Académique Documents
Professionnel Documents
Culture Documents
After carefully testing several data warehouse platforms, we chose the Oracle Database Machine. Oracle Exadata was able to speed up one of our critical processes from days to minutes..
Brian Camp, Sr. VP of Infrastructure Services
Agenda
Overview Exadata Based Product Offerings Exadata Architecture and Features Best Data Warehousing Machine Best OLTP Machine Best Consolidation Machine
The Products
Exadata Storage Server & Database Machine
Exadata Storage Server
Storage Product Optimized for Oracle Database Extreme I/O and SQL Processing performance Combination of hardware and software
Storage Bottlenecks
Data Bandwidth limits severely restrict performance for data warehousing Random I/O bottlenecks limit performance of OLTP applications
Enterprise Technology Center
Oracle solves random I/O bottlenecks using Exadata Smart Flash Cache
Increase random I/Os by factor of 20X
Enterprise Technology Center
Sun Oracle Database Machine delivers the first and only complete grid architecture for all data management needs
InfiniBand Network
40 Gb/sec fault-tolerant unified server and storage network
Extreme Performance
InfiniBand Network
880 Gb/sec aggregate throughput
Enterprise Technology Center
Quarter Rack
Half Rack
Full Rack
Scalable
Scales to 8 rack database machine by just adding wires
Multi-petabyte databases
Enterprise Technology Center
Months to Days
Hardware by
Software by
Pre-installed Software: InfiniBand QDR (40Gb/s) dual port card Oracle Exadata Storage Server Software 4 x 96GB Sun Flash PCIe Cards Oracle Enterprise Linux Drivers
1 Admin Cisco Ethernet switch Keyboard, Video, Mouse (KVM) hardware Redundant Power Distributions Units (PDUs) Single Point of Support from Oracle 3 year, 24 x 7, 4 Hr On-site response
1 Admin Cisco Ethernet switch Keyboard, Video, Mouse (KVM) hardware Redundant PDUs Single Point of Support from Oracle 3 year, 24 x 7, 4 Hr On-site response
2 Sun Fire X4170 Oracle Database servers 3 Exadata Storage Servers (All SAS or all SATA) 2 Sun Datacenter InfiniBand Switch 36 36-port Managed QDR (40Gb/s) InfiniBand switch 1 Admin Cisco Ethernet switch Keyboard, Video, Mouse (KVM) hardware Redundant PDUs Single Point of Support from Oracle 3 year, 24 x 7, 4 Hr On-site response
Customer supplied x86 64-bit Linux Database Servers Hardware installation more complex No single point of support for entire deployment
24 TB
384 GB
72 TB
1.1 TB 6 TB 21 TB
168 TB
2.6 TB 14 TB 50 TB
336 TB
5.3 TB 28 TB 100 TB
SAS SATA
2 TB 7 TB
1 Raw capacity calculated using 1 GB = 1000 x 1000 x 1000 bytes and 1 TB = 1000 x 1000 x 1000 x 1000 bytes. 2 - User Data: Actual space for end-user data, computed after single mirroring (ASM normal redundancy) and after allowing space for database structures such as temp, logs, undo, and indexes. Actual user data capacity varies by application. User Data capacity calculated using 1 TB = 1024 * 1024 * 1024 * 1024 bytes.
SAS SATA
1 Bandwidth is peak physical disk scan bandwidth, assuming no compression. 2 - Max User Data Bandwidth assumes scanned data is compressed by factor of 10 and is on Flash. 3 IOPs Based on IO requests of size 8K 4 - Actual performance will vary by application.
