This is a preparation blog post before I travel to Boston for Tech Field Day 11. You can find more information about Tech Field Day, the other presenters, and the other delegates here on the TFD web site.
Disclosure: TFD are paying my airfare and accommodation to attend TFD11 and I’m sure there will be gifts and catering from the presenters and TFD while we’re there. Everything I write about what I learn at TFD will be my opinion and will not be reviewed by TFD or the presenters. There is also no obligation or expectation that I will write about any or all of the presenters.
Plexistor presented at Storage Field Day 9 and the impression I got was that delegates were very confused as to what the product does. That is quite understandable as it’s not like other storage systems. The first thing to understand is that it is a kernel module for Linux. The next is that it delivers a file system directly to the Linux system. There are no LUNs or shares with Plexistor. The file system is backed by non-volatile(NV) RAM and flash (UltraDIMM, NVME or SSD) in a tiered model. Frequently accessed data in NVRAM and less frequently accessed on SSD. Multiple Linux systems can be clustered for high availability and for increased capacity, although both options reduce performance. Using NVRAM means that RAM is a tier, rather than a cache. Writes are persistent when they reach the NVRAM, so very fast.
So what can you do with this file system? You could put databases on the file system & get very high performance. Imagine how a MongoDB implementation would perform with its documents accessible at RAM speed. You could put KVM based VMs and get that same great performance. You could also use the POSIX mmap command to map an application’s RAM to a file on the filesystem. Since the file could be as large as the sum of NVDIM plus flash tiers this could deliver a lot of RAM. Think of an in-memory database like Redis or SAP HANA with 1TB of RAM.
Unfortunately Crunchbase does not have a lot of information on Plexistor. They were founded in December 2013 and have one round of funding for an undisclosed amount. The two investors are well known which suggests that Plexistor did not have to work too hard to get funding.
I am expecting that Plexistor will come to TFD11 with a much more coherent and clear story. The product does look interesting and upcoming products like Intel’s 3d X-Point solid state storage will certainly help. My concern is that tiering products rely on the workload above being inefficient. If applications like HANA or Mongo get better at using different speeds of storage then the market here is lost.
© 2016, Alastair. All rights reserved.