First Impressions With PowerMax 2500 and PowerMax 8500
In this blog
Thirty-two years ago, a company called EMC released the Symmetrix 4400 Intelligent Cached Disk Array. Since then, it has been deployed in the most demanding data and security environments. The platform has continually reinvented itself to stay at the leading edge of the enterprise storage world. Introducing the world to synchronous (SRDF/S) and async (SRDF/A) replication, flash storage, and tiering (FAST), Symmetrix and its descendants have set the bar for what enterprise arrays are capable of.
That statement is not to diminish any of the achievements of the midrange systems; they all have a lot to offer, but at the scale at which Symmetrix/VMAX/PowerMax operate and the workloads they service, it operates on a different level. It has provided true active/active controllers, scale-out upgrades to add additional processing power to the system and nonstop operations for decades. That brings us to the latest addition to the family, PowerMax 10, the tenth version of the platform, and its two models, PowerMax 2500 and PowerMax 8500.
Announced publicly at Dell Tech World 2022 in Las Vegas, today it is officially available for ordering. See our Dell Tech World debrief for more of the key new features. While building on everything that has come before, it is (to this writer) the most significant platform update since VMAX3 in 2014. Read on to learn why I believe this is such a big update.
As a participant in Dell's beta hardware and software programs, WWT has spent the last nine months testing various pieces of the new system, including a 2500 array. In addition to geeking out with pre-release hardware, we provide feedback to Dell's engineering teams on what works well and what we think needs tweaking; the goal is to provide our customers with a better overall experience. It is an excellent example of our partnership with Dell and a wonderful experience. All that is to say, the perspective you'll get is from an actual user.
PowerMax 10 carries forward the same management systems it's had for years: Unisphere, SYMCLI and the REST API.
SYMCLI will be familiar to those who have been around the platform for any amount of time. It can work with gatekeepers (small control devices) allocated to hosts or VMs, or in client-server mode. As before and maybe not well-known, client-server mode can use the array's control station as the server in the relationship, meaning you don't need to allocate gatekeepers to anything.
Likewise, the REST API is a full-featured way of programmatically interacting with your array. You can use the API as-is or work with Dell's wonderful PyU4V. Furthermore, they have Dell-built and supported Ansible modules.
Unisphere 10 is the latest iteration of the array's graphical web-based management system. A tall order is to design an interface to provide simple, visual management of an array capable of 64,000 volumes and 64 million snaps, mainframe, open systems and multi-site replication. Unisphere simplifies all standard provisioning tasks, including metro-replicated volumes. It also provides access to the system's deep performance metrics to monitor nearly every aspect of the system and speed time-to-innocence when a performance issue arises.
There are some significant changes to the guts of the system. The product line has always been cache-centric; it's how the array can provide great read and write response times and is where the system's metadata lives. To get the system to scale larger while keeping excellent performance, they had to find a way to add more cache into the system. Intel's Apache Pass PMEM is how they accomplished it. By adding significant amounts of Optane onto the memory bus, they moved metadata onto it, allowing less DRAM to be needed while storing significantly more metadata.
Dynamic media enclosures
The simple DAE has gotten an upgrade. Instead of a tray of disks captive to a director pair, PowerMax 8500's dynamic media enclosures (DME) are connected to the backend fabric. The fabric has been upgraded to 100Gb/s NVMe RDMA over Infiniband, meaning the directors' CPUs have less work to perform due to the nature of RDMA. Also new are self-encrypting drives; the controllers are no longer responsible for encrypting the data on disk. The most significant change is that any disk director can contact any DME to perform IO. This move frees up the drive enclosures to be independent of the nodes' director boards.
This is the first edition of the Symm/VMAX/PowerMax product family that disaggregates RAID data protection from the drives. Like Unity's Dynamic Pools and PowerStore, the array manages data placement, maintaining parity as it sees fit, rather than a RAID set belonging to a group of drives. Furthermore, since RAID isn't dependent on a set number of drives, you can now upgrade the system one drive at a time if you wish; it will rebalance data after an upgrade. Additionally, instead of dedicated hot spares, the system now carves spare space off the top of each drive. This allows for many-to-many rebuild operations in the event of a disk loss. Dell claims about 1TB of rebuild in ten minutes or an entire 3.84TB drive rebuilding in 38 minutes, which is 50 percent faster than the previous generation's already outstanding rebuild performance.