Enterprise Technology Center
Exadata Configuration
Single-Instance Database RAC Database
InfiniBand Switch/Network
Exadata Cell
Exadata Cell
Exadata Cell
Each Exadata Cell is a self-contained server which houses disk storage and runs the Exadata software Oracle Databases are deployed across multiple Exadata Cells Oracle Database enhanced to work in cooperation with Exadata Storage Server No practical limit to number of Cells that can be in the grid
Enterprise Technology Center
Exadata Architecture
Single-Instance Database DB Server DB Instance DBRM ASM RAC Database DB Server DB Server DB Instance DBRM ASM DB Instance DBRM ASM
Enterprise Manager
InfiniBand Switch/Network
OEL
CELLSRV
OEL
CELLSRV
CELLSRV
I/Os Issued
Smart Scan Example: Telco wants to identify customers that spend more than $200 on a single phone call Rows Returned The information about these premium customers occupies 2MB in a 1 terabyte table With traditional storage, all DB Host reduces database intelligence resides terabyte of data to 1000 in the database hosts customer names that are returned to client Very large percentage of data returned from storage is discarded by database servers Discarded data consumes valuable resources, and I/Os Executed: impacts the performance of 1 terabyte of data other workloads
returned to hosts
customer_name and required rows where amount>200 are are returned to hosts
Smart Scan identifies rows and columns within terabyte table that match request
Enterprise Technology Center
11.2
All data mining scoring functions offloaded to Exadata Up to 10x performance gains Reduced CPU utilization on Database Server
11.2
Index
Exadata Storage Indexes maintain summary information about table data in memory
Store MIN and MAX values of columns Typically one index entry for every MB of disk
Min B = 1 Max B =5 Eliminates disk I/Os if MIN and MAX can never match where clause of a query Min B = 3 Completely automatic and transparent Max B =8
8
3
Select * from Table where B<2 - Only first set of rows can match
Enterprise Technology Center
11.2
50X
Up To
Archive Compression
Optimized for Space 15x average storage savings
Up to 50x on some data
50 45 40 35 30 25 20 15 10 5 0
43 29 16 19 19 19 20 21
10
10
10
11
Columnar compression ratios Query = 14.6X Archive = 22.6X Vary by application and table
Enterprise Technology Center
Ideal Solution
Keep most data on disk for low cost Transparently move hot data to flash Use flash cards instead of flash disks to avoid disk controller limitations Flash cards in Exadata storage
11.2
4 x 96 GB Flash Cards
Database Server
InfiniBand Switch/Network
Exadata Cell
Exadata Cell
Exadata Cell
Ensure different users and tasks within a database are allocated the correct relative amount of I/O bandwidth
Database A: Reporting: 60% of I/O resources ETL: 40% of I/O resources Database B: Interactive: 30% of I/O resources Batch: 70% of I/O resources
InfiniBand Switch/Network
Exadata Cell
Exadata Cell
Exadata Cell
Single-Instance Database
RAC Database
InfiniBand Switch/Network
Exadata Cell
Exadata Cell
Exadata Cell
ASM disk
Grid Disk n
Sys Area
ASM disk
Exadata Cell
Exadata Cell
Hot Cold
Hot Cold
Hot Cold
Hot
Cold
ASM hot and cold Disk Group Two ASM disk groups created across the two sets of grid disks ASM striping evenly distributes I/O across the disk groups ASM mirroring is used to protect against disk failures ASM failure groups are used to protect against cell failures
Enterprise Technology Center
11.2
Grid Disk 1
Hot Data, Cold Data
Grid disks are optionally split and interleaved to place frequently accessed data in all grid disks on higher performing outer tracks All applications benefit from higher performance outer tracks of disks
Grid Disk 2
Hot Data, Cold Data
Enterprise Technology Center
Comprehensive CLI
Local Exadata Storage cell management Distributed shell utility to execute CLI across multiple cells
Data Guard provides disaster protection and data corruption protection Flashback provides human error protection Recovery Manager (RMAN) provides backup to disk
Database Server
Exadata
Non-Exadata
Online migration if currently using ASM and ASM redundancy Migration can be done using RMAN or Data Guard
Enterprise Technology Center
Online Migration
OLAP
ETL
OLAP, Statistics, Spatial, Data Mining, Real-time transactional ETL, Efficient point queries
Data Mining
New
Enterprise Technology Center
11.2
A single database machine has over 400GB of memory usable for caching Database release 11.2 introduces parallel query processing on memory cached data
Harnesses memory capacity of entire database cluster for queries Foundation for world record 1TB TPC-H
315,842
Exadata Hybrid Columnar Compression enables multi-terabyte tables or partitions to be cached in memory
ParAccel Exasol Oracle
Source: Transaction Processing Council, as of 9/14/2009: Oracle on HP Bladesystem c-Class 128P RAC, 1,166,976 QphH@1000GB, $5.42/QphH@1000GB, available 12/1/09. Exasol on PRIMERGY RX300 S4, 1,018,321 QphH@1000GB, $1.18/QphH@1000GB, available 08/01/08. ParAccel on SunFire X4100 315,842 QphH@1000GB, $4.57 /QphH@1000GB, available 10/29/07.
