CAVE : Channel-Aware Buffer Man-agement Scheme for Solid State
DiskSAC '11
Sungkyu Park, Youngwoo Park, Gyudong Shim and KyuHo Park From KAIST
2011. 06. 02 (Thu)Kwangwoon univ. SystemSoftware Lab.
HoSeok Seo1
2
Introduction The General Architecture of SSD
Maximize the performance with the parallelism utilizing 4 channels and 2 Ways
3
Introduction Multi-channel FTL Scheme
To Achieve the maximum of parallelism- I/O operations are performed in a clustered page unit- Erase operations are performed in a clustered block unit
Related FTL schemes- MCsplit Scheme, the SubGroup Scheme, etc…
1-channel FTL Scheme Low garbage collection overhead instead of a multi-
channel FTL
Related FTL Schemes- BAST, FAST, LAST, etc…
4
Introduction 1-channel FTL Scheme vs Multi-channel FTL
Scheme
5
What is CAVE? CAVE (Channel-aware victim eviction) is
Buffer Management Scheme proposed to achieve good write performance in 1-ch FTL like multi-ch
FTL Scheme used with 1-channel FTL and other buffer Management Scheme
What is Different between CAVE and other buffer man-agement?
Other buffer management - Focus on which data have to are evicted ( = eviction order )- Related buffer scheme : CFLRU, BPLRU, FAB, REF, CLC
CAVE- Focus on How many victims have to are evicted ( = eviction rate )
6
CAVE Algorithm
7
Experimental Setup Used scheme list for experiment
FTL- 1-Channel FTL Scheme : BAST- Multi-Channel FTL Scheme : MCsplit
Buffer Management Scheme- Page-level LRU : CFLRU- Block-level LRU : BPLRU
8
Experimental Setup
- Using Trace-driven simulator- Evaluation Matrix : Hit Rate, Performance
9
Results
Not bad about Hit Ratio with CAVE compared with Conventional Method
10
Results
Good Performance with CAVE compared to Conventional Method
11
Results
w/o CAVE was evaluated with Multi-channel FTL w/o CAVE with Page LRU result is not 100%, it is approximately 100%
12
Summary This paper focus on exploiting full I/O parallelism.
CAVE scheme can reduce it by 55.5%-97.4% in block-level LRU using SYSMARK compared to the conventional method.
