Big-Socket Systems With Intel® Optane™ Technology Drive HPC Use Cases
In This Article
It used to be that four-plus socket server systems were commonplace hardware for data centers. But with the swing to distributed, non-specialized systems some 30 years ago and the emergence of virtualization with more VM per-server density for application availability, the focus shifted to one- and two-socket systems -- and data center requirements began to change with the times.
Today, with the massive growth in business demand for increased functionality and performance of memory, storage, networking and compute, the traditional single- and dual-socket systems have reached their growth limit. They cannot expand with the times, and the complexity and requirements around clustering single- and dual-socket servers can result in an inefficient solution. Whereas big-socket solutions provide a better ROI than single- or dual-socket solutions, costing less and requiring less power, cooling, rack space and IT upkeep.
The reality is, IT is evolving from a compute-centric environment to one that's data-centric and increasingly real-time, in which resources must be close to the data. In other words, we're transitioning from an age of big data to the age of big memory with data in motion. And yet, you cannot solve the challenge of data in motion with a hyper-converged approach; in fact, most OEM-engineered solutions require additional software licenses and professional services engagements to place them in the customer's data center. What's more, those customers are at the mercy of their vendor when it comes time for upgrades.
Instead, the best way to answer that demand for big memory is to resurrect a classic model with a contemporary twist: the big-socket server system, supercharged with the revolutionary memory and storage performance of Intel® Optane™ technology and other Intel® product innovations for the data center. For anyone who might question this approach, consider these reasons and use cases illustrating the value-add that big-socket brings to high-performance computing (HPC).
1. Address data in flight more effectively.
Data volumes are expanding by the second and IT organizations need to perform real-time analytical processing (RTAP) on that data to glean actionable insight and support digital transformation. Whereas a few years ago there was a push for building "big data," today we are seeing a move to "big memory" -- in-memory databases that can deliver the "data now" that customers are demanding. But that demand cannot be met by single- and dual-socket servers, which simply lack the horsepower to handle all that data in motion. Instead, it requires that RTAP be done in memory because of the latency, and that requires the power and capacity of big-socket systems.
2. Protect data with less threat exposure.
It's simply a matter of numbers: since big-socket systems involve fewer servers to monitor and fewer points of attack from cybersecurity threats, they reduce the potential for costly, disruptive breaches. That's a huge benefit to IT and business. In addition, going from, say, 100 servers down to 20 reduces complexity for data center management. Now apply this change at the rack- or data center-scale and you begin to appreciate the difference.
3. Reduce the total cost of ownership (TCO).
WWT endorses this sensible belief: if you can keep it simple, keep it simple. That bit of wisdom certainly applies to big-socket systems, which reduce attack surfaces and vectors, and are easier and less time-consuming to manage, conserving IT and resources. Instead of worrying over how to fit single- and dual-socket systems into limited rack space, increase capabilities by moving from a vendor-centric approach to a cost-optimized big-socket solution that ultimately delivers a long-term TCO dividend.
4. Increase reliability.
A long-touted advantage of single- and dual-socket systems is the old "don't place all your eggs in one basket" problem: if one goes down, they all go down. Advancements in recent years have significantly boosted reliability of big-socket systems -- they're designed for rapid failure recovery, helping deliver the greater uptime, performance and security from the Intel® products that power them. Which leads to our final reason...
5. Maximize HPC advantages with Intel® technology.
In addition to the above four inherent advantages, big-socket systems benefit further from the proven reliability of their Intel® technology components for memory, compute, networking and more. Intel technologies work securely together to drive virtually all WWT data center solutions, including these products for HPC:
- Intel® Optane™ persistent memory is the key to big-socket system performance, sitting between memory and storage to bring large memory closer to the CPU, delivering huge capacity to accommodate demanding workflows. Its versatility enables Intel Optane persistent memory to be used like conventional DRAM or as persistent storage. It also accommodates a Key Management Server (KMS) solution when required for persistent storage. On a four-socket system, Intel Optane persistent memory can provide up to 36 TB of system-level memory capacity when combined with standard DRAM.
