StreamFast SSD Concept Cuts Memory Use by Ditching FTL and DRAM
▼ Summary
– Hammerspace founder David Flynn proposes a new SSD architecture called StreamFast that eliminates the Flash Translation Layer (FTL) and its controller DRAM.
– The current SSD model requires a large amount of DRAM, roughly one byte per kilobyte of flash, creating a significant memory overhead for high-capacity drives.
– StreamFast instead uses device-assigned sequential addresses, writing data streams in order and returning those addresses to the host file system.
– This design drastically reduces memory needs to about one byte of RAM per megabyte of flash, a thousand-fold improvement, and also cuts write amplification and heat.
– The cooler, simpler drives could suit power-limited environments, and the concept is being developed with partners in the flash ecosystem.
A new solid-state drive architecture promises to revolutionize data storage by dramatically cutting memory requirements and power consumption. The StreamFast concept, proposed by Hammerspace founder David Flynn, eliminates two core components of traditional SSDs: the Flash Translation Layer (FTL) and the controller’s DRAM. Instead, it employs a file-system-centric design where the device itself assigns sequential addresses to incoming data streams. This fundamental shift claims to reduce the memory needed for large-capacity drives by a factor of one thousand, potentially enabling cooler, simpler, and more reliable high-capacity storage solutions.
The current standard SSD model faces a significant scaling challenge, often referred to as the DRAM crunch. Flynn explains that controllers typically require one byte of DRAM for every kilobyte of NAND flash memory. This ratio becomes staggering at petabyte scales, demanding a terabyte of DRAM alongside a petabyte of flash, a costly and power-intensive proposition. This overhead is exacerbated by industry trends shifting DRAM production toward high-bandwidth memory for AI and GPU applications, creating supply and cost pressures. Flynn’s proposal tackles this by removing the FTL entirely, allowing the host file system to interact directly with the flash memory in a more native way.
The core innovation of StreamFast lies in its handling of data writes. Instead of the controller mapping data to arbitrary physical locations using the FTL, the SSD accepts incoming data as a sequential stream. The device assigns its own consecutive addresses to these data strings and returns those addresses to the host’s file system for tracking. This method turns a random write process into an ordered, sequential one. The major advantage is in failure recovery; if an interruption occurs, the host system can simply replay the data stream from the last known point rather than relying on a massive in-drive address map held in volatile DRAM.
This architectural change yields profound efficiency gains. Flynn states that with the StreamFast file system, the requirement drops to roughly one byte of RAM for every megabyte of flash. For a hypothetical one-petabyte SSD, this translates to needing only about one gigabyte of host memory, compared to the terabyte of DRAM that would traditionally be embedded within the drive itself. Beyond the massive memory savings, removing the FTL and its associated DRAM also reduces write amplification, a process that wears out flash cells, and lowers heat generation, as the controller DRAM is often a primary thermal hotspot.
The implications extend to drive reliability and deployment scenarios. A simpler controller with less heat output could lead to more robust hardware. Flynn suggests these cooler, less complex drives could be ideal for power-constrained environments, such as sealed server racks or even futuristic orbital data centers where thermal management is critical. The development is being pursued under a new StreamFast business entity collaborating with the Open Compute Platform’s Open Flash Project, while Hammerspace continues its work on global file system software. While Flynn remained discreet about specific manufacturing partners, he hinted at active engagement with key players in the flash memory ecosystem. The concept represents a significant rethinking of SSD fundamentals, aiming to pave the way for the next generation of high-capacity, energy-efficient storage.
(Source: TechRadar)
