Transcript Oracle Exadata with ZFS NAS Appliance for faster platform

Oracle EXADATA and ZFS
NAS Appliance for Faster
Platform as Service
11
Pranilesh Chand
Lead Engineer - Database
Oracle Certified Professional 9i, 10g,11g
Speciality:
Performance Tuning
Oracle RAC/HA
Oracle Exadata
Disaster Recovery
22
Agenda
•
•
•
•
•
Technology Overview
Why we chose this solution
How we did it
Issues
Q&A
33
SUN Oracle Database Machine
•
2 Sun Fire™X4170 Oracle Database servers
•
3 Exadata Storage Servers (All SAS or all SATA)
•
2 Sun Datacenter InfiniBand Switch
•
Preconfigured out of the box high performance machine
•
Balanced performance configuration
•
Oracle database 11.2 preinstalled
•
No ASM Instance
•
Smart Scan
44
Sun ZFS Storage 7420
•Only NAS appliance that supports Hybrid Columnar Compression
•Cluster Ready System for HA
•Can be easily integrated with Oracle database
•ZFS Filesystem
•Inline de-duplication and compression to save space
•Filesystem snapshot and cloning
55
Why We Chose This
Solution
66
Problem Presented
•Performance Slowness
•Batch jobs taking 9hrs to run
•User experience affected as some had to work in weekends
•Database backups
•OLTP reports not able to run and times out
•Some important reports had to run in weekend as it took 14hrs
•No High Availability
•Database Refresh
•Takes almost 16-20hrs to refresh
77
Problem Analysis
Performance
•Database IO Bound
•Server Low RAM
Environment Refresh
Cost
88
Solution Accepted
•Platform as a service to customer
•Oracle Exadata for production DB hosting
•Consolidation of all production databases
•Sun ZFS and Sun Fire servers for test/dev environment
•RAC environment to be created
•Consolidation of all test/dev databases
99
Exadata Benefits
Data Warehouse Jobs
•From 9hrs to 90mins
•From 8hrs to 40mins
•ETL now running in minutes from hours before
OLTP Performance
•Reports running in minutes from hours
Business Improvement
•Staff engagement
•Cost
Faster RMAN Backup
Consolidation
10
10
Oracle ZFS Appliance Benefits
Hybrid Columnar Compression
Quick Refresh of Test/Dev from standby database
Consolidated platform for test databases
Attached to Exadata via IB
Clone DB requires very small space
11
11
How We Did It
12
12
Backup: High Level Design
OEM Server
Exadata Server
OEM Repository/
RMAN Catalog
Tape
Backup
OEM agent
10G Ethernet
Backup Server
/backup (DNFS mount)
Compute node Operating System
Oracle RDBMS/RAC binaries
OCR and voting disk backup
Database RMAN backups
Regular IB configuration export
SAN
13
13
Sample: OEL Backup Script
# Create snapshot backup of filesystems /, /u01, /boot
lvcreate -L1G -s -n root_snap /dev/VGExaDb/LVDbSys1
e2label /dev/VGExaDb/root_snap DBSYS_SNAP
mkdir -p /root/mnt
mount /dev/VGExaDb/root_snap /root/mnt
lvcreate -L5G -s -n u01_snap /dev/VGExaDb/LVDbOra1
e2label /dev/VGExaDb/u01_snap DBORA_SNAP
mkdir -p /root/mnt/u01
mount /dev/VGExaDb/u01_snap /root/mnt/u01
# Backup the snapshots to the NFS mount /OS_backup
cd /root/mnt
# Check for any significant errors reported in the /tmp/backup_tar.stderr file
tar -pjcvf /OS_backup/node1_backup.tar.bz2 * /boot --exclude node1_backup.tar.bz2 >/tmp/backup_tar.stdout
2>/tmp/backup_tar.stderr
# Remove snapshots
cd /
umount /root/mnt/u01
umount /root/mnt
/bin/rm -rf /root/mnt
lvremove -f /dev/VGExaDb/u01_snap
lvremove -f /dev/VGExaDb/root_snap
14
14
Backup: IB Switch
15
15
Monitoring: High Level Design
OEM Server
ASR Manager
Triggered
Alert
SMS Sent
OEM Agent
Breached
Hardware Alert
Breached SoftwareAlert
SR Details
Exadata
Machine
Gen-I Engineer
Oracle Support
Exadata Storage cells registered in ASR
Exadata compute nodes registered in ASR
Compute node and Storage Cell ILOM registered in ASR
16
16
DR – High Level Design
Primary Site
DR Site
OEM Server
OEM Server
OEM repository/
RMAN Catalog
OEM repository/
RMAN Catalog
Redo Shipping
OEM Agent
10G Ethernet
Load Balancer
Dark Fiber
OEM Agent
10G Ethernet
RAC Database
Exadata
Exadata
IB
RAC
Database
Redo Shipping
Rac1
Test/DEV
DB
ZFS 7420
