processed by the onboard CPUs
Secondly, software improvements were needed as well. In fact, new code functions had to be developed to properly leverage the hardware improvements brought forth by the NVMe architecture, to implement in code the changes to the dedupe / compression functions, and last but not least, Machine Learning (ML) was introduced with PowerMax, a first for Dell EMC storage arrays.
Those comprehensive changes and the tight coupling of the major functions with the hardware make it difficult, if not impossible, to backport the PowerMax OS (which is actually called PowerMaxOS, whee!) to the VMAX platform. For the time being, Dell EMC have decided that the VMAX series will not get the benefit of the ML functions. Other additions such as the hardware implementation of compress & dedupe make it also more difficult to happen.
A glance at the PowerMax features and architecture
Obviously, being an end-to-end NVMe AFA means that the PowerMax operates in the Tier-0 market (as stated above) where extremely high IOPS (in the millions to tenth of millions range) and low latencies are the norm. Compared to a VMAX AFA using SAS Flash, the PowerMax achieves a 25% improvement in response time with current NVMe Flash, and should be able to achieve up to 50% better response times once NVMe SCM modules will become available (Dell EMC is currently evaluating Intel Optane, per the informations I have on the matter). Thanks to the use of InfiniBand, the entire system can sustain a throughput of 150 GB/s.
From an architectural perspective, the PowerMax platform builds on what has made the VMAX the success story and reliable platform it is (see my blog post on the VMAX 250F/950F to learn more). Systems start with 1 brick (13TB, 10U) and scale up to 8 bricks with up to 4PBe per system, based on whether the customer goes for the PowerMax 2000 or 8000. The PowerMax 2000 offers 1.7M IOPS, 1PBe capacity, 1 to 2 Power Bricks; the PowerMax 8000 offers 10M IOPS, 4PBe capacity, from 1 to 8 PowerBricks. Also worth noting, the 2000 is a dual-controller platform, while the 8000 is a multi-controller platform.
On the data reduction front, the inline dedupe & compression algorithm, a function that is now offloaded to a dedicated module, helps achieve up to 5:1 dada reduction ratios. The array also allows for this feature to be turned off/on granularly, at the application level – which I suppose will mean at the dataset / LUN / whatever container level. Last point worth noting, the use of NVMe allows for an up to 40% lower power consumption.
PowerMax – The Rise of Autonomous Storage
A couple years ago I was debating about what the future of storage arrays would look like. It seems we are slowly getting there: Dell EMC leveraged the data they collected from their largest VMAX deployments to build their own Machine Learning solution.
The ML engine (which doesn’t seem to have a name yet) leverages predictive analytics and pattern recognition to make data placement decisions. Depending on how much a PowerMax array is utilized, the ML engine can drive up to 6 billion decisions a day. While we do not have more precisions, Dell EMC states that the ML engine analyses and forecasts 40 million data sets in real-time (is that locally?) and is currently analyzing over 425 billion data sets in real-time. I hope to get more clarifications on the matter to improve this section of the article.
Obviously, most of the data is processed and kept internally by PowerMaxOS, with only relevant data that helps enrich the ML engine being sent back to Dell EMC.
Until now, Dell EMC customers lacked a Tier-0 offering and had to look for alternative options; Dell EMC was cut off from a market that, while still relatively niche, is deemed to keep growing and growing as the data explosion trend keeps following its hyperbolic course with no correction in sight. With the PowerMax, Dell EMC makes a noted and most welcome return into the most prestigious segment for storage, Tier-0 – a segment that they had pioneered with the powerful but short-lived DSSD array, a solution that was in advance for its time and too early on the market. With the PowerMax platform, Dell EMC now has a full portfolio offering for every storage type & tier.
Customers evaluating the PowerMax will be happy to know that thanks to the experience coming from the VMAX especially on the long-standing reliability of that platform, which has been use for many years as the foundation to run mission-critical workloads. PowerMax expands these capabilities into the Tier-0 world; it offers enterprise-class support and the guarantee of a strong brand to customers who look at emerging startups in the field with suspicion and fear. It’s not that large companies never kill off products (no need to look far with the DSSD story) but the market has reached the maturity for such solutions, demand is there, and PowerMax comes out just at the right time.
In my view, there are very good chances for PowerMax to thrive, and eventually in the mid to long run to supplant the VMAX offering as ML gradually becomes a standard feature of storage arrays, and as NVMe adoption grows. I’m convinced that the technology we see today in PowerMax will make its way, when relevant, to the Tier-1 range (if it doesn’t ends up supplanting it) and eventually even in the midrange storage, at least for ML / predictive analytics functions.
I was invited to Dell Technologies World by Dell. Dell covered the cost of travel and accommodation during my time at the event. I have not been financially compensated for participating; the invitation was made without any obligation to blog or to produce content. Any content you see posted here is created willingly and is the exclusive product of my mind. Content aims to be objective and will be produced without any moral obligation to please Dell or any of its subsidiaries.