Benefits Multiply
1 TB with compression
5 to 7 GB/sec file system I/O throughput Load into database using External Tables
ETL
Exadata Flash
Solves the Random I/O Bottleneck
11.2
Sun Oracle Database Machine has 5+ TB of flash storage Exadata Smart Cache caches hot data Database Machine achieves: 20x more random I/Os Over 1 million per second 2x faster sequential query I/O 50 GB/sec 10x better I/O response time Sub-millisecond Greatly Reduced Cost 10x fewer disks for IOPS Lower Power
WAN
Protection from
Server Failures Storage Failures Network Failures Site Failures
Configuration Management
Access Control
Audit Vault
Total Recall
Database Vault
Encryption and Masking
Label Security
Advanced Security
Enterprise Technology Center
Secure Backup
Data Masking
Consolidation
ERP CRM ERP
CRM
HR
Warehouse
Warehouse
Data Mart
Data Mart HR
The Sun Oracle Database Machine handles any combination of workloads with extreme performance
And predictable response times
Consolidate Storage
Exadata and ASM allow all storage servers to be shared across databases
ERP CRM Warehouse Data Mart HR
Shared Configuration
Advanced data striping spreads every database across all storage servers Eliminates hot-spots and captive unused space Full storage grid performance available to all databases Database or cluster level storage security
Predictable Performance
Exadata I/O resource manager prioritizes I/Os to ensure predictable performance At user, job, application, or database level No need for isolated storage islands
Consolidate Servers
Many databases can run on Database Machine servers Shared Configuration
CRM Applications connect to a database service that runs on one or more database nodes Services can grow, shrink, & move dynamically Large databases can span nodes using RAC Multiple small databases can run on a single node
ERP
Warehouse
HR
Data Mart
Predictable performance
Instance caging provides predictable CPU resources when multiple databases run on the same node Restricts a database to subset of processors
Resources
Oracle.com:
http://www.oracle.com/exadata
http://www.oracle.com/technology/products/bi/db/exadata/pdf/migr ation-to-exadata-whitepaper.pdf
Marketing US
Up to 10x faster for queries than Sybase IQ
A query using where clauses for de-duplication ran 10x faster on a Database Machine A query using Soundex comparisons ran 10x faster on a Database Machine
Telecommunications APAC
Up to 11x faster than customer set targets
Database Machine purchased by one of the largest Teradata customers in APAC Target time for a set of CDR queries is 3 hours. Database Machine Half Rack ran workload in 47 minutes, 3.8x faster Data load rates of ~3 TB/hour
Retail EMEA
80x faster than Oracle-on-IBM for cross-selling analysis
A complex PL/SQL analysis taking 12 hours or more ran in 12 minutes on a Half Rack Database Machine, 80x faster Index builds were on average 16x faster
Telecommunications EMEA
Full rack DB Machine is 5x faster than Oracle on Sun M-series
One of the most run queries is 19x faster on a Database Machine
An ETL batch job that required well over 3 hours runs in 50 minutes, 3.6x faster
Retail APAC
A full rack DB Machine delivers on average 3x better performance than Netezzas 2 rack NPS10200
A full rack database machine outperforms a Netezza NPS10200 by 9x on analyzing 3TB for cross selling analysis Database machine loads data on average 25% faster
Transportation EMEA
On average 6x faster than an 8 node Teradata 5450*
50 queries on 3TB that run 8 hours on Teradata, run just over 1 hour on a Database Machine, 6x faster Scale out test query on 6x data volume (18TB) did not complete on Teradatas 5550, query ran in 3 minutes on a full rack database machine
* Two Teradata systems were involved, a 5450 and a 5550 (see notes for details)
Telecommunications APAC
Up to 80x faster than Greenplum on a Sun X4500
177M rows in daily SMS query runs in 1 minute, compared to 16 minutes on Greenplum, 16x faster Capabilities to run real-time inserts of 37M rows with individual commits in 27 minutes. Impossible on Greenplum.
