Vous êtes sur la page 1sur 14

HTC

Storage POC

敦陽科技 HDS
郭勝利 孫逸芳 陳治愷 馮怡碩
Victor Yvonne Tony Essau
Agenda
Architecture
Environment
Testing items
Testing procedure
Q&A

2
Architecture
HDS 9570V

5 0

13 12

Benchmark Factory 9
Sun V60x 8
172.32.83.165 1
L2 Switch 4
100 Mb/s UTP 0B 1A 1B
0A
SAN Switch
2 Gb/s FC
HDS Tuning Manager Sun F6800 172.32.100.87
Sun V60x 172.32.83.96
172.32.83.161

HDS AMS2100
Sun F4800 172.32.12.240
3 172.32.83.97
Environment
Oracle DB Server: Sun Fire 6800
Solaris 8, Oracle DB 8.1.7
750 MHz x8, 16GB RAM
100 Mb/s Ethernet, 2 Gb/s FC
HDS AMS2100
4 GB cache, 300GB 15K rpm x 20, 450GB 15K rpm x9
Total 140 x 24 = 3360 IOPS (or 180 x 24 = 4320 IOPS for ideal value)
LUN configuration

Oracle Type HD Configuration Mount Point Size (usable)


Redo log Raid 1 (1+1) (450GB) /ora_redo 50GB
Archive log Raid 1 (1+1) (450GB) /ora_archive 50GB
/ora_index, 350GB,
Index, Sys temp Raid 1+0 (2+2) *2 (300GB)
/ora_temp 150GB
Data Raid 1+0 (2+2) *3 (300GB) /ora_data 850GB
Clone Raid 5 (4+1) (450GB) N/A N/A

4
Testing Items
1. Storage utilization monitoring
2. Data migration moving time
3. Multi-LUN snapshot/DB consistency
4. Load balancing—path, controller
5. Failover—path, controller
6. Single table full table scan
7. 2 HBA vs. 4 HBA performance
8. Two tables full table scan

5
1. Storage utilization monitoring
HDS Tuning Manager report:
Controller CPU utilization
Cache write pending
Port IOPS

6
2. Data migration moving time
Copy 30GB (2GB x 15) data from 9570
to AMS 2100 independent LUN
Both one path only (9570-to-F6800,
F6800-to-AMS2100)
It take about 10 minutes
預估 copy 800GB 需要 800/30 x 10 ≈
270min.=4.5 hrs
安裝軟硬體、設定、測試等等約需 3.5hrs
容錯時間 2hrs
總共停止服務時間 4.5+3.5+2=10 hrs
7
3. Multi-LUN snapshot/DB consistency
All LUNs of Oracle pairs have created
already (HDS ShadowImage)
Insert time stamp record to a independent
table while running stress
Split pair, record current time
F4800 mount all secondary LUN
Startup Oracle on F4800
Check time stamp record
Join pairs then split again
Check time stamp record for resync. status
8
4. Load balancing
Run Quest Benchmark Factory to
generate loading (TCP-B in 100 users, the
same follows)
Observe port IO from SAN switch
Port 0,1,4,5,8,9,12,13 should have value
Unplug FC cable from port 0, 12  port
0,5,12,13 IO drop to zero (path failover
also)
Plug FC cable back to port 0,12  port
0,5,12,13 IO get value
9
5. Failover--controller
Run Benchmark Factory to generate
loading
Observe port IO from SAN switch
Port 0,1,4,5,8,9,12,13 should have value
Unplug controller 1  port 0,5,12,13 IO
drop to zero
Plug controller 1 back to port 0,12 
port 0,5,12,13 IO get value (after 3 min.
boot up)

10
6. Single table full table scan
Run Benchmark Factory to generate
loading
SQL command to run single table full
table scan
Observe HDS Tuning Manager port
IOPS or iostat command on F6800
Gather statistics from r/s and w/s of
iostat output , it reached a maximum
of 3200 IOPS
11
7. 2 HBA vs. 4 HBA performance
Run Benchmark Factory to generate
loading
Run iostat command on F6800
Observe IO busy status b%
It showed around 35% busy for 4
HBA
Unplug FC cable from port 0, 12
b% value will raising, it showed
around 70% busy for 2 HBA
12
8. Two tables full table scan
Run Benchmark Factory to generate
loading
SQL command to run two tables full
table scan
Observe HDS Tuning Manager port
IOPS or iostat command on F6800
Gather statistics from r/s and w/s of
iostat output , it reached a maximum
of 4000 IOPS
13
Q&A

Vous aimerez peut-être aussi