Frontend director convergence
Going back in history, VMAX assigned CPU cores to functions: one core went to a frontend port, and one core went to a backend port. When one CPU got overworked, there was nothing you could do about it. This led to ~60 percent being the maximum utilization you'd want to see for any given system resource. VMAX3 created core pools to be used by the various emulations in the system. A set of CPU cores was captive to their assigned emulation. Dell's support and engineering teams could move these around to tune for certain workload environments, but there was a downside. Users could not spin up or spin down emulations; Dell created the emulations, and after that, users could assign ports to those emulations. In practice with core pooling, you could now run your frontend emulations at 80 percent utilized without too much worry.
PowerMax 10 allows users to have a system starting with a single frontend emulation and then set new emulations to other ports. For example, I started with a pair of Fibre Channel ports on a SLIC pair. Then, I assigned two ports on each card to SRDF without an existing SRDF emulation running. The same is true for IP cards, now running at 25Gb/s; I can have no emulations running on them and assign iSCSI to one pair, NVMe over TCP to another and SRDF replication to a third pair.
NVMe over TCP
NVMe dramatically reduces the workload on the system when it comes to performing IO operations. In Dell's primary storage platforms, it launched early this year in PowerStore 2.1. PowerMax 10 gets this new frontend emulation and, coupled with 25Gb/s IP ports, it brings powerful and cost-competitive performance to standard IP networks, with the large number of frontend ports for which the product line has been. NVMe over TCP setup is simple. There is a wizard that walks you through all of the steps in just a couple of minutes. PowerMax integrates with Dell's SFSS, which provides service discovery for IP networks (if you're familiar with Fibre Channel zoning, the concept is similar).
The same codebase used to drive NAS on PowerStore is now running virtualized within PowerMax 10. It is tightly integrated into the system in the same interface used for managing the rest of the array. NAS has been promoted to a first-class citizen. SDNAS runs active/active on each node pair and can utilize the same IP port cards you use for iSCSI or NVMe over TCP. As you learned above, with director convergence, emulations are spun up and down when you need them. The same is true of SDNAS: it's not running until you configure it, but there is no work on Dell's part to make that happen. In our testing, NAS was performant and made use of the no-impact underlying data reduction systems in the system. I don't think most users will acquire PowerMax as their NAS array, but it provides a nice sidecar to an excellent block platform, and it scales well.
SDNAS provides a 64-bit file system, allowing for up to 64TB file systems, with 256TB file systems planned in the future. It also is fully integrated with SRDF and SnapVX, using the same underlying software to manage replication.
VMAX AFA was the first iteration to significantly shrink the size of the system, with the 250F/FX packing two two-engine arrays into a single rack; larger flash drives and data reduction technologies helped system density immensely. PowerMax 10 cuts the array size in half! The engines fit two directors into three rack units and only 2RU for the 48-drive DME. With the 8500, I can put six node pairs, four DMEs and the backend fabric switches in a single rack.
Since Unisphere 9, the arrays could send telemetry data to Dell's SaaS monitoring portal, CloudIQ. With Unisphere 10 and the workload intelligence in CloudIQ, it can flag unexpected access patterns and alert users, hopefully allowing them to mitigate any cyber-threats before they do too much damage. Combine that with Secure Snaps, the immutable copies with a time-to-live, the array's ability to take very granular snapshots, and you've limited the overall attack depth on your storage infrastructure.
PowerMax can integrate with a customer's existing RSA SecurID environment to make it harder for bad actors to access the system. Infrastructure attacks have driven our customers to adopt a more assertive security posture. Arrays have been connecting with authentication providers like LDAP and Active Directory for a while; granular RBAC has been added more recently. Multi-factor authentication is the next tool the storage OEMs are deploying to enhance system security. PowerMax supports SecurID, PowerScale supports Duo, and I'd expect to see this deployed across the rest of the portfolio before too long.
PowerMax 10 shows the power of its legacy. An already robust platform that has provided true active/active controllers, scale-out upgrades and a secure foundation adds features that keep it at the forefront of the enterprise storage world. Performing up to 2x better than PowerMax 9, it scales more than 4x higher in effective capacity in a reduced footprint. There is much to talk about and even more to love.