Startup MemVerge’s Big Memory software converges different types of memory into a single pool for applications.
Two much larger companies made recent software announcements concerning memory tiering and virtualisation: VMware’s Project Capitola and Samsung’s SMDK to accompany its CXL Expander. Will established players like these walk away with the market? We asked MemVerge CEO Charles Fan how he views these events.
Blocks & Files: Could you review MemVerge’s approach to expanding memory?
Charles Fan: MemVerge software today includes two important parts. The first part is the memory virtualization layer that manages heterogeneous memory types including DRAM, Persistent Memory (PMem) and in the near future CXL-connected memory. This layer includes intelligent tiering between these memory types and delivers the combination of their capacities as software-defined memory service to the applications without requiring application change. This part in effect “expands memory”.
There is also a second part of MemVerge technology that delivers an in-memory snapshot on top of the memory virtualization that can capture an application’s running state and allows the application to roll-back to this point-in-time. Such snapshots can also be moved in the background to another server, to a storage service, and the application can be restarted from this point-in-time, anytime from anywhere.
How does MemVerge’s technology virtualize (if that is the right term) the different memory types in a server, such as DRAM and NAND, for an application’s use?
As described above, MemVerge uses intelligent tiering between different memory types.
How does MemVerge view VMware’s Project Capitola? Does it render MemVerge’s technology redundant? How might MemVerge work with a Project Capitola version of ESXi and multiple memory tiers, such as DRAM + Optane PMem?
We are excited to hear VMware’s announcement of Project Capitola. (Please see our blog). The biggest bottleneck of our customer adoption over the last year was the lack of customer awareness of Big Memory software. Having VMware joining this area really validated this emerging market.
Yes, conceptually Project Capitola may be considered a competitor to the first of the two MemVerge technologies. However, MemVerge is different from Project Capitola in the following ways:
- Project Capitola is only for vSphere environments, while MemVerge Memory Machine can run in all environments, including in AWS, Azure or GPS cloud;
- Project Capitola can set different DRAM/PMem ratios per VM, while MemVerge Memory Machine can set different DRAM/PMem ratios per application process;
- Project Capitola will lose the persistence capabilities of PMem, while MemVerge Memory Machine does not. In fact, the MemVerge in-memory snapshot capability takes advantage of PMem’s persistence capability.
- Project Capitola will only be available later next year, while MemVerge Memory Machine is available now as Version 2.0 and has been on the market for over a year getting mileage from dozens of customers.
How does MemVerge view CXL in general and does CXL support have a role in MemVerge’s technology roadmap?
MemVerge is very excited about the advent of CXL. We believe CXL will be the new memory fabric that enables memory disaggregation and makes memory a first class-citizen in a datacentre for the first time — meaning memory can be scaled independently to the compute. This will increase the need for software to manage and scale memory.
We are currently working with a number of hardware companies which are working on CXL products, and it is in our roadmap to support these CXL products as they come to market.
Would MemVerge be able to support and integrate Samsung’s CXL Expander and its CXL SMDK? Does it agree memory tiering should not necessitate application software change?
It is in our plan to support all CXL memory cards, including Samsung’s. We are still learning about SMDK. In our current discussion with CXL hardware companies, we all agree memory tiering should not necessitate application software changes.
How does MemVerge see an industry standard emerging to cover memory tiering and virtualization? Is there a role for the SNIA here?
We believe it is an interesting area. Perhaps similar to how storage was disaggregated from compute in the early ’90s with SAN and the late ’90s with NAS, a similar phenomenon will occur with memory disaggregation with the emergence of industry standards or de facto standards. I believe there is a role for SNIA or other standards groups.