Rac2
17
17
NFS Shares
Key Notes (762374.1)
•ZFS Shares mounted as NFS on Exadata DB nodes
•DNFS configured for each share used by DB
•To configure DNFS, run the following for each Oracle Home
•make -f ins_rdbms.mk dnfs_on
•Restart each instance from the Oracle Home
•Add NFS shares to /etc/oranfstab for global access
server: heada
path: 192.168.20.10
export: /export/db_logs mount: /u01/app/oracle/db/reco
server: headb
path: 192.168.20.11
export: /export/db_data mount: /u01/app/oracle/db/data
•Once DNFS is enabled, the following message will be displayed in alert log file
•Oracle instance running with ODM: Oracle Direct NFS ODM Library Version 2.0
•Following query can be used to check if DNFS is used
•select * from v$dnfs_servers;
18
18
NFS Mount Options (RAC)
Binaries
Mount options for Oracle
Datafiles
Mount options for CRS
Voting Disk and OCR
Operating System
Mount options for
Sun Solaris
rw,bg,hard,nointr,rsize=32768, rw,bg,hard,nointr,rsize=32768, rw,bg,hard,nointr,rsize=32768,
wsize=32768,proto=tcp,noac,
wsize=32768,proto=tcp,noac,
wsize=32768,proto=tcp,vers=3,
forcedirectio,
vers=3,suid
noac,forcedirectio
vers=3,suid
AIX (5L)
rw,bg,hard,nointr,rsize=32768, cio,rw,bg,hard,nointr,rsize=3276 cio,rw,bg,hard,intr,rsize=32768,
8,
wsize=32768,proto=tcp,
wsize=32768,tcp,noac,
wsize=32768,proto=tcp,noac,
vers=3,timeo=600
vers=3,timeo=600
vers=3,timeo=600
HPUX 11.23
rw,bg,vers=3,proto=tcp,noac,
hard,nointr,timeo=600,
rsize=32768,wsize=32768,suid
rw,bg,vers=3,proto=tcp,noac,
forcedirectio,hard,nointr,timeo=
600,
rsize=32768,wsize=32768,suid
Windows
Not Supported
Not Supported
rw,bg,vers=3,proto=tcp,noac,
forcedirectio,hard,nointr,timeo=
600
,rsize=32768,wsize=32768,suid
Linux x86
Not Supported
rw,bg,hard,nointr,rsize=32768,
rw,bg,hard,nointr,rsize=32768, rw,bg,hard,nointr,rsize=32768,
wsize=32768,tcp,noac,actimeo=
wsize=32768,tcp, vers=3,
wsize=32768,tcp,actimeo=0,
0,
timeo=600, actimeo=0
vers=3,timeo=600
vers=3,timeo=600
Linux x86-64
rw,bg,hard,nointr,rsize=32768, rw,bg,hard,nointr,rsize=32768, rw,bg,hard,nointr,rsize=32768,
wsize=32768,tcp,vers=3,
wsize=32768,tcp,actimeo=0,
wsize=32768,tcp,noac,vers=3,
timeo=600, actimeo=0
vers=3,timeo=600
timeo=600,actimeo=0
Linux - Itanium
rw,bg,hard,nointr,rsize=32768, rw,bg,hard,nointr,rsize=32768, rw,bg,hard,nointr,rsize=32768,
wsize=32768,tcp,vers=3,
wsize=32768,tcp,actimeo=0,
wsize=32768,tcp,noac,vers=3,
timeo=600, actimeo=0
vers=3,timeo=600
timeo=600,actimeo=0
MOSS: 359515.1
19
19
DB Refresh
1. Suspend managed recovery of standby DB
Primary DB
SQL> alter system set log_archive_dest_state_2=DEFER sid=’*’;
DR DB
SQL> alter database recover managed standby database cancel;
2. Take snapshot of the of the filesystem
3. Resume managed recovery of standby DB
Primary DB
SQL> alter system set log_archive_dest_state_2=ENABLE sid=’*’;
DR DB
SQL > alter database recover managed standby database using current logfile disconnect;
4. Create clone of snapshot
20
20
DB Refresh
5. Mount the filesystems on both the test RAC servers
6. Create a softlink to refer old location of standby redo logs
Update db_file_name_convert and log_file_name_convert
7. Startup DB in mount state
8. Drop the standby redo logs
9. Activate the database
SQL> alter database activate standby database ;
SQL> shutdown immediate ;
SQL> startup
10. Use NID to rename the database.
21
21
DB Refresh
22
22
Technical Issues
23
23
Issues: Exadata
Exadata IPoIB protocol died
•The cause is still not known fully
•This happened when ZFS was hooked to Exadata via IB cable
Workaround
•Oracle support has confirmed that ZFS can be hooked to
Exadata via IB
•Transmit queue length to be increased to 1000 for the IB
interfaces
#ifconfig ib0 txqueuelen 1000
#ifconfig ib1 txqueuelen 1000
24
24
Issues: ZFS NAS Appliance
Memory leak at the server heads
•Cause known to be with SNMP service
•Firmware to be patched until Oracle releases it
•Workaround is to disable SNMP service
HCC on ZFS needs SNMP service to be enabled
25
25
Q&A
26
26