Software-Defined Memory Arrived

This article was published on April 7, 2022, written by Ken Clipperton, lead analyst, DCIG, LLC.

Software-Defined Memory in the Wild

The era of large, software-defined memory is upon us. In fact, it is “in the wild” now and is already delivering dramatic performance and cost benefits to many organizations.

Software-Defined Memory Arrived
The virtualization of compute, storage, and network resources over the last 20 years has enabled organizations to reap many benefits. Those benefits include improvements in data center Capex and Opex efficiency with enhanced resiliency. Now software-defined memory, in the form of the MemVerge Memory Machine, brings the benefits of virtualization to computer memory, including DRAM and persistent memory.

Early adopters are achieving success across multiple demanding use cases
MemVerge reached availability in the fall of 2020. Early its use cases include life sciences research, video animation, special effects production (VFX), and-surprisingly-banking. By virtualizing multiple memory tiers into a single addressable memory pool, the company makes it possible for much larger working sets to reside in memory and achieve much better economics than are possible with DRAM-only solutions.

Wins On-premises
We don’t have to get to exotic workloads to discover the benefits of software-defined memory. VMs are a standard element in both cloud and on-premises data centers. DRAM is often the constraining resource limiting the number of VMs a physical server can host. MemVerge technology breaks through the DRAM constraint.

Case studies and articles on the MemVerge website document multiple wins for on-premises workloads in multiple industries and across a range of workloads. For example, MemVerge’s benchmark testing revealed that VM density could be increased by as much as 8x without significant performance degradation in a standard VMware or KVM environment running MySQL VMs.

Increasing VM density per server or increasing application density per VM results in savings on hardware, software licenses, power, space, and cooling. Increased performance density can reduce the TCO, especially where data center power, space, or cooling resources are constrained.[i]

Beyond reducing TCO, software-defined memory enables workloads and application performance that would otherwise be impossible or too costly to be practical. In turn, this creates opportunities to reimagine and redesign business processes and create business value on-premises.

Genomics
Analytical Biosciences found that genomics data loaded 800x faster and that overall task completion time for the workload was cut from approximately 6.5 hours to under 3 hours. The solution also dramatically improved infrastructure efficiency, with an overall 10x increase in project throughput per server. Another MemVerge client cut their time-to-discovery by 37%.

M&E
Visual effects (VFX) and animation applications are processor and memory intensive and are becoming more so as resolutions increase and as more and more of what ends up in the final product is computer-generated rather than live-action content. As artists push the boundaries, application crashes can occur.

The in-memory data management services provided by MemVerge can transform an hour of crash recovery downtime into just second. Artist time is expensive, and production timelines are unforgiving. The firm enables artists to think differently about working on scenes and elevate their art within the constraints of budgets and deadlines.

Financial Services
The truism that time is money is perhaps better understood in the financial services industry than any other. Thus, it is not surprising that the Members Exchange (MEMX) stock exchange, the fastest-growing US equities stock exchange[ii], was an early MemVerge customer.

In a March 2021 Bloomberg interview, MEMX CEO Jonathan Kellner was asked: “What’s your edge?” He answered: “One area where we are really differentiated is the fact that we do our whole exchange on one server rack. We are able to put a lot of computing, storage, and networking in a small amount of space. A much smaller footprint is much more cost-effective for us. Second, you have fewer hops across the exchange, so you have fewer points of failure and a more stable exchange. And you also have a much shorter distance from when the customer gets to our exchange to when they get to the matched order.“[iii]

Another financial services firm that embraced MemVerge for a very different application is Banca Intesa Sanpaolo, a banking group in Italy and one of the top banking groups in Europe with 19 million customers in approximately 40 countries worldwide. It uses MemVerge to support its online banking and real-time recommendation engines to promote offers to customers under specific conditions. The ability to run large databases in memory is critical to such real-time applications.

“With all relevant information in memory in an in-memory data grid, there is no need to traverse a network to remote storage for transaction processing. The difference in speed is significant – minutes vs. sub-millisecond response times for complex transactions done millions of times per second”.[iv]

“MemVerge provides a very fast backup of the landscape of the memory with a fast snapshot and then the ability to recover in seconds,” said Nicola Carotti, Banca Intesa Sanpaolo, head of cloud and hybrid datacenter. “The first benefit we had is to backup all the data every 30s with no impact on production and to guarantee a one-second recovery.“[v]

Wins in Cloud
Genomics
The Memory Machine Fault Tolerant Service (FTS) makes it feasible to run stateful, long-running, genomic analytic pipelines in the cloud using lower-cost spot instances rather than on-demand instances. This reduced the cost of running a SplAdder analytic pipeline from $1,559 to $263, a savings of 83%.

