PPT

Report
Leverage Similarity and Locality to Enhance
Fingerprint Prefetching of Data Deduplication
Yongtao Zhou, Yuhui Deng, Junjie Xie
Department of Computer Science, Jinan University, Guangzhou, 510632,
P. R.China
ICPADS 2014
1
Agenda
• Introduction
• Related work
• Motivation
• System overview
• Evaluation
ICPADS 2014
2
Introduction
• IDC: 95% redundant data in the backup systems; 75% redundant data across the
digital world
 Consumes IT resources and expensive network bandwidth
• Data deduplication: eliminate redundant data by storing only one data copy
Fingerprint
Chunk
algorithm
Files
Fingerprints to large too
store all fingerprints in
memory
MD5
Data blocks
Hash Table
B+ Tree
A
B
G
ICPADS 2014
E
C
3
Introduction
• Querying fingerprints incurs disk bottleneck
 The size of fingerprints are too large too be cached in memory
 Cache hit ration very low (lack temporal locality)
 The IOPS of disk drivers is limited
800TB unique data
MD5 signature Avg.8KB
Chunk
A large portion of fingerprints have to be stored on disk drives
ICPADS 2014
4
Locality based strategies: DDFS
• Locality: segments tend to reappear in the same or very similar sequences with other
segments. This is because most data from previous backup has a slight modifications.
RAM
Prefetching!!!!
Index
A B C A
...
Bloom Filter
...
A E I A
ABC
DE F
H I J
Poor deduplication performance
when there is little or no locality in
datasets
?
ICPADS 2014
5
Similarity based method: Extreme Binning
RAM
DISK
Similarity Index
C1 C2 Cn
Bin(C1)
Files
Bin(C2)
Bin(Cn)
Fail to identify and thus remove
significant amounts of redundant data
when there is a lack of similarity among
files
It uses a two-leve lindex structure made up of similarity characteristic value and the granularity of bin.
Extreme Bining stores the similarity characteristics value in RAM.
Extreme Bining only identifies the redundant data in the same bin, even though neighbouring bins
may have identical data blocks.
This results in some redundant data blocks so as to degrade the deduplication ratio.
the deduplication ration of Extreme Bining heavily relies on the similarity degree of data streams.
ICPADS 2014
6
SSD based approach
• Fingerprint lookup disk bottleneck
 The IOPS of disk drive is limited
HDD VS SSD
• Some studies alleviate disk bottleneck by using SSD
 Dedupv1, ChunkStash
• SDD is still very expensive in contrast to disk drives.
• The performance of random and small writes becomes a new bottleneck of SSD
ICPADS 2014
7
Our approach
• A fingerprint prefetching approach by using the file similarity to enhance the
deduplication performance
• The locality of fingerprints are maintained by arranging the fingerprints in terms of the
sequence of the backup data stream
• The overhead of different similarity identification algorithms are investigated, and the
impacts of those algorithms on data deduplication are evaluated in contrast to previous
studies Extreme Binning, Silo, FPP
• This approach does not impact the deduplication ration
ICPADS 2014
8
System architecture
Implementation in LessFS
Implementation in Tokyo Cabinet
ICPADS 2014
9
Storage structure for fingerprints
The locality of fingerprints
are maintained by arranging
the fingerprints in terms of
the sequence of the backup
data stream
Loss the locality of
fingerprints
ICPADS 2014
10
The process of fingerprints prefetching
ICPADS 2014
11
Evaluation
• Implement a real prototype based on LessFS and Tokyo Cabinet
• Three similarity identification algorithms FPP, PAS and Simhash are implemented in
the Similar File Identification Module
• Ubuntu operation system(Kernel version is 3.5.0-17) ,1GB memory, 2:4GHz Intel(R)
Xeon(R) CPU
• We take four full backups to evaluate the system like what DDFS does.
• Four data sets backup1, backup2, backup3 and backup4 to perform the evaluation
 10GB, 15GB, 20GB and 25GB, and the numbers of files are 3073, 4694, 6539 and 9910,
respectively.
 We choose fixed-size chunk algorithm. The chunk size is 4KB, 8KB, 16KB, 32KB, 64KB and 128KB
ICPADS 2014
12
FPP and PAS
ICPADS 2014
13
Simhash
• Simhash is a member of the local sensitive hash
• Simhash has the property that the fingerprints of similar files differ in a small number of
bit positions
• Actual runs at Google web search engine
ICPADS 2014
14
Data sets
The file size distribution matches the previous studies.
