On Fri, Mar 21, 2014 at 7:32 AM, Matias Bjørling <m@xxxxxxxxxxx> wrote: > Hi, > > Users that have custom firmware SSDs, may choose to expose their flash directly. > This allows the host to control logical to physical address mappings, garbage > collection strategy, wear-leveling, and so on. > > This is beneficial when you either want to strip the cost of a costly controller > or want detailed control over an SSD. Such as implementing key-value stores, > object-stored, atomic I/Os, etc. > > LightNVM implements the host-side core. It supports two modes. Users that > doesn't have hardware available and want to emulate an SSD, and users with a > physical SSD, that has a LightNVM compatible firmware. > > The behavior of the core can be exposed through the various components that make > up an FTL. Thus, it is not single implementation, but instead possible to > mix the best algorithms for the user-space application workloads. > > Currently, LightNVM implements a page-based FTL, cost-based GC and simple > wear-leveling. Additionally, it allows simulation of flash timings when hardware > isn't available. There's work on getting it to the OpenSSD platform with a > custom firmware and getting it stable for broad use. > > There is still much work to do. I'm looking for feedback on the approach, dm > integration, and more. Any feedback is greatly appreciated. > > A presentation was given at the recent Non-Volatile Memory Workshop (NVMW) > workshop. Slides are available at: http://bjorling.me/NVMW2014-LightNVM.pdf > > Major todo's: > * Patch has TODO and FIXME in places that needs to be cleaned. > * LightNVM compatible firmware for the OpenSSD platform and integration. > * Performance regressions during GC. > * Durability during power failure. This sounds very interesting! Is there also a way to expose the flash directly as MTD device? I'm thinking of UBI. Maybe both projects can benefit from each others. -- Thanks, //richard -- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
