Texas Memory Details Its Secret Sauce
Today’s flash storage systems have a lot of work to do. There is the lowest level flash management that includes error correction and wear leveling. Then there is the more advanced functions that manage the infrastructure that surrounds the flash NAND. Essentially flash storage systems need an operating system to make sure that all of these parts work as expected and work with each other to make sure maximum data reliability it achieved.
Today, Texas Memory provided details on the operating system that has driven its flash based system for over five years, starting with the RamSan-500. Unlike a typical operating system this OS is designed from the ground up to manage flash and to work on a cluster of CPU nodes and Field Programmable Arrays (FPGAs) that are distributed throughout a RamSan Flash appliance.
The OS provides flash management, data reliability management as well as manage a software stack that exists in CPUs and FPGAs. The data reliability management ranges from wear leveling and garbage collection to managing Texas Memory’s Variable Stripe RAID that we discussed in a prior article. In the newer Highly Available Systems that Texas Memory has brought to market, the OS manages the fail over process, also.
The use of FPGAs is a key ingredient for all the RamSan systems and a key differentiator. Many vendors have chosen to leverage a software only model where the operating system runs on a CPU inside the appliance or uses the CPU of an attached host. This approach, while simpler to design can cause some performance variability depending on the load of the appliance or host and the speed of its processors. Texas Memory has taken a more challenging route, using FPGAs to execute the most commonly run time-critical software functions. This gives them the effect of running in hardware and speeds up their repeated execution. Higher level, more complex operations are still run on the onboard CPU’s inside the RamSan appliance. This combination should provide the user a more consistent performance condition while under load while at the same time providing field upgradability as that code is improved.
In addition to detailing its RamSan-OS, Texas Memory also announced an eMLC version of the RamSan-720 which was Texas Memory’s first HA product. The RamSan-820 offers the same HA capabilities as the 720 but by using eMLC memory it can now hold 24-TBs of application data. We describe how the flash architectures are designed below:
Storage Swiss Take
Texas Memory can no longer be considered the quiet little memory company down in Texas. The announcement of a highly available eMLC based flash appliance is a bold move into the mainstream storage market place. As we discuss in our article “What is a Storage Hypervisor” combining these capabilities with the storage intelligence in the hypervisor makes the 820 a compelling storage alternative for virtual desktop infrastructures and virtual server environments as well as the more traditional use case of application performance enhancement.
The real boldness in this announcement is the detailing of the RamSan-OS. Many customers assume that all flash appliances are created equal. Texas Memory’s detailing of the RamSan-OS is clear evidence that this is not the case. More so it shows confidence in their approach to operating system design and that their combined use of CPUs and FPGAs as a unique differentiator in their products. Since TMS has always been traditionally a hardware design house, putting some of its RamSan-OS into hardware is a natural evolution of its flash products.
Texas Memory Systems is a client of Storage Switzerland