ICPADS 2014
15
Deduplication ratio
• We measure the size of unique data blocks by using three different similarity
identification algorithms including FPP, PAS and Simhash with four full backups
• When the chunk size is 4KB, the unique data blocks are 14GB, and the data
deduplication ratios are 3.93 across the three cases.
• The performance is the same as that of the baseline system LessFS.
ICPADS 2014
16
Time overhead of fingerprint lookup
•  : the time of similarity detection
•  : the time of fingerprint prefetch
•  : the time of fingerprint lookup
• The overall overhead of fingerprint lookup  =  +  + 
• For Base has  =  = 0
ICPADS 2014
17
Time overhead of fingerprint lookup
ICPADS 2014
18
CPU utilization
ICPADS 2014
19
Memory utilization
ICPADS 2014
20
Conclusion
• Proposes a fingerprint prefetching approach by preserving the locality of fingerprint in
the form of backup data stream as well as taking advantage of file similarity
• The proposed method can effectively alleviate the disk bottleneck with acceptable
overhead of CPU, memory, and storage when performing fingerprint lookup, thus
improving the throughput of data deduplication
• Does not impact the data deduplication ratio
ICPADS 2014
21
Reference
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
SSD: http://en.wikipedia.org/wiki/Solid-state_drive
HDD vs SSD: http://www.diffen.com/difference/HDD_vs_SSD
D. Bhagwat, K. Eshghi, D. D. Long, and M. Lillibridge, “Extreme binning: Scalable, parallel deduplication for chunk-based file backup,” in Modeling, Analysis
& Simulation of Computer and Telecommunication Systems, 2009. MASCOTS’09. IEEE International Symposium on. IEEE, 2009, pp. 1–9.
W. Xia, H. Jiang, D. Feng, and Y. Hua, “Silo: a similarity-locality based near-exact deduplication scheme with low ram overhead and high throughput,” in
Proceedings of the 2011 USENIX conference on USENIX annual technical conference. USENIX Association, 2011, pp. 26–28.
A. Z. Broder, M. Charikar, A. M. Frieze, and M. Mitzenmacher, “Min-wise independent permutations,” Journal of Computer and System Sciences, vol. 60,
no. 3, pp. 630–659, 2000.
Y. Zhou, Y. Deng, X. Chen, and J. Xie, “Identifying file similarity in large data sets by modulo file length,” in Proceedings of the 14th International Conference
on Algorithms and Architectures for Parallel Processing. IEEE, 2014.
D. Meister and A. Brinkmann, “dedupv1: Improving deduplication throughput using solid state drives (ssd),” in Mass Storage Systems and Technologies
(MSST), 2010 IEEE 26th Symposium on. IEEE, 2010, pp. 1–6.
B. Debnath, S. Sengupta, and J. Li, “Chunkstash: speeding up inline storage deduplication using flash memory,” in Proceedings of the 2010 USENIX
conference on USENIX annual technical conference. USENIX Association, 2010, pp. 16–16.
Y. Deng, “What is the future of disk drives, death or rebirth?” ACM Computing Surveys (CSUR), vol. 43, no. 3, p. 23, 2011.
B. Zhu, K. Li, and R. H. Patterson, “Avoiding the disk bottleneck in the data domain deduplication file system.” in Fast, vol. 8, 2008, pp. 1–14.
J. Gantz and D. Reinsel, “The digital universe decade-are you ready,” IDC iView, 2010.
S. Quinlan and S. Dorward, “Venti: A new approach to archival storage.” in FAST, vol. 2, 2002, pp. 89–101.
M. Ruijter, “Lessfs,” http://www.lessfs.com/wordpress/.
F. Labs, “Tokyo cabinet,” http://fallabs.com/tokyocabinet/.
M. Lillibridge, K. Eshghi, D. Bhagwat, V. Deolalikar, G. Trezis, and P. Camble, “Sparse indexing: Large scale, inline deduplication using sampling and
locality.” in Fast, vol. 9, 2009, pp. 111–123.
G. S. Manku, A. Jain, and A. Das Sarma, “Detecting nearduplicates for web crawling,” in Proceedings of the 16th international conference on World Wide
Web. ACM, 2007, pp. 141–150.
ICPADS 2014
22
Thank you!
Question?
ICPADS 2014
23

similar documents