Aerospike and Intel: Continuous Focus on Best-in-Class Cost-Performance along with Ensuring Zero-Downtime Migration to Newer Hardware Technologies
Since 2019, Intel® Optane™ PMem has been a new class of memory and storage technology, that is a little slower than RAM but has the virtues of persistence, high IOPS, and an attractive value proposition for some of Aerospike’s customer use cases. That’s why PMem was incorporated in Aerospike’s mix of offers to store our Primary Index and data in different combinations:
- Both stored in DRAM (In-Memory)
- Both stored in Persistent Memory (PMem)
- Indexes in DRAM, Data in Flash Drive/SSD (Hybrid Memory)
- Indexes in PMEM, Data in Flash Drive/SSD (Hybrid Memory)
Obviously, there is (Performance + Cost) vs. Scale trade-off that dictates the ideal index/data combination for a specific use case. Commercial deployment by a major telco is a testament that PMem can be trusted to handle Telcos’ critical workloads supporting key use cases such as “Customer 360.” Our blog post on the Intel Network Builders platform during Mobile World Congress 2022 highlighted this. However, continuous improvements on SSD performance and the rapidly falling price of DRAM shed light on the effectiveness of the 3rd choice above - our patent-winning Hybrid Memory Architecture (HMA).
Last summer Intel announced the winding down of its Optane business in favor of more promising technologies.
A key alternative to Optane PMem is CXL (Compute Express Link), in which Intel is already heavily invested. Intel is a founding member of the CXL consortium which was formed in 2019 to develop an open standard for next-generation memory capabilities. CXL1.0 specification came out in March 2019 and Intel donated the code that formed the basis of that spec.
In this Register article, Jim Pappas, CXL consortium chairman and Intel’s director of technology initiatives, articulated that CXL defines a common, cache-coherent interface for connecting CPUs, memory, accelerators, and other peripherals. He also made the observation that CXL would benefit AI/ML workloads by enabling a much more intimate relationship between the CPU, AI accelerator, and/or GPU than is currently possible over PCIe.
Intel’s view is that the industry is embracing memory expansion, tiering, and pooling to optimize cost, capacity, and performance. This originally influenced Intel’s development and productization of PMem. Along those lines, CXL usage models and software solutions are comparable to the usage models and software that Intel Optane PMem supports. Therefore, Intel insists That even if enterprises use Intel Optane PMem today to cost-efficiently meet memory expansion and tiering needs, they can then migrate to CXL-based solutions as the CXL 2.0 ecosystem matures over the next two to three years.
Now, the question becomes – what does Aerospike do to smoothly handle this transition?
Following a recent Aerospike and Intel cooperative effort to help Aerospike customers through their respective Optane PMem transitions, Aerospike Founder and CTO, Srini Srinivasan commented:
“At Aerospike, we help our customers meet and exceed stringent real time SLAs for Operational Database Applications while delivering predictable performance and reliability at unlimited scale. We have a number of Aerospike customers who have large scale production deployments utilizing Intel Optane PMem technology today. We have engaged with Intel on an approach for our customers to ensure they continue to meet their SLAs with Aerospike at unlimited scale and guide them through the planning and migration off Optane with zero downtime.”
In this regard, it is important to remember that Intel has committed to utilize existing PMem inventory to fulfill orders through 2025. Additionally, Intel has pledged to support Optane for 5 more years after 2025 end of sales. As Optane successors (case in point CXL) take hold, Aerospike and Intel will continue to work together and ensure that our joint customers will have access to software that is fully tested and integrated on upcoming Intel technology thus continuing to benefit from a compelling value proposition.
Srini further stated:
“For the majority of Aerospike customers running Optane today, the above plan will be sufficient. However, for those Aerospike customers who wish to move off Optane prior to 2025, or those that need to scale out their existing Optane-infused clusters and do not wish to further their investment into the Optane technology, I’m happy to announce an Aerospike certified approach to migrating and scaling your Aerospike production clusters off Optane PMem without downtime, holding firm on your SLAs and with improved TCO. We have architected and documented our server compression approach leveraging Aerospike’s HMA deployment model and the NUMA pinning feature set to provide our customers additional TCO- related savings while maintaining predictable SLAs at unlimited scale without specialized hardware components.”
As mentioned earlier, HMA deployments get an additional boost from consistent recent decline in DRAM prices and it makes HMA the clear leader (over PMem) from cost-performance perspective for large scale Aerospike deployments.
Meet us at MWC 2023 Barcelona
Aerospike’s representatives from its enterprise solutions, telecom vertical, and partner/alliance team will be attending Mobile World Congress (MWC) 2023 in Barcelona and take part in an interview session to highlight the message above. Post MWC 2023, we also intend to present in an Intel Network Builders webinar to provide further details and our latest status on this topic. Learn more about the Aerospike Real-time Data Platform at aerospike.com.