with GPU in Hybrid Storage Systems Prince Hamandawana, Awais Khan, - - PowerPoint PPT Presentation

with gpu in hybrid storage systems
SMART_READER_LITE
LIVE PREVIEW

with GPU in Hybrid Storage Systems Prince Hamandawana, Awais Khan, - - PowerPoint PPT Presentation

Accelerating the Data Deduplication Performance with GPU in Hybrid Storage Systems Prince Hamandawana, Awais Khan, Changgyu Lee , Sungyong Park, Youngjae Kim Department of Computer Science and Engineering Sogang University, Seoul, Republic of


slide-1
SLIDE 1

Accelerating the Data Deduplication Performance with GPU in Hybrid Storage Systems

Prince Hamandawana, Awais Khan, Changgyu Lee, Sungyong Park, Youngjae Kim Department of Computer Science and Engineering Sogang University, Seoul, Republic of Korea PDSW-DISCS 17 WIP session November 13, 2017, Denver, USA

1

Laboratory for Advanced System Software

slide-2
SLIDE 2

Inline Deduplication in Cloud Storage System

 To achieve high space utilization in Tiered Cloud Storage System, following techniques are discussed in community

1. Compression 2. Erasure Coding

 Can’t remove replicated data across cluster  Difficult to deploy inline mode

3. Inline Data Deduplication

 Higher Storage Efficiency by removing replicated data across cluster  Eliminating duplicated data in Cache tier  But, overhead of inline deduplication directly affects to performance.

 In Hybrid Storage system, Cache-tier nodes equip SSDs and inline deduplication can reduce amount of writes to SSD.

→ Lower Write Amplification, Longer Lifetime

2

slide-3
SLIDE 3

Inline Deduplication Framework on Ceph

3

Storage Node #1 Storage Node #2 Storage Node #3 Cache Node #1 Cache Node #1 Cache Tier (SSD) Storage Tier (SSD)

Fingerprint Index Fingerprint Index

Object CRUSH Algorithm

slide-4
SLIDE 4

1 2 3 4 Object

Inline Deduplication Framework on Ceph

4

Storage Node #1 Storage Node #2 Storage Node #3 Cache Node #1 Cache Node #1 Cache Tier (SSD) Storage Tier (SSD)

Fingerprint Index Fingerprint Index

Chunking

slide-5
SLIDE 5

1 2 3 4 1 2 3 4

Inline Deduplication Framework on Ceph

5

Storage Node #1 Storage Node #2 Storage Node #3 Cache Node #1 Cache Node #1 Cache Tier (SSD) Storage Tier (SSD)

Fingerprint Index Fingerprint Index

Fingerprinting

slide-6
SLIDE 6

Fingerprint Index

1 2 3 4 1 2 3 4

Inline Deduplication Framework on Ceph

6

Storage Node #1 Storage Node #2 Storage Node #3 Cache Node #1 Cache Node #1 Cache Tier (SSD) Storage Tier (SSD)

Fingerprint Index

Deduplication Check Not Duplicate

slide-7
SLIDE 7

Fingerprint Index

1 2 3 4 2 3 4

Inline Deduplication Framework on Ceph

7

Storage Node #1 Storage Node #2 Storage Node #3 Cache Node #1 Cache Node #1 Cache Tier (SSD) Storage Tier (SSD)

Fingerprint Index

Deduplication Check Duplicate

slide-8
SLIDE 8

Fingerprint Index

1 3 4 3 4

Inline Deduplication Framework on Ceph

8

Storage Node #1 Storage Node #2 Storage Node #3 Cache Node #1 Cache Node #1 Cache Tier (SSD) Storage Tier (SSD)

Fingerprint Index

Deduplication Check Increase Reference Count

slide-9
SLIDE 9

Fingerprint Overhead and GPU Acceleration

9

 Deduplication overhead consists of  Chunking  Calculating Fingerprint  Fingerprint Query  We observed fingerprint overhead is more than 70% in total deduplication overhead.  To reduce fingerprinting overhead, we propose to use GPU Acceleration for fingerprinting.

slide-10
SLIDE 10

GPU

Accelerating Fingerprint Calculation with GPU

10

Storage Node #1 Storage Node #2 Storage Node #3 Cache Node #1 Cache Node #1 Cache Tier (SSD) Storage Tier (SSD)

Fingerprint Index Fingerprint Index

GPU Fingerprinting 1 2 3 4

slide-11
SLIDE 11

GPU

Accelerating Fingerprint Calculation with GPU

11

Storage Node #1 Storage Node #2 Storage Node #3 Cache Node #1 Cache Node #1 Cache Tier (SSD) Storage Tier (SSD)

Fingerprint Index Fingerprint Index

GPU Fingerprinting 1 2 3 4 1 2 3 4

slide-12
SLIDE 12

GPU

Accelerating Fingerprint Calculation with GPU

12

Storage Node #1 Storage Node #2 Storage Node #3 Cache Node #1 Cache Node #1 Cache Tier (SSD) Storage Tier (SSD)

Fingerprint Index Fingerprint Index

GPU Fingerprinting 1 2 3 4 1 2 3 4

slide-13
SLIDE 13

Experiment Setup

 Ceph Jewel v10.2.5  CUDA Toolkit 8.0  4 OSD server

 Intel Xeon ES-2640 v3 @ 2.60GHz  32GB memory  12GB NVIDIA Tesla K80 GPU  2 SSDs (Cache Tier), 4 HDD (Storage Tier)

 Ceph RBD Client  Total 1GB size random 4MB writes using fio benchmark

13

slide-14
SLIDE 14

Preliminary Results

14

2 4 6 8 10 12 14 16 18 20 CPU GPU CPU GPU CPU GPU CPU GPU 128 256 512 1024 Total Time (sec) Chunk Size (KB) Chunking Fingerprint Fingerprint Query

65% Reduced

 GPU Fingerprinting reduced about 65% of fingerprint

  • verhead.

 Total Deduplication overhead is reduced to 52%.

slide-15
SLIDE 15

Q&A

15

 Contact: Changgyu Lee (changgyu@sogang.ac.kr) Department of Computer Science and Engineering Sogang University, Seoul, Republic of Korea