• No results found

Is there any alternative to Exadata X5? March 2015

N/A
N/A
Protected

Academic year: 2021

Share "Is there any alternative to Exadata X5? March 2015"

Copied!
26
0
0

Loading.... (view fulltext now)

Full text

(1)

Is there any alternative to Exadata X5?

(2)

1 About Benchware Ltd.

2 Licensing 3 Scalability

4 Exadata Specifics 5 Performance 6 Costs

7 Myths

8 Conclusion

Contents

(3)

1 About Benchware Ltd.

2 Licensing

3 Scalability

4 Exadata Specifics 5 Performance 6 Costs

7 Myths

8 Conclusion

Contents

(4)

 Exadata now supports Oracle VM with trusted partitions

 Much more flexible licensing, e.g. for specific options

 Separation of network zones

 Separation of classes of database, e.g. with different service level

 Benchware recommendation

 Use database virtualization [1] instead of server virtualization

 Use RAC One Node to move database services between server

Oracle VM

CPU: 2 sockets, 36 cores

X5-2 Server System

Virtual Server 1 Virtual Server

2

Virtual Server 3

(5)

 A minimum of 40% of all server cores must be licensed

 Example

 1/8 rack has 18 cores

 Minimum of 4 cores per server must be licensed

 Entry-level Exadata now available

Capacity on Demand

CPU: 2 sockets, 36 physical cores, 18 available

X5-2 Server System Eighth Rack

(6)

1 About Benchware Ltd.

2 Licensing 3 Scalability

4 Exadata Specifics 5 Performance 6 Costs

7 Myths

8 Conclusion

Contents

(7)

 Exadata provides limited scale-up features

 Memory capacity default 256 GByte

 Memory extension to 512 GByte or maximum 768 GByte

Scale-Up versus Scale-Out

CPU: 2 sockets, 36 cores Memory: 256 – 768 GByte

X5 -2 Se rv er S ys tem

CPU: 2 sockets, 36 cores Memory: 256 – 768 GByte

X5 -2 Se rv er S ys tem

(8)

 Some applications need

 Large scale-up server

 More than 2 sockets and 36 cores

 More than 768 GByte memory

 But not certified for RAC

 X4-8 with Oracle VM and Capacity on Demand (CoD)

 Minimum 48 cores per server (out of 120)

 There are some drawbacks with Intel E7 processors . . .

Scale-Up versus Scale-Out

CPU: 8 sockets, 120 cores, minimum 48 cores licensed with CoD Memory: 2 – 6 TByte

X4 -8 Se rv er S ys tem

(9)

 Building Blocks

 Foundation is 1/8 or 1/4 rack

» 2 Exadata Database Servers

» 3 Exadata Storage Servers

 Add Exadata Database Server as needed

 Add Exadata Storage Server as needed

 Even from different Exadata generations!

 Oracle maintains interfaces, patch levels, security, etc.

Workload dependent Elastic Configurations

X5-2 Database Server

X5-2 Storage Server HC

X5-2 Storage Server EF

(10)

1 About Benchware Ltd.

2 Licensing 3 Scalability

4 Exadata Specifics

5 Performance 6 Costs

7 Myths

8 Conclusion

Contents

(11)

 Exadata V2, X2, X3, X4

 HDD with flash cache

 Uses auto-tiering mechanism

 Auto-tiering algorithms work fine in most situations - but not all situations

 Also true for shared storage systems

 Exadata X5

 Replacement for “High Performance”

Storage Server

 Benchware recommendation

 EF for active OLTP data for predictable Oracle performance

 HC for DWH systems or inactive data

 Instead of auto-tiering use Oracle 12 ADO

Flash Storage, Flash Cache and Auto-Tiering

(12)

 Compression

 Hybrid Columnar Compression

» Different options dependent on workload profile

 Part of the Exadata storage software license

» 10’000 USD per hdd disk

» 20’000 USD per flash module

 Encryption

 Advanced Security Option

 License on database server

» 15’000 USD per core

 Some flash storage vendors offer inline deduplication, compression and

encryption for free

Compression and Encryption

Exadata versus competition

(13)

1 About Benchware Ltd.

2 Licensing 3 Scalability

4 Exadata Specifics 5 Performance

6 Costs 7 Myths

8 Conclusion

Contents

(14)

 Extract from Oracle Marketing presentation

Intel Processors

(15)

 CPU performance has two components [2]

 Single-thread performance

 Throughput

#Cores SPECint_ base2006 (speed)

SPECint_ base_rate2006 (throughput)

Oracle CPU calibration sys.aux_stats$

(speed)

