Around the Storage Block

Accelerate transactional workloads with Storage Class Memory

What sort of performance improvement can you expect from an SCM-accelerate array? As part of the NimbleOS 5.2 blog series; discover how HPE Nimble Storage Class Memory can give you a speed advantage with transactional workloads. HPE’s most recent new technology announcement: hybrid arrays built using flash SSDs as the persistent storage and Storage Class Memory (SCM) as the cache.

shutterstock_332053760.jpgA decade ago, when HPE Nimble Storage launched, it made a name for itself by – among other things -using Adaptive Flash arrays to provide storage performance which was similar to All-Flash arrays, but at a fraction of the cost. The magic formula for that success was using just enough solid-state disk (SSD) as a read cache. The combination of SSD speed and HDD cost-efficiency, coupled with a host of enterprise-ready features and world-class support, led to massive adoption of the product by thousands of customers in record time.

This long history of excellence in hybrid arrays contributes substantially our most recent new technology announcement, hybrid arrays built using flash SSDs as the persistent storage and Storage Class Memory (SCM) as the cache.

The storage class memory used with Nimble arrays is an Optane SSD from Intel. Optane uses 3D XPoint memory, which is dramatically faster than the NAND flash used by conventional SSDs. As with flash-based SSDs ten years ago, SCM is too expensive for most of us to use directly as storage.  We could build an all-SCM array, but such a product would be unsuitable for all but the largest budgets.

That’s why the approach HPE Nimble Storage pioneered with their line of Adaptive Flash arrays just makes sense. You can take a relatively small amount of SCM, use it as a cache along with a persistence layer based on Flash SSDs and achieve the best of all worlds – blazing speed,  enterprise features, and reasonable cost.

For HPE Nimble Storage arrays, we’re now making 1.5TB of SCM available as an option for your AF60 and AF80 All-Flash arrays

As long as you are running NimbleOS version 5.2.1 or greater and have an available PCIe slot, you can add an SCM SSD to your AF60 or AF80 array. The kit comes with one for each of the two controllers in the array. You can also now configure them from the factory for newly purchased arrays.

What sort of performance improvement can you expect from an SCM-accelerate array? Primarily this will manifest as significantly decreased response time for random read operations. Since incoming writes are staged in non-volatile memory (NVDIMM) before being written to disk, there isn’t a big improvement to be had with write operations; they are already incredibly fast. Nor will there be a significant performance boost for sequential I/O, though there will be some nominal improvement. The big payoff will be for the random reads. As such, there are certain workloads which are perfect for SCM accelerated arrays and those workloads will see the most benefit. If you have a highly transactional workload, such as an order entry system or a real-time interactive application, perhaps that coordinates a production floor or an oil refinery, you’ll see significant reduction in transaction latency with SCM. While sub-millisecond storage latency has become expected from All-Flash arrays, SCM can push that latency as low as 150 microseconds. In general, we expect about a 2x reduction in latency.

Take a look at the following example of an order entry system, both with and without SCM.



As you can see, with SCM the storage latency is cut in half, meaning significantly less time that your database is waiting for data. This means your transactions are faster, your customers have a much better experience checking out from your online cart, and you can transact more business in a shorter amount of time.

In summary, using tried and proven caching methodologies hybrid SCM + All Flash arrays can significantly improve your application throughput, reduce storage latency, and provide a notable decrease in read response time for your transactional applications. You can have the speed of SCM at a moderate cost, all without sacrificing the enterprise data storage features you’ve come to rely upon.

Talk to your HPE Nimble Storage representative to help determine if it’s a good fit for your own applications.

Want to learn more? Read the press release, and please check out these blog articles:

HPE Discover Virtual Experience

As a reminder, registration for HPE Discover Virtual Experience 2020 is FREE – and with your registration, HPE will donate $10 to your choice of one of 10 featured causes, so join us and give back to organizations helping people, businesses and communities around the world recover from the COVID-19 crisis. Read more about these organizations and the great work they are doing to support COVID-19 recovery around the world. (Terms and conditions apply.)

Starting June 23rd, please log in to the HPE Discover Virtual Experience and search by Session ID (listed above) or by area of interest to view a session.

We look forward to seeing you virtually at HPE Discover 2020!

StephenDaniel_2120720 small.jpg




Meet HPE Distinguished Technologist and Blogger, Stephen Daniel. Stephen is a 20-year veteran of the storage industry, and has worked extensively in the fields of storage performance, data reduction, and high availability.


Storage Experts
Hewlett Packard Enterprise

0 Kudos
About the Author


Our team of Hewlett Packard Enterprise storage experts helps you to dive deep into relevant infrastructure topics.