Formulus Black launches software stack to vastly increase power of In-Memory compute

Formulus Black is a software rethinking of technoloigy originally developed by its predecessor company, Symbolic IO, designed to make DRAM non-volatile, inexpensive and usable for storage.

Formulus Black has launched its ForsaOS, a complete Linux-based software stack built specifically to run all applications in memory. It keeps data in Persistent Memory, safe against power loss, by making DRAM non-volatile and less expensive to use. It also amplifies memory by a variable but generally impressive amount, to make it highly efficient and cost effective.

Formulus Black is a resurrection – and reworking – of technology originally developed by Symbolic IO, a predecessor company of Formulus Black which imploded and rebranded a year ago. The then-CEO and founder, who is no longer connected with the company, had been arrested on a domestic violence charge. But the company’s whole route to market – although not the technology under the hood – also underwent a major restructuring.

“Symbolic IO was a hardware appliance, and was very capital intensive,” said Wayne Rickard, Formulus Black’s Chief Marketing and Strategy Officer. “So that was abandoned out of the gate, and they restructured and recapitalized the company so that it produced a much more streamlined software product. I joined the company five months into that, when the software was well on the path to being removed from the hardware dependence that was imposed by the Symbolic IO platform. We then had to add a licensing model, provisions for software activation, and other things to turn it into a software product. We also moved from a direct sale model to a channel model. The changes mean that now it is sold through the channel, and as a software product, can work off any modern Intel platform.”

Rickard said that the core technology remains the same as in the Symbolic IO days however – which explains the company’s large roster of patents inherited from the old company. They include 15 Issued US Patents, 7 US Utility Patents, 8 US Design Patents, 3 in Japan and single ones in Australia, Canada, Korea, Mexico, and Taiwan

“The bulk of what we do hasn’t changed,” Rickard said. “We still have the same in-memory architecture execution architecture story, where everything lives in the memory channel and main memory.”  Their Forsa Operating System is a Linux-based software stack that runs all applications in memory, to generate huge gains in compute efficiency.

“In the old days, DRAM was fast and not very big, while hard drives were super-slow and super-cheap,” Rickard said. “That meant things that had to be executable now went into in DRAM and everything else on drives. Back then, a CPU had 1 core and 1 path to memory. Today they can have 36 cores and multiple memory channels, and you can put 3 TB of DRAM on a 6-socket server. But it is still expensive and volatile. Our innovation is we that have made DRAM less expensive and non-volatile.

Formulus Black does this with patented algorithms that process application codes to find and remove repetitive bit patterns. They are replaced by Formulus Bit Markers (FbMs), which results in lossless, mathematically optimal storage in-memory, without the use of compression, decompression or dedupe.  Their ForsaOS proprietary OS stack includes a built-in hypervisor that works with the FbM technology to present the amplified memory to each guest OS as virtual storage, accessable at memory channel speeds. The result is a much smaller data memory footprint, with greater usable capacity than the installed physical memory.

“The more that a pattern is replicated, the more we can use that one Bit Marker to point to it,” Rickard said. “The effect of that is typically 3x to 4x amplification, although I have seen it go as high as 24, with the worst case being 1.2x. This solves the size problem and solves the cost problem. That was the first innovation.”

Rickard said that the second problem Formulus Black solves is persistence.

“Because DRAM has been volatile, it has never been appropriate for storage,” he stated. “We keep it safe against power loss, because if we detect a power failure, we will take system memory from DRAM and migrate it to external SSDs. This couldn’t be done before, because hard drives were too slow for this migration. You would run out of power before the whole thing was copied. Now with our amplification, we can do it well within the envelope, treat the whole system like a big NVDIMM, not give up the capacity, and do it at a much cheaper cost point.”

The third problem address was that system architectures still don’t recognize the memory channel as having a place for storage, and just view the IO bus for storage.

“We solved this issue by modifying KVM and the Ubuntu kernel so the hypervisor can see system memory as either memory or storage,” Rickard said. “So we can create a  virtual disk out of memory. The application thinks it is talking to storage, when it’s actually amplified memory. This also eliminates another system artifact, that memory is only used for cache. Since all storage is now in memory, we’ve eliminated the need to provision any cache at all.”

Rickard also stressed that this technology is fundamentally different from memory-driven computing.

“They are talking about composable infrastructure, which independently scales infrastructure, storage and compute to optimize,” he said. “It requires a different management plane to make it work and all that stuff has to be invented. Our model requires no changes to either the application or the deployment model or the way we fit into the applications. There is no heavy lifting and no code rewrite.” It plays nicely with existing hardware, loading as an image on top of a standard modern Intel platform – Broadwell Haswell or Skylake – or AMD Epyc.

“Once you load an image, we present a standard virtual machine hypervisor to the world,” Rickard said. “Anything that runs on a virtual machine will run on our platform, and we provide a GUI similar to VMware ESX or Nutanix.”

The Go-to-Market strategy is to license the software stack to enterprises, OEMs and cloud providers.

“We are targeting strategic partnerships primarily in the backup space, notably Veeam and Rubrik, although we don’t have formal relationships with them,” Rickard said. “There are other logical partnerships with other ISVs that make sense going forward, around things that create a lot of memory, and with ISV platforms that address very specific verticals. This is like fairy dust to these vertical applications. We can make Greenplum databases look a lot better.

Formulus Black has six channel partners at this point, and are looking for more, with a strong vertical focus.

“We are the underpinning, but we would be almost invisible to their customers,” Rickard indicated. “We have a channel program well underway. We will make our partners’ offerings look better against their vertical competitors. These partners with vertical focus give us that front line expertise. It is so much more efficient than a direct model where every application is new.”

Pricing is per socket.

“We will work with a single socket or dual socket license, fully loaded with memory or not, and with no extra cost for applications,” Rickard said. “The model is about as straightforward as it can be.”