Although known primarily for their innovation in the CPU universe, Intel® has recently been making waves with a new type of memory known as Intel® Optane™ persistent memory (PMem), which has blurred the lines between memory and storage. Branded by Intel® as Optane™, and previously referred to as DCPMM, PMem is already impacting the enterprise and data centers everywhere, with Supermicro leading the market in building servers supporting the highest density of PMem per system. So, what does this really mean, how do the different modes of PMem work, and why did Intel invest $5 billion in Optane development?
PMem – What it is and is not
So let’s start by describing PMem. Like Flash, PMem is solid-state, and is inherently non-volatile, maintaining its memory even when powered off. PMem offers significantly lower latency than Flash – particularly since there is no need to erase before re-writing. This can translate to DRAM-like performance, especially when processing a high volume of small files, like OLTP. PMem was designed to augment memory and to cache frequently accessed data, thus reducing the number of disk I/O swaps, resulting in better performance.
Intel® Optane™ comes in two “flavors”: Intel® Optane™ PMem, used to expand the memory capacity of systems, and Intel® Optane™ SSD, designed to improve storage performance. In the hierarchy of IT media, think of Intel® Optane™ as fitting between DRAM and NAND SSD, with PMem configurations offering 100s of GB of memory expansion with sub-microsecond access, and SSD configurations offering terabytes of capacity with sub 10 microsecond access. There is just one caveat – servers will require Intel® 3rd Gen Xeon® Processors CPUs or later to take advantage of all the latest features that Optane™ PMem offers.
Data center use cases
For storage applications, Optane™ offers an SSD-like form factor that delivers up to 8x the performance of traditional NAND SSDs. In addition, Optane™ also offers a data center PMem configuration with 2x the density of DDR4 DRAM for those extremely memory hungry workloads that prioritize capacity over speed, such as SAP HANA and servers with high virtual machine (VM) density that are constantly vying for limited memory resources. Intel® Optane™ persistent memory modules are configurable into three different modes:
- Memory Mode turns DRAM into a non-addressable L4 cache, so that all user-addressable capacity is what’s in the PMem modules. In this mode, there is no special programming needed but data in PMem is volatile, just like the DRAM that is caching it. This delivers a large memory pool with low but effective DRAM cache investment.
- App Direct Mode lets the system address DRAM and PMem independently. Apps that demand the fastest memory can use DRAM while the rest can use Optane™. In this mode, the data in the PMem is persistent, which can greatly reduce memory reload time after a power failure or reboot. This, however, does require apps to be optimized for the App Direct mode.
- Dual Mode is a sub-set of App Direct Mode that allows some PMem resources to operate in Memory Mode and the remainder in App Direct.
Optane™ SSDs further blur the lines between SSD and DRAM with the ability to employ Intel® Memory Drive Technology – basically turning an Optane™ SSD into the equivalent PMem in Memory Mode. In this manner, Optane™ SSDs can greatly expand the memory pool to help contain costs on projects requiring memory capacity above other attributes.
Supermicro offers maximum-density Intel® Optane™ PMem support in a wide range of systems optimized for Intel® 3rd Gen Xeon® Processors. From standard rackmount servers such as the Ultra series to multi-node systems that reduce power consumption, such as the BigTwin®, Intel® Optane™ PMem helps enable larger models to be run, more VM’s to be executed, and faster database recovery. The Supermicro SuperBlade® which packs up to 40 CPUs, as 20 independent servers with extremely fast networking, in just 8U and is designed for HPC and Enterprise applications, can incorporate Intel® Optane™ PMem in each node for the ultimate in memory capacity for memory dependent applications and for recovery operations. Similarly, Supermicro’s GPU optimized systems with Intel® Optane™ give AI applications the additional memory to increase performance when working with massive amounts of data.
Real-world benefits
So, what does it all mean? Intel® Optane™ PMem simply excels at handling a large volume of data demanding low latency applications, such as:
- Terabyte-plus datasets where adding DRAM alone is economically unfeasible
- Memory-bound workloads that have already hit the limits of VMs and suffer performance penalties as a result
- Reducing the physical server count in the datacenter, since Optane™ memory can handle more VMs per server than DRAM alone
- Fast restarts of memory-heavy applications thanks to the persistence of Optane™ modules. This can reduce application restart time from hours to minutes
Which applications get the biggest bang for the Optane™ buck? Candidates include SAP HANA, Apache Spark, live content streaming, databases, high performance computing –virtually any application that has lots of data “in memory”. Furthermore, Optane™ SSDs operating in Memory Mode can further accelerate use cases including HPC Flex memory, KVM memory extension, Spark, or communications service provider (CSP) use cases.
As for non-volatile Optane™ SSDs, workloads that can see real benefits would include VMware vSAN, Microsoft SQL, and countless other high-volume applications.
If I/O bottlenecks are the problem, Intel® Optane™ SSD may be the best answer. Optane™ can break physical memory size barriers without breaking the IT budget. That is why organizations worldwide are increasingly turning to Supermicro to bring Optane™ into their data centers.
Supermicro consults with every organization and use cases to help determine which Intel® Optane™ configuration works best, tailoring each solution to the exact needs of every application.
Learn more at www.supermicro.com