Abstract
It is usual in many applications that a large fraction of data pages is, once created, actively updated for a while, and then eventually no longer updated. In flash storage, these frozen pages tend to spread over physical flash blocks and thus are repeatedly relocated upon garbage collection, exacerbating the write amplification. To address the problem of scattered frozen pages, we propose the interface, which allows users or applications to inform the flash storage of frozen pages explicitly. Upon receiving the call, FTL clusters and isolates frozen pages into a physically separate space so that they are no longer relocated upon garbage collection. Experimental results confirm that can reduce the write amplification factor from 25% to 40% of the conventional SSDs. The scheme is novel in the way it opens up a model case of physical tuning knobs in flash storage.
Highlights
Flash memory storage devices have been replacing hard disks as the main storage in most computing platforms, including mobile phones, desktops, enterprise servers, and data centers, thanks to their high IOPS/$ and energy efficiency [1], [2]
GATHER: DESIGN AND IMPLEMENTATION As a solution to the problem of scattered frozen pages, we propose a novel interface for flash storage, called GATHER, which clusters the scattered frozen pages into frozen blocks, a special type of flash blocks distinguished from normal flash blocks
Basic Idea If we know that a set of pages have been newly frozen and can control to gather the pages scattered over different flash blocks and pack them into separate flash blocks (i.e., denoted as frozen blocks in Figure 1.(b)), these frozen blocks will consist of only never-to-be-updated frozen pages so that those frozen flash blocks are unlikely to be chosen as the victim for garbage collection under the greedy victim selection policy [3]
Summary
Flash memory storage devices have been replacing hard disks as the main storage in most computing platforms, including mobile phones, desktops, enterprise servers, and data centers, thanks to their high IOPS/$ and energy efficiency [1], [2]. It is usual in OTLP applications that a large fraction of data pages is, once created, actively updated for a while but eventually no longer updated These frozen pages amount to 80% of total data capacity in the TPC-C benchmark. Because of log-structured writes in flash storage, they tend to be scattered over flash blocks and collocated with non-frozen pages in the same blocks, and are repeatedly relocated upon garbage collection, amplifying writes. This is the first solution to address the problem of scattered frozen pages. In an analogy to defragmentation tools that reduce the disk head movements by reorganizing fragmented files into contiguous areas in harddisk, GATHER reduces the write amplification in flash storages by physically freezing the logically frozen pages from the perspective of garbage collection
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
