Paper Title: Re-think the Design of RAID upon the Arrival of Computational Storage Drives

Paper Abstract: This proposed talk will present a new RAID design solution that can achieve much more favorable reliability vs. storage cost trade-offs by utilizing computational storage drives (CSDs) with built-in transparent compression. When deploying RAID, users must choose and stick with one specific RAID level after (often painfully) deliberating the speed vs. cost trade-offs of different RAID levels: RAID-1/10 achieve very high IOPS and short rebuild time at the penalty of higher storage cost; in contrast, RAID-5/6 reduce the storage cost by sacrificing the IOPS and rebuild time. The emerging CSDs with built-in transparent compression brings unprecedented opportunities to make RAID deployment much less painful for users. This proposed talk will present a SW Flex-RAID solution that can leverage runtime data compressibility to dynamically maximize the speed performance and minimize rebuild time while maintaining the same fault tolerance. This proposed talk will elaborate on the key data structures and algorithms, and present experimental results that well demonstrate the effectiveness of the SW Flex-RAID solution. This design solution can also be readily integrated into HW RAID card.

Paper Author: Jiangpeng Li, Director of Software Engineering, ScaleFlux

Author Bio: Jiangpeng Li joined Scaleflux at 2015 and now he is leading a software team in ScaleFlux on computational storage product.