Cloud/colocation data centers
phoenixNAP is a global IT services provider offering security-focused cloud infrastructure, dedicated servers, colocation, and specialized IaaS technology solutions. The company is working with MemVerge to enable simplified deployments of Memory Machine on its automation-driven Bare Metal Cloud and provide an infrastructure solution optimized for big memory workloads.[vi]

Up to this point, many large stateful applications have not run cost-effectively in the cloud. Now MemVerge is enabling these apps to move to the cloud by encapsulating stateful apps into AppCapsules that include the memory state of the application. Coupled with MemVerge orchestration services, this enables cloud fault tolerance, cloud bursting, and cloud-to-cloud mobility even for stateful, non-fault-tolerant, long-running apps using low-cost spot instances.

These innovations align with the growing enterprise adoption of containerized applications and the cloud.

Emergence of Memory-as-a-Service (MaaS)
Perhaps due to enterprise familiarity with virtualization and software-defined data center concepts, memory virtualization is progressing rapidly up the path to becoming another “as-a-service” component of cloud environments, whether private or public.

We are seeing the emergence of MaaS, where the amount of memory for a workload is no longer depends on the server instance type. Instead, MaaS allocates memory independent of CPU. Much as storage and networking are today. This MaaS capability, in concert with PCIe 5 and CXL technologies, will enable a new level of infrastructure composability and efficiency.

Features Underpinning Memory-as-a-Service with MemVerge

Transparent Memory Service virtualizes and automatically tiers DRAM and non-volatile memories into a large pool of software-defined memory with QoS.

Container support
Containers are a fundamental building block for cloud-native applications. MemVerge supports multiple Kubernetes-based container platforms, including OpenShift and VMware Tanzu.

In-memory data management services
It solves what have been insurmountable issues of storage physics through memory snapshots. These ZeroIO Snapshots are foundational to its in-memory data services, which it calls Time Travel, AutoSave, Thin Clone, and App Migration.

Orchestration services
The solution captures an in-memory snapshot of all application data needed to restart an app into what MemVerge calls an AppCapsule. Cloud mobility services provided by the company integrate with cloud service schedulers and cloud-native application orchestration platforms to enable cloud bursting and cloud-to-cloud workload migration. 

Benefits of MaaS
MemVerge technologies offer benefits to enterprises and cloud service providers.
• Makes the cloud both possible and cost-effective for a broader range of enterprise applications, including long-running stateful applications.
• Enables long-running apps to take advantage of spot instance pricing.
• ZeroIO memory snapshots protect and rapidly recover big memory applications.
• AppCapsules simplify and accelerate application migrations in hybrid cloud and multi-cloud environments.
• Transparent Memory Service makes large pools of software-defined memory available to applications without requiring changes to the applications. This enables data sets that were too large for memory to gain the performance benefits of in-memory processing-more transactions in less time, at a lower cost.
• Enables more efficient use of all performance resources in the data center by consolidating more work onto fewer servers.
• Accelerates backup and recovery, providing data protection for big high-value in-memory data.

Guidance for Enterprise Architects and Business Decision-Makers
Software-defined memory and MaaS may initially sound like an exotic only-for-the-Fortune-100 technology, but this is a misperception. These technologies can accelerate mainstream applications and streamline infrastructures cost-effectively. From the animation workstation to the containerized application, software-defined memory can increase performance 10x and even 100x without increasing the initial cost.

Any organization running compute or memory-intensive workloads should investigate these solutions. Organizations seeking to eliminate data centers through consolidation or migration to the cloud would do well to evaluate MemVerge, as would any organization approaching full utilization of existing data centers.

As noted in the opening paragraph of this report, the era of large, software-defined memory is upon us. In fact, it is now “in the wild” and is already delivering dramatic performance and cost benefits to many organizations.

Beyond software-defined memory, MaaS is on the near horizon. Enterprise architects and other business decision-makers should take it into account as they formulate technology plans that reach beyond the 2022/2023 watershed for many of the enabling hardware and software technologies.

[i] https://memverge.com/software-defined-memory-platform-for-virtual-machines/, accessed February 22, 2022.
[ii] https://www.businesswire.com/news/home/20220113005181/en/Fastest-growing-US-Equities-Stock-Exchange-MEMX-Uses-Azul-to-Power-Its-High-Performance-Java-based-Trading-Platform, accessed February 22, 2022.
[iii] https://www.bloomberg.com/news/articles/2021-03-30/jonathan-kellner-has-big-money-backing-his-memx-stock-trading-start-up, accessed February 22, 2022.
[iv] https://memverge.com/wp-content/uploads/2021/03/Case-Study_Banca-Intesa.pdf, accessed February 22, 2022.
[v] https://www.youtube.com/watch?v=abv3SA4CPmk, accessed February 22, 2022.
[vi] https://phoenixnap.com/company/press/memverge-memory-virtualization-in-bare-metal-cloud, accessed February 22, 2022.