Our first session note from Storage Visions 2014 comes from the session, “Please Don’t Go Away: Bringing Non-Volatile Memory to Tomorrow’s Storage Architectures”, sponsored by SNIA. As the title implies, the session discussed Non-Volatile Memory (NVM) and its potential impact on storage architectures. NVM is essentially memory that can act as RAM or storage that doesn’t lose the data it is storing when power is removed.
The Storage Networking Industry Association (SNIA) has created a technical working group (TWG) on NVM. The purpose of the TWG (pronounced “twig”) is to deliver specifications describing the behavior of a common set of software interfaces that provide access to non-volatile memory (NVM). The TWG goal is to encourage a common ecosystem for NVM-enabled software without limiting the ability to innovate.
In essence, the TWG is working on how we can communicate with something that may be 1000 times faster than flash is today? Today, if you break down where the latency is in each I/O to flash storage, the overwhelming majority of that latency is in the program/erase cycle of the flash media itself. When NVM becomes available, the majority of the latency will be caused by the storage I/O interface and the application itself. As a result, the work that SNIA is doing on software interfaces is critical to realizing the full potential of flash storage.
The Bridge To NVM
In reality, NVM won’t be ready for at least five years and probably won’t be viable in mass use for 10 years. Think of it as similar to the progression of flash-based storage. First flash was only justifiable to the performance fringe, now its practical for almost every data center.
There are bridging technologies that will service us well as we move closer towards the real application of NVM technology. As we discussed in a recent webinar, SanDisk and Diablo Technologies both have Memory Channel Architectures that allow Flash Storage to access the performance and bandwidth of the memory bus. Viking Technologies has NVDIMM, which as we discussed in our article, “How NVDIMM Can Protect Flash SSD Appliances“, integrates Flash with DRAM and a capacitor to provide a non-volatile aspect to RAM memory. Both of these will require server manufacturers to modify their BIOS to be useable, but these modifications should be relatively simple. Today we will have to rely on PCIe or SAS based SSDs.
Storage Swiss Take
One of the benefits of attending sessions like these it that it makes you think. For example, how will we use storage that is 1000x faster than what we consider the fastest storage available today? With NVM we are basically talking about storage that operates at memory speeds! There is the obvious low hanging fruit, clearly there are environments that need more IOPS than we can deliver with flash technology today, like high frequency trading and real-time analytics. Beyond that though, NVM will spawn a whole new generation of applications that we haven’t even dreamt of yet.
The big question for me is if I have storage that is as fast as memory why do I need memory. Or why do I need storage? It seems that one of the two become unnecessary in the future data center. After all storage exists because memory is non-volatile and is too expensive to have in large scale; NVM may not have those problem. It is possible that NVM may, at a minimum, end up merging RAM with Tier 0 and Tier 1 storage.
Can you imagine a world with just RAM and Tape? I can, we used to call that a mainframe.