cloudxtreme/destor

An experimental platform for chunk-level data deduplication. Key words: DDFS, Sparse Index, Extreme Binning, SiLo, Sample Index, BLC; CBR, CFL, CAP, HAR; ASM, OPT; GC, Cumulus

General Information

Destor is a platform for data deduplication evaluation.

Features

1. Container-based storage;
2. Chunk-level pipeline;
3. Fixed-sized chunking, Content-Defined Chunking (CDC) and an approximate file-level deduplication;
4. A variety of fingerprint indexes, including DDFS, Extreme Binning, Sparse Index, SiLo, etc.
5. A variety of rewriting algorithms, including CFL, CBR, CAP, HAR etc.
6. A variety of restore algorithms, including LRU, optimal replacement algorithm, rolling forward assembly.

Related papers

1. The chunking algorithm:

   a) A Low-bandwidth Network File System @ SOSP'02.

   b) A Framework for Analyzing and Improving Content-Based Chunking Algorithms @ HP technical report.

   c) AE: An Asymmetric Extremum Content Defined Chunking Algorithm for Fast and Bandwidth-Efficient Data Deduplication @ IEEE Infocom'15.

2. The fingerprint index:

   a) Avoiding the Disk Bottleneck in the Data Domain Deduplication File System @ FAST'08.

   b) Sparse Indexing: Large Scale, Inline Deduplication Using Sampling and Locality @ FAST'09.

   c) Extreme Binning: Scalable, Parallel Deduplication for Chunk-based File Backup @ MASCOTS'09.

   d) SiLo: A Similarity-Locality based Near-Exact Deduplicatin Scheme with Low RAM Overhead and High Throughput @ USENIX ATC'11.

   e) Building a High-Performance Deduplication System @ USENIX ATC'11.

   f) Block Locality Caching for Data Deduplication @ SYSTOR'13.

   g) The design of a similarity based deduplication system @ SYSTOR'09.

3. The fragmentation:

   a) Chunk Fragmentation Level: An Effective Indicator for Read Performance Degradation in Deduplication Storage @ HPCC'11.

   b) Assuring Demanded Read Performance of Data Deduplication Storage with Backup Datasets @ MASCOTS'12.

   c) Reducing impact of data fragmentation caused by in-line deduplication @ SYSTOR'12.

   d) Improving Restore Speed for Backup Systems that Use Inline Chunk-Based Deduplication @ FAST'13.

   e) Accelerating Restore and Garbage Collection in Deduplication-based Backup Systems via Exploiting Historical Information @ USENIX ATC'14.

   f) Reducing Fragmentation for In-line Deduplication Backup Storage via Exploiting Historical Information and Cache Knowledge @ IEEE TPDS.

4. The restore algorithm:

   a) A Study of Replacement Algorithms for a Virtual-Storage Computer @ IBM Systems Journal'1966.

   b) Improving Restore Speed for Backup Systems that Use Inline Chunk-Based Deduplication @ FAST'13.

   c) Accelerating Restore and Garbage Collection in Deduplication-based Backup Systems via Exploiting Historical Information @ USENIX ATC'14.

5. Garbage collection:

   a) Building a High-Performance Deduplication System @ USENIX ATC'11.

   b) Cumulus: Filesystem Backup to the Cloud @ FAST'09.

   c) Accelerating Restore and Garbage Collection in Deduplication-based Backup Systems via Exploiting Historical Information @ USENIX ATC'14.

The Destor paper

[FAST'15] Design Tradeoffs for Data Deduplication Performance in Backup Workloads.

This paper presents the parameter space in data deduplication that guides the design of Destor. It then gives the overall architecture and the backup/restore pipeline in Destor. Finally, we did an entensive experimentation via Destor to find reasonable solutions. You can find the paper in doc directory.

Recent publications using Destor

1. Min Fu, Dan Feng, Yu Hua, Xubin He, Zuoning Chen, Wen Xia, Fangting Huang, and Qing Liu. Accelerating Restore and Garbage Collection in Deduplication-based Backup Systems via Exploiting Historical Information @ USENIX ATC'14.
2. Jian Liu, Yunpeng Chai, Xiao Qin, and Yuan Xiao. PLC-Cache: Endurable SSD Cache for Deduplication-based Primary Storage @ MSST'14.
3. Yucheng Zhang et al. AE: An Asymmetric Extremum Content Defined Chunking Algorithm for Fast and Bandwidth-Efficient Data Deduplication @ IEEE Infocom'15.
4. Min Fu et al. Reducing Fragmentation for In-line Deduplication Backup Storage via Exploiting Historical Information and Cache Knowledge @ IEEE TPDS.

Environment

Linux 64bit.

Dependencies

1. libssl-dev is required to calculate sha-1 digest;

2. GLib 2.32 or later version

   libglib.so and glib.h may not be found when you first install it.

   The header files (that originally locate in /usr/local/include/glib-2.0 and /usr/local/lib/glib-2.0/include) are required to be moved to a searchable path, such as "/usr/local/include".

   Also a link named libglib.so should be created in "/usr/local/lib".

3. Makefile is automatically generated by GNU autoconf and automake.

INSTALL

If all dependencies are installed, compiling destor is straightforward:

./configure

make

make install

To uninstall destor, run

make uninstall

Running

If compile and install are successful, the executable file, destor, should have been moved to /usr/local/bin by default. You can create a config file, destor.config, in where you run destor. A sample destor.config is in the project directory.

NOTE: run rebuild script before destor to prepare working directory and clear data.

destor can run as follows:

1. start a backup task,

   destor /path/to/data -p"a line as in config file"

2. start a restore task,

   destor -r /path/to/restore -p"a line as in config file"

3. show the current statistics of system,

   destor -s

4. show help

   destor -h

5. make a trace

   destor -t /path/to/data

Configuration

A sample configuration is shown in destor.conf

To find what the parameters in destor.conf exactly mean and how to configure an existing solution (such as DDFS), please read the paper Design Tradeoffs for Data Deduplication Performance in Backup Workloads in doc/. The parameter space is based on the taxonomy proposed in the paper. (Note: The paper is somewhat difficult to follow. I am sorry about that, still working on improving the readability.)

Bugs

1. If the running destor is crashed artificially or unexpectedly, data consistency is not guaranted and you'd better run rebuild script.

2. Do NOT support concurrent backup/restore.

3. If working path in destor.config is modified, the rebuild script must be modified too.

4. CMA assumes the backups are deleted in FIFO order.

   If you set a backup-retention-time, the expired backup is deleted automatically.

Author

Min Fu

Email : fumin at hust.edu.cn

Blog : fumin.hustbackup.cn

(Feel free to contact me if you have any questions about Destor. I would appreciate bug report.)