GrassHopperFS(GHFS) 일반 구성도

GHFS는 Metadata 서버가 없는 No Metadata구조로 Elastic Hashing 알고리즘을 통해 Metadata Server 구조의 분산 파일 시스템의 단점인 SPOF를 보완하여 진정한 분산 처리 스토리지를 구현하고 있습니다.

모니터링 및 관리영역		GHFS 영역

Performance for GlusterFS

How to using iozone

-> iozone -w -c -e -i 0 -+n -r 64k -s 2g -t {1-9} -F /mnt/Storage/f{0,1,2,3,4,5,6,7,8}.ioz

• -i : test type, 0=write/rewrite, 1=read/re-read, 2=random-read/write, 3=read-backward, 4=Re-write-record
• -r : data transfer size, block size (64KB, 2MB)
• -s : per-thread file size
• -t : number of thread
• -w, -c

Sample Write

Sample Read

glusterFS vs Ceph vs HDFS

Test Methods (HDFS vs Ceph vs GlusterFS)


테스트 방법

• # iozone -r 128k -i 0 -i 1 -i 2 -t 24 -s 10G

테스트 조건

• HDFS
- 20 datanodes, 1 namenode
• Ceph
- V0.61 : 3 Monitors nodes, 1 Metadata Server, 120 Object Storage Daemon(20 Nodes)
- V0.67 : 3 Monitors nodes, 1 Metadata Server, 95 Object Storage Daemon (19 Nodes)
• GlusterFS
- v3.3 : 21 nodes, v3.4 : 20 nodes

 출처 : http://iopscience.iop.org/1742-6596/513/4/042014/pdf/1742-6596_513_4_042014.pdf

Test Results (HDFS vs Ceph vs GlusterFS)

• Test results using iozone. MB/s

	HDFS		Ceph		GlusterFS
	24 Threads	36 Threads	v0.61	v0.67	v3.3	v3.4
Initial Write	239.72	N.A.	52.49	18.93	234.06	306.34
Re-Write	ERROR	ERROR	54.05	19.31	311.75	407.90
Random Write	ERROR	ERROR	ERROR	13.96	326.89	406.33
Read	155.18	195.65	95.38	53.40	621.08	688.06
Re-Read	151.33	207.43	102.04	57.29	662.92	711.46
Random Read	29.06	39.98	ERROR	5.13	242.75	284.00