120 likes | 243 Views
This paper discusses the implications of flash memory as a replacement for traditional disk storage in Database Management Systems (DBMS). While flash offers advantages such as efficient random access, challenges like performance instability, shorter device lifespan, and higher costs persist. It explores potential re-architecting of DBMS to leverage flash's capabilities, focusing on buffer management, indexing, and logging techniques. The paper also considers future perspectives on persistent memory technologies and the need for adaptive software stacks to fully exploit solid-state storage solutions.
E N D
Solid-state storage & DBMS CIDR 2013 Manos Athanassoulis
Flash is replacing disks Flash not a drop-in replacement for disks But, random writes on flash • Instability in performance • Shorter device lifetime And flash capacity is expensive Is flash ready/suitable for secondary storage? Reverse: Howcan DBMS exploit flash?
Disks all over the place Re-architect DBMS needed for solid-state storage Buffers • DBMS modules assumes disks • Storage • Buffer management • Indexing • Logging • Lower-level: • Block APIs • Storage software stack
How to use flash in a “disk-world”? Buffers Updates • As a specialized write cache • Hides update overhead
How to use flash in a “disk-world”? Buffers • As a specialized write cache • Hides update overhead • As secondary storage • Offers efficient random access
How to use flash in a “disk-world”? SSDBuffer Buffers • As a specialized write cache • Hides update overhead • As secondary storage • Offers efficient random access • As a “regular” cache level
Flash Wall Capacity: Density will plateau (soon) Endurance (already)
PCM, memristor and co More than flash: How to use persistent main memory?
Flash Wall(s) Persistent main memory OS & FS too slow Thank you! Manos Athanassoulis