- Intel® Optane™ SSDs are a new class of non-volatile memory with the high speed and density to eliminate processing bottlenecks and improve performance in big-socket systems. Deployed with Intel Optane persistent memory, these SSDs close the gaps of capacity, cost, and performance between DRAM and 3D NAND SSDs, advancing data-intensive workflows while increasing uptime and flexibility for HPC.
- 2nd Gen Intel® Xeon® Scalable processors deliver outstanding big-socket compute performance to drive flexible solutions across a broad range of mission-critical, data-centric workflows. And they include acceleration for HPC and artificial intelligence (AI).
- Intel® networking technologies, featuring 100Gbe Ethernet adapters, scale to meet network demands in data-centric environments, accelerating high-priority applications, packet processing and latency-sensitive workflows.
- Intel® FPGAs (field-programmable gate arrays) provide acceleration that can be quickly deployed and re-configured after manufacture to suit a given environment or workflow. These SmartNIC devices are part of the ecosystem paradigm shift in computing to move aspects of computation off the motherboard and onto composable resources such as networking, memory and storage. This supports a dynamic, real-time, load-balancing "just-in-time" compilation of complex workflows.
WWT integrates these Intel technologies into today's big-socket servers for cost-effective HPC with unparalleled scalability and low latency. Together, we are introducing HPC solutions across a number of industry use cases leveraging cloud, edge and on-premises environments.
- Risk management capabilities for transaction processing, investment monitoring, hedge funds and other financial services operations that require near-instant processing, especially when your decision window is measured in microseconds. The forever war against fraud and theft keeps defenses on continuous high alert, with no tolerance for latency vulnerabilities to exploit. Data must be readily available to evaluate multiple risk factors, which is why risk management is the number-one use case for the speed and security of big-socket systems.
- Engineering sciences run incredibly complex calculations -- everything from combustion simulations of rocket engines to modeling geology for oil exploration. They can only scale so much before encountering a memory bandwidth problem and then reach network saturation. But a big-socket system is able to clock at higher speeds not limited by the network. What's more, when Intel Optane persistent memory is added, the entire model can be placed in-memory so that everything runs faster with increased memory bandwidth.
- In-memory systems such as SAP HANA and ERPs -- as well as REDIS, MemSQL, MinIO and others that deal with rapidly exploding databases -- must put that data in-memory. Consequently, single- and dual-socket systems fall woefully short of the challenge; they simply need more sockets to support memory. Once again, a big-socket solution is what's required.
- Advanced artificial intelligence is another example of the data-in-motion challenge -- how to deal with the speeding freight train of raw data bearing down on the world. One game-changing innovation for AI application developers is Intel® oneAPI: for the first time, they and other engineers will have a single, unified and open programming model to build and optimize code across diverse architectures, including processors and accelerators. Intel oneAPI's "no developer left behind" approach is simplifying development across all architectures and is agnostic whether it runs on single socket or big-socket.
- Digital experiences encompass multi-user, smart spaces, 3D remote visualizations, remote video sessions, interactive collaboration with interdisciplinary teams, or any endeavor that needs to process data quickly and drive models interactively -- it's all driving the need for larger systems to manage those interactions. Big-socket servers enable the move from workload to workflows, where individuals doing tasks give way to interdisciplinary teams engaging collaboratively to complete the work.
Big-socket servers might be "old school" solutions from yesterday's data center, but in fact they are today more relevant than ever, in on-prem and public cloud infrastructures alike. As workflow challenges grow more complex, the need for big-socket systems based on Intel technology is growing exponentially.
This is WWT's opportunity to focus on its "first principles": correctly identify the current problem and then solve that problem -- and in this case, the problem concerns data in flight, data in motion, data now. As more organizations rethink their data center infrastructure requirements, look beyond single- and dual-socket replacements and take a second look at the benefits of big-socket server solutions.
WWT can help: explore the HPC workshops and briefings in our Advanced Technology Center (ATC) or request a Server Infrastructure briefing and workshop. And check back in Q1 2021 for more information on our High-Performance Architecture Lab.