X2-2 12 40.8 367 2’795

X3-2 16 54.3 630 1’751

X4-2 24 88.1 806 3’074

X5-2 36 65.2 1’380 2’890

Intel Processors

Other benchmark results

(16)

 PL/SQL performance

 Native compiled SIMPLE_INTEGER

 Arithmetic addition

 SQL performance

 Cached table (no I/O)

 Light OLTP

Intel Processors

#Cores Speed

[Mops]

Throughput [Mops]

X2-2 12 452 5’414 (451)

X3-2 16 423 8’465 (530)

X4-2 24 434 13’300 (554)

X5-2 36 409 21’070 (585)

Benchware benchmark results

#Cores Speed

[tps@servicetime]

Throughput [tps]

X2-2 12 19’160@58µs 270’600 (22’550)

X3-2 16 18’380@48µs 373’400 (23’337)

X4-2 24 18’150@55µs 545’700 (22’738)

X5-2 36 42’900@23µs 1’215’000 (33’750)

(17)

1 About Benchware Ltd.

2 Licensing 3 Scalability

4 Exadata Specifics 5 Performance 6 Costs

7 Myths

8 Conclusion

Contents

(18)

1 About Benchware Ltd.

2 Licensing 3 Scalability

4 Exadata Specifics 5 Performance 6 Costs

7 Myths 8 Conclusion

Contents

(19)

 Exadata is only for large companies  Early Warning Systems GmbH (Zurich)

 Subsidiary of FIFA

 10 employees, 3 people in IT

 Running Exadata X3-2 QR

 External support for patching

 Oracle 12c since January 2015

Myth

(20)

 Exadata means vendor lock-in  All Exadata features can be used without changing application code

 Usage of HCC needs changes in DDL commands when migrating back to conventional platform

 Oracle databases can easily be

migrated between Exadata and non- Exadata platforms

Myth

(21)

1 About Benchware Ltd.

2 Licensing 3 Scalability

4 Exadata Specifics 5 Performance 6 Costs

7 Myths

8 Conclusion

Contents

(22)

 Extreme efficient engineered system

 Uses state-of-the-art hardware technology (processors, memory, flash, protocols, …)

 Highest I/O throughput per database server

 Modular expansion

 Based on building blocks

 Flash storage available

 Workload specific configurations possible

 OLTP -> memory for buffer cache, flash storage

 Data Warehouse -> memory for PGA, flash cache

 Analytics -> memory for column store

Conclusion

(23)

 Flexible licensing

 Virtualization and trusted partitions supported

 Capacity on demand

 High license costs for storage server

 Lowest operational cost for Oracle database platforms

Conclusion

(24)

 Ideal platform for DBaaS infrastructures

 Elastic configuration

 Highest density of databases

 Extreme performance on demand – controlled by comprehensive Resource Manager

 One management console for complete platform – Oracle Enterprise Manager

 Lowest operational cost

Conclusion

(25)

 1 Day Discovery Workshop

 Preliminary for system architects and decision makers

 Neutral overview of actual and important technology trends in the field of Oracle

platforms. A lot of benchmark results are provided for an in depth understanding and facilitate decisions based on facts

 Evaluation criteria and key performance metrics of Oracle platforms for DBaaS infrastructures

More information

(26)

www.benchware.ch

swiss precision in performance measurement

References

Related documents

Extreme Performance by Offloading Data Intensive Processing The Oracle Database and the Exadata Storage Server Software includes a unique technology that offloads data intensive

The High Performance SAS based Exadata Storage Servers provide up to 3.25 TB of uncompressed usable capacity, and up to 1.8 GB/second of raw data bandwidth.. The High Capacity

Ø  A single Exadata storage server is also known as a “cell” Ø  A cell is the building block for the Exadata storage grid Ø  More cells provide greater capacity and

On top of the capacity benefits provided by smart caching, the Exadata Storage Server Software 11.2.3.3 and above, introduces the Exadata Smart Flash Cache Compression feature, which

Keys to Speed and Cost Advantage Exadata Hybrid Columnar Compression Exadata Intelligent Storage Grid Exadata Smart Flash Cache.!. • Data Intensive processing runs in Exadata

Keys to Speed and Cost Advantage Exadata Hybrid Columnar Exadata Smart Flash Cache Exadata Intelligent Storage Compression Grid.. Exadata Intelligent

When added to an Exadata Database Machine X9M-2, a full rack of Exadata Storage Expansion X9M with 19 Extreme Flash storage servers, using the persistent memory accelerator, can

First, the random matrix models are simple enough to provide analytical formulas relating observables like the microscopic level density of the Dirac operator to low energy constants