Introducing Intel Optane Persistent Memory – the new foundation for data-centre performance
Meet the chip tech that holds more than DRAM, is faster that flash and hard disks
Sponsored The data centre industry is undergoing a transformation: we're all generating and processing more and more information, from Internet of Things sensor readings to real-time analytics and machine-learning analysis. We need somewhere to store all these bytes – but where, and how?
One class of technology in particular from Intel® is well positioned to hold all of this data: Optane™, which is a form of persistent memory. Specifically, it is a type of non-volatile random-access memory (NVRAM) with a range of properties, all designed to boost performance. The concept of Optane™ is relatively simple: it holds more data than DRAM, it is faster than flash and magnetic hard disks, and it is non-volatile.
Look at it this way: imagine a pyramid, as pictured below, where DRAM sits at the top: low latency though expensive and volatile. At the bottom, relatively slow flash solid-state drives and hard disk drives. Sandwiched between the system memory and storage is persistent memory, a balance of speed and capacity.
The Optane™ family is a hybrid of storage and memory, offering the best of both worlds: it improves the handling of data workloads by keeping data closer to processors, speeding up processing time considerably without busting the bank. Intel® Optane™ DC Persistent Memory provides a new foundation for data centre performance, delivering up to 36TB of system-level memory capacity in an eight-socket system when combined with traditional DRAM.
Intel® Optane™ DC persistent memory is renowned for its exceptional high-processing speeds, and can reduce system restart times from minutes to seconds. There are even capacity choices for IT buyers: 128GB, 256GB, and 512GB per module.
Modes to success
One of the main advantages of Optane™ DC Persistent Memory is that it has the ability to operate in two distinct modes, which the provisioning manager can set depending on application needs. This flexibility is going to change the way data centres operate.
In the first of these modes, Memory Mode, the applications and operating system perceive the NVRAM as a huge pool of volatile memory, as if they were drawing on banks of normal DRAM. When running in this mode, no modifications are needed for deployed applications: they should run as is, accessing a much larger pool of system RAM, backed by NVRAM.
To do this, the real DRAM acts as a cache for the most frequently accessed data while paired banks of Optane™ DC Persistent Memory provide the large memory capacity. When data is requested from RAM by a program, the hardware controller first checks the DRAM cache and, if the data is present, the data is quickly returned, and the latency would be as if the memory address were read in DRAM. If the data is not in that DRAM cache, it’s read from the persistent memory, although this would mean a slightly longer latency. CIOs who are dealing with virtualized database deployments and big-data analytics applications will generally opt for Memory Mode.
The other mode is App Direct Mode, in which applications and the operating system can load and store data from and to NVRAM directly, choosing themselves whether to use persistent memory or normal DRAM DIMMs. This means that operations that don’t require a high latency and don’t need permanent data storage – such as database scratch pads – can be executed in DRAM, while structures that are very large can be routed to persistent memory.
Intel® envisages that the type of applications that will benefit from App Direct Mode include in-memory databases and analytics frameworks, while ultrafast storage applications would also be a good fit.
Keeping things secure
Another advantage of Optane™ DC Persistent Memory is that it’s been integrated with an AES-XTP encryption engine, to provide additional security in both Memory Mode and App Direct Mode.
Intel® is taking the persistent memory concept a stage further. In June of this year, the company announced a new storage architecture for supercomputers that combines Intel® Optane™ DC Persistent Memory and Distributed Asynchronous Object Storage (DAOS). DAOS is an open-source software-defined scale-out object store technology providing high-bandwidth, low-latency, and high I/O operations to serve the demands of AI and HPC workloads.
It’s this sort of approach that demonstrates where persistent memory is going to have the biggest impact. The need to process huge volumes of data quickly and efficiently is going to be at the heart of many applications in the coming years. Intel® Optane™ DC Persistent Memory chips, with their relatively low latency and fast processing, will be geared to handle the complex databases that will be a feature of many organisations in the coming years. Conventional chipsets will struggle to cope with the many gigabytes of data that will be produced by these applications. And with data centres operating close to capacity, the option to introduce more servers is often a non-starter – the only option is to improve performance.
The Intel® Optane™ Persistent Memory technology has immediately found a home at one large research organisation. CERN – home of the world’s largest and most powerful particle accelerator, the Large Hadron Collider – has been a keen user of Intel® products and an Intel® partner. As a preeminent nuclear physics research establishment, it’s an organisation that is producing massive amounts of data every day. As Alberto Pace, head of storage at CERN, points out, the organisation uses a huge amount of in-memory databases that were all reaching the end of their limitations. “We were waiting tens of minutes, if not hours for results," he said. "The use of Intel® Optane™ DC Persistent Memory has drastically reduced waiting time. The analysis of data using machine learning and AI really depends on this type of storage. It’s a technology with a lot of potential.”
CERN is perhaps an extreme example: it’s handling data sets well beyond the imagination of many organisations. These sorts of data volumes have been the preserve of just a select number of organisations in the past but they’re increasingly going to be the type of applications that will be handled by any number of companies. All of them will be looking at ways to give their data centres a boost.
The introduction of Intel®’s Optane™ DC Persistent Memory has created a new way of handling new paradigms that wouldn’t have been possible before. As CERN’s Pace puts it: “By bridging the gap between RAM and solid-state disks, this could be a breakthrough for computer technology in general.”
Sponsored by Intel®.