1. Understanding Intrinsic Characteristics and System Implications of Flash Memory based Solid State Drives
Feng Chen, David A. Koufaty, and Xiaodong Zhang
2009 ACM SIGMETRICS/Performance
Embedded Lab.
Kim Sewoog

2. Motivation Solid State Drive(SSD) “pivotal technology”
HDD
SSD

3. < SSD Block Diagram >
SSD Internals
Array of flash memory packages
Flash Translation Layer(FTL) in the SSD controller
Hybrid mapping, Over-Provisioning,
Using the original host interface(SATA) for compatibility
Interleaving, DMA for data processing, low power comsumption, etc…
< SSD Block Diagram >

5. Belief and betrayal of SSD
The common belief
Accesses to SSD are uncorrelated with access patterns!
Betrayal
Unexpected performance issues
Uncertain behavior
SSD is not just another ‘faster’ disk!
We need to understand
intrinsic limits
unexpected performance behavior

6. Experiments and analysis
7 questions
Access Patterns of workloads
Random writes
Caching for optimizing performance
Interference between read and write operations
Background operation effects
Internal fragmentation
System implications

7. Measurement environment
Solid State Drives
Experiment System
DellTM PowerEdgeTM 1900 server
Benchmarks
Intel® Open Storage Toolkit : generate various types of I/O workloads
blktrace / blkparse : trace and parse I/O activities (completion event)

8. General Tests Bandwidths
4 distinct workloads : Random/Sequential Read/Write
Random workload : 4KB request size, 1024MB storage space
Sequential workload : 256KB request size
32 parallel jobs, direct I/O, 30 seconds
Comparison with harddisk (WD1600JS)

9. Micro-benchmark Workloads
Various combinations of factors
3 access patterns : Sequential / Random / Stride
10 seconds running, one job, synchronous I/O
Full utilization for initialization (using 256KB sequential write)

10. Distribution of access latencies
Read operations on the SSD
SSD-L : uniform distribution of latencies
SSD-M/H : non-uniform distribution of latencies
Reason :
specification
a readahead mechanism
multi-plane operations
interleaving
65%
65%

11. Distribution of access latencies
Write operations on the SSD
SSD-L : non-uniform distribution of latencies
SSD-M/H : uniform distribution of latencies
Independent of workload access patterns
88%
over 90%

12. Distribution of access latencies
Sequential vs. non-sequential writes on SSD-L
(seems to) use a small buffer
Sequential write : stripped
Non-sequential write : no-stripped
64 requests
initiate the prog. process & write data into flash memory in parallel
from buffer to register
from host to buffer

13. Disk cache effect of SSDs
Large RAM cache(disk cache)
Using hdparm tool to enable and disable the disk cache
Disk cache off : increase of latencies both SSD-M/H
Performance comparision between SSD-M/H without disk cache
SSD-H is good performance -> SLC

14. Interference between read/write operations
Writes : high-cost internal operations
Cleaning and asynchronous write-back of dirty data from the disk cache
Negatively affect foreground read operations
Reads : competition for buffer space with writes
Break sequential patterns
4 workload patterns
Read(n) + Write(n)
Write(n) + Read(n)
Read(n) + Write(n+1)
Read(n) + Write(n+4MB)
only non-sequential pattern
simultaneously, sequential pattern

15. Interference between read/write operations
SSD-L
Substantial degradation
SSD-L optimizes performance for sequential writes
Non-sequential write
Non-shared buffer

16. Interference between read/write operations
SSD-M/H
readahead effect
random read latency
asynchronous write-back

17. Background operation effects
Writes lead almost background operations
Sequential workload using request size of 4KB
Request type : random (50% write requests)
interval time : 10ms
disk cache
Background operations are completed during the idle periods !

18. 16MB: individual mapping unit
Workload randomness effects
Randomness effects (only SSD-L)
Random write : random range from 1GB to 30GB
Stride write : stride step from 4KB to 128MB
Request size : 4KB
16MB: individual mapping unit
metadata synchronization
log block merging

19. Internal fragmentation
Invalid pages in flash memory blocks
Cleaning efficiency : block num x valid page num - read/write, block num - erase
Non-continuous physical pages : readahead mechanism is not effective
Over-provisioning
(25% of the SSD capacity)
No readahead effect

20. Conclusion Many well understood features of SSDs
Many unexpected performance issues
Access Patterns of workloads
Random writes
Caching for optimizing performance
Interference between read and write operations
Background operation effects
Internal fragmentation
System implications

21. Reference
N. Agrawal, V. Prabhakaran, T. Wobber, J. D. Davis, M. Manasse, and R. Panigrahy. “Design tradeoffs for SSD performance”, In Proc. of USENIX’08, 2008.
Blktrace.
S. Lee, D. Park, T. Chung, D. Lee, S. Park, and H. Song. “A log buffer based flash translation layer using fully associative sector translation”. In IEEE Tran. on Embedded Computing Systems, 2007.
M. Mesnier. Intel open storage toolkit.
V. Prabhakaran, T. L. Rodeheffeer, and L. Zhou. “Transactional flash.” In Proc. of OSDI’08, 2008.

22. Thank you !