Memory

Persistent Memory Virtualized with vSphere 6.7

By Brett Williams - 8.23.18

You need to return a purchased item to the local store. Wrong size, wrong color, it just isn’t right for you.

You’re dreading the return line, but when you get to the store there’s no line. That seems strange. You walk up to the counter, where the clerk scans the item then calls you by your full name, states that you’ve been reimbursed fully, thanks you, and sends you on your way. You are done!

What just happened? What you just experienced is an in-memory database being able to determine what you bought, when you bought it, identifying you and refunding you, all in seconds of compute time.

So, what is persistent memory? It is the fastest DRAM but with a memory that doesn’t go away even if the power does. Micron NVDIMM runs at 2933Mbps performance, but the data is persistent even without power.

There is a convergence in the market between storage and memory. This is due to the demand for faster access to information, or "lower latency" of data access. The drive for lower latency and faster throughput is exacerbated by the artificial intelligence/machine learning/deep learning "wildfire."

Memory databases are the trending upward. Memory is measured in nanoseconds, NVMe drives in 10's of milliseconds, and SSD drives in 100's or microseconds. Getting more data into persistent memory provides 1000x improvement in latency, which is great for analytics. Data turns into intelligent information more quickly.

Vsphere 1

75 percent of servers in the market are now virtualized. Until now none of those virtualized servers could take advantage of NVDIMM persistent memory. Getting lots of data to and from memory required slower SSD transactions with bulk reads/writes – it took longer to find the valuable bits of data.

VMware® now supports NVDIMM for vSphere 6.7®, either in block mode, supporting older applications, or in direct byte mode for newer performance-driven applications. This results in a 1.8x improvement in throughput, meaning you are getting more work done with NVDIMM compared to SSD. This is a 2.3x improvement in latency, so your applications won’t be waiting around for reads/writes to finish.

Vsphere 2

Source: Computer Profile, Sept. 2017

The applications affected by this trend appear across many markets. Even more of the market can accelerate workloads using NVDIMM. Applications like relational databases, scale-out storage, virtual desktops, big data, in-memory data bases, high performance computing, and more. With NVDIMM now enabled, transactions can happen up to 400x faster, resulting in lower latency and faster data access rates.

Micron will be discussing this more at VMworld 2018, so keep an eye on @MicronTech and @MicronStorage on Twitter, and micron.com/virtualization for more information!

Brett Williams

Brett Williams

Brett Williams is Senior Business Development Manager at Micron. Previous positions include Applications Engineer, Applications Engineering Manager General Manager of the Platform Architecture Group and Senior Marketing Manager. Williams also launched Crucial Technology (Crucial.com), a subsidiary of Micron.

Prior to Micron, Williams worked at Intel from 1981 to 1989. He earned a BSEE degree from Brigham Young University in 1981.

 


+