Weekly paper read

Week9:

Search for papers 1. * (reference) title: Improving Performance and Capacity of Flash Storage Devices by Exploiting Heterogeneity of MLC Flash Memory author: Sungjin Lee; Jihong Kim from: ieee transactions on computers content: (1) a mlc nand flash memory, each memory cell can be programmed as a single-level cell or a multi-level cell at runtime (2) present a flexible flash file system, called FlexFS, (3) To provid E high performance and high capacity simultaneously, FlexFS employs a dynamic free-space management (DFM) technique. (4) FlexFS also adopts a novel dynamic lifetime management (DLM) technique, which manages the storage lifetime by controlling the use of SLC-mode programming. (5) In order to achieve the SLC performance, FlexFS writes as your data as possible toflash memory using fast SLC-mode program Ming. * (6) Free-space reclamation -- idle time (7) The dynamic free-space management (DFM) technique resolves the problems caused by improper management of free space by maintaining minimal but sufficient free space (8) The dynamic life time management (DLM) technique adaptively controls the wearing rate offlash memory (which is accelerated by free-space reclamation) so that a reasonable storage lifetim E can be provided (9) FlexFS is based on a JFFS2file system (10) same pages? (11) 2 write buffer & log (12) free-space reclamation is already Med only when there are no user I/O activities and on-demand free-space reclamation is not required (see Section 3.2 ). free-space reclamation is already Med only when there are no user I/O activities and on-demand free-space reclamation is not required. (13) If an observed idle period is longer than a certain threshold value, FlexFS triggers free-space reclamation, expecting that there will be a long idle period. [50 MS] (14) delayed freespace reclamation does not trigger free-space reclamation unless available free space is smaller than TH. (15) a delayed free-space reclamation policy that delays free-space reclamation as long as possible so that each data are invalidated in the SLC region. [SLC invalid page] (16) DLM [w]

2. title: A Workload-Aware-Design of 3D-NAND Flash Memory for Enterprise SSDs author: Chao Sun (Japan) from: 15th Int 'l Symposium on Quality Electronic Design (2014) content: (1) SCM/NAND flash hybrid

3. title: Differentiated Storage Services autor: Michael Mesnier, Feng Chen, Tian Luo from: SOSP '11 content: (1) I/O classification architecture (2) As examples, the "middle" of a disk can be used to reduce seek latency, and the "outer tracks" can be used to improve transfer speeds. (3) I/O is classified as metadata, journal, directory, or file, and file I/O is further classified by the file size (e.g ., ≤4kb ≤16kb,...,> 1 GB ). (4) we modify the OS block layer so that every I/O request carries a classifier. we copy this classifier into the I/O command. in this way, a storage system can provide block-level differentiated services (performance, reliability, or security) (5) This paper focuses on priorities (6) our storage systems implement a priority-based performance policy, so we map each class to a priority level (7) For the FS, the priorities reflect our goal to reduce small random access in the storage system, by giving small files and metadata higher priority than large files. for the DB, we simply demonstrate the flexibility of our approach by assigning caching policies ies to common data structures (indexes, tables, and logs ).

4. * title: Hystor: Making the Best Use of Solid State Drives in High Performance Storage Systems autor: Feng Chen David Koufaty from: ICS '11 (2011) content: * (1) improvements to LRU disadvantages (cache pollution problem) (2) a special data structure, called block table. for large-scale history recode. (3) automatically learns * (4) identifies semantically-critical blocks (e.g. file system metadata) (5) Hybrid HDD and SDD (6) request size, frequency, seek distance, reuse distance, (7) use the blktrace tool (8) (frequency/request_size) is found to be the most either tive one

Thought (1) Multiple Priority dimensions, such as updating faster than writing? --- Order (2) comparison, better than LRU, better than size threshold, can be from the correct allocation ratio, page collection (3) SLC1: for random; SLC2: for cache (4) reads and writes in SLC first, and then does not need to be scheduled to TLC (5) request size, frequency, // logical move distance, reuse distance, size change // on time requests, on time workload (6) SLC/TLC Range (7) [write paper] --- Details Introduction (8) SLC ---> TLC idle time [find a suitable time] & 1.14 & 1.15 [paper1: 3.21, fig7]

 

Week10:

 

Weekly paper read