Retail - US
Exadata improved IBM DB performance by more than 20X
The move to Exadata allowed daily PoS analysis to be completed by 7:00 am, instead of 12 Noon Projected to result in $25M annual revenue growth and bottom line improvement of $1.8M Also projected to reduce inventory investment capital by $4.5M CIO quote Exadata is as advertised, and then some
Retail - US
Exadata improved query performance by approximately 170X
The move to Exadata will allow this retailer to scale Point-of-Sale data from 1TB to 100TB Gathering daily instead of weekly PoS data will allow better inventory replenishment and reduction of out-of-stock occurrences Exadata completes queries that currently take 6 hours in a little over 2 minutes
Pharmaceuticals US
Exadata beat the current client timings by 218X
Exadata beat current client timings by 218X, 19.8 minutes to current 72 hours Client target was 24 hours, which Exadata handily beat by 73X This was executed with 50GB, 1.6B rows across 17 tables
Insurance US
Multiple queries were executed, up to 2.6X faster than Teradata
In this competive benchmark, Exadata beat both Netezza and Teradata This was executed with 5.0TB of data, with a full rack, and SATA drives
Telecommunications - US
Exadata was 4.5X faster than in-house for total batch cycle, savings of 33 hours
Exadata outperformed in-house benchmarks in all areas
initial data load, by 2.3X daily run, by 12X large DB query, 18.7X overall, 4.5X (9.5 hours to 42.5 hours)
Telecommunications - US
Average speed-up due to Exadata was 186X faster
Exadata executed one test query in 10 seconds
Healthcare - US
On average, Exadata was 7.9X faster for a national healthcare research institution
26 "Reverse POV" queries were run
Financial - US
Exadata ranged from 14X to 752X faster than client for query sets
One query within a set that ran in 4 hours and 40 minutes, was run by Exadata in 17 seconds, 988X faster Largest single table created and loaded was 330GB
Federal Financial - US
Exadata ranged from 15X to 186X faster than client
Application was a data warehouse application
Exadata completed one particular query in 49 seconds, the customer completed the same query in 3.5 hours
Speed-up of 256X
Healthcare - US
Exadata 8X faster than IBM P570
For one query, Exadata was 26X faster than IBM Over 26 queries, Exadata averaged 8X faster than IBM All 26 queries faster on Exadata POV was a Reverse POV, with 200GB
90
80 70 60 50 40 30 20 10 0
78
DB Machine
IBM P570
Entertainment (V2) US
Exadata was 74X faster than current environment
(combination of Oracle and SQL Server)
One query that could not finish in 5 hours, Exadata completed in 52 seconds Exadata queries were 29x faster than SQL Server One query currently takes 81 minutes, Exadata completed in 19 seconds (256x faster) Current environment is Oracle10g and SQL Server 2005 POV simulated 25 concurrent users, on Exadata V2
Retail (V2) US
Exadata V2 up to 13.5X faster than Teradata
A series of queries that Teradata completed in 4 hours, Exadata completed in 17.6 minutes, Exadata 13.5X faster Exadata outperformed Teradata in all three benchmark tests
o 49 sequential queries 3.0X faster o 695 concurrent queries 1.6X faster o 5 surprise ad hoc queries 13.5X faster
238.0
17.6
0.0
DB Machine
Teradata
7000
6000
6000 5000
4000
4000 3000 2000 1000 0
DB Machine
IBM Pseries