I’m very excited to announce that the evolution of HPE Nimble Storage is now available.

The new line is called Alletra 6000, and is completely interoperable with Nimble and manageable both from its own GUI/CLI/API and Data Services Cloud Console. For the Alletra 9000 (the evolution of Primera) there will be a separate post.

All the usual goodies of Nimble are still there (6-nines guaranteed uptime, fancy direct-to-L3 support, InfoSight with infrastructure AI recommendations, 100% headroom even if running on one controller, SCM cache, Triple+ RAID, Cascade Multistage Checksums etc). What’s different is mostly the vastly increased speeds in real-world workloads, and the shorter form factor.

Re-Engineered and Refactored

Instead of the usual incremental improvements most vendors go for, we decided to make far more significant architectural changes – which resulted in class-leading results.

Hardware

Remember my post about lack of PCI lanes in most modern storage systems? Well – we did something about that.

We decided to base our new platform on AMD processors and PCIe Gen4.

This has allowed a vast increase in PCI lane count and throughput all over the system – plus many more slots for HBAs, which is always a good thing. And a lot more CPU cores to play with.

We also made the new system physically shorter, it’s now about 31” deep and will fit in any standard rack.

The changes include a different and much faster way to mirror data between the controllers (which also results in much higher write speeds, since writes won’t be acknowledged to the host before they’re mirrored three times).

A welcome security addition is TPM (Trusted Platform Modules) – four total: two in the chassis and one in each of the controllers.

Software

From a code standpoint we had to refactor a large portion of the code to take advantage of the new hardware.

It takes a lot more effort than just recompiling to properly do this stuff.

There were major scheduling, queuing and memory management changes, path length optimizations, tons of new driver work, etc.

From an OS standpoint, some significant enhancements for the release accompanying this hardware are:

• Encryption for Peer Persistence setups (sync replication with auto failover)
• LDAP integration
• Full networking setup automation and auto-checks for dHCI and select switches
• A tenant-aware API, with Kubernetes support
• Non-disruptive upgrades from X10 (i.e. xxx000) to Gen5 (i.e. xxx00)

The Performance Enhancements

I will refrain from posting hero marketing numbers – any vendor can make crazy claims (like using 4K sequential reads to measure IOPS – sorry, no, that’s not a thing for any real workload).

Instead, I will share certified performance testing done for SAP HANA.

The way it works, SAP has a brutal performance test suite that hammers arrays in ways that make it hard to “cheat”. Various I/O sizes, random and sequential, overwrites, deletes, various phases etc etc. In addition, it’s not “easy” data like all zeroes or all ones that can be reduced to nothing (and go fast as a result). It really pushes the systems.

Depending on the results, a number of HANA nodes are certified to safely run on the array.

I like this approach since it’s tied to a real enterprise application, and the results directly correspond to how much workload one can run for that application instead of some useless number like 100% reads of a single, fixed block size read in a single way.

Before I share the results, it’s important to note that Alletra 6000 achieves these numbers using conditions that would make most other arrays show far smaller numbers. For instance:

• All Alletra 6000 performance is shown while the array has a full 100% headroom still available, which reduces customer risk even in extremely adverse situations. This means a controller can be unavailable (for instance, undergoing maintenance like a firmware upgrade) and the performance shown will be entirely unaffected. Most other dual controller systems show performance with zero headroom, so if a controller is unavailable you’re looking at half the speed…
• The data protection is Triple+ RAID and Cascade Multistage Checksums. This again lowers risk and affords an extreme level of data integrity hundreds of thousands of times greater than the protection used by other systems.
• The best quality of Triple + RAID isn’t that it protects against the loss of any 3 drives. The cool part is the “+”. It allows a system to have no data integrity issues even if every single drive in the system is experiencing sector read errors in parallel and three drives have been lost! That’s a level of resiliency against time-correlated errors that no other array in the world even comes remotely close to offering (see here for more info on why this is important).
• All this crazy reliability, headroom and performance density is possible in just 4U.

Without further ado, here are some numbers (all current as of May 18th, 2022 – updated a year after publication to reflect an increase from 96 nodes to 120 on the HPE A9080 after some code enhancements, and update the other vendors as well):

You can see the individual results here: Alletra 9000, Alletra 6000, Pure, PowerMax, NetApp, PowerStore. FYI, some vendors like IBM will show clustered results (saying “with max. supported IO Groups”- this means that’s a 4-HA pair cluster – 8 total controllers – so divide that number by 4 to get the speed of a single HA pair).

One can also run in scale-out mode 4x Alletra 6090 in parallel (in a single cluster/group entity) for a cool 216 HANA nodes, but I wanted to clearly show single array speeds in this example for an apples-to-apples comparison.

As you can see, the difference is stark – for instance, Dell’s PowerStore 9200T can do 28 nodes and that’s with 2 controllers maximally contributing to performance (zero headroom) and perhaps RAID5.

Remove a controller from each system and the per-controller speed of the PowerStore 9200T would be 14 nodes, whereas the Alletra 6090 would remain at 54 nodes – and the Alletra 6090 uses the incredibly more resilient Triple+ RAID.

That’s almost a 4x per-controller speed delta in favor of the Alletra 6090. Nobody knows what kind of speed one could expect from a PowerStore 9000T with Triple+ RAID and Cascade Multistage Checksums because that’s just not available  🙂

It’s safer to invest in something that’s both faster in adverse conditions and has orders of magnitude higher resiliency.

Summary

The Alletra 6000 represents a major architectural evolution of the Nimble design.

It has retained all the nice things that made Nimble a customer favorite in the first place, in a refreshed package offering significantly higher performance.

The more interesting part of course isn’t the “box”, as nice as the box is. It’s the ease of consumption, interesting data services, automation and orchestration that will be part of the whole experience, via the HPE Data Services Cloud Console.

D