Supermicro and AMD Achieve Outstanding Linear Workload Scaling for Oracle Database 19C

1

Solution Brief

SOLUTION BRIEF

SUPERMICRO AND AMD ACHIEVE

OUTSTANDING LINEAR WORKLOAD SCALING

FOR ORACLE DATABASE 19C

AMD Powered Supermicro A+ Servers Accelerate Oracle Database 19c Performance

Executive Summary

Supermicro, working closely in partnership with AMD, has been a part of the AMD EPYC™ processor journey since the first

launch of AMD's “Zen” microarchitecture in 2017. Supermicro was one of the first server vendors to bring the 1

st

Gen AMD™

EPYC -processors to market with our H11 platforms. In 2019, Supermicro launched its first family of H12 generation AMD

processor-powered Supermicro A+ servers optimized to deliver a new level of integration and superior performance for

modern datacenters with the AMD EPYC™ 7002 Series Processors. The new A+ servers, powered by the 2nd Gen AMD EPYC

7002 series processors, deliver up to two times the performance with up to double the core count compared to 1st Gen AMD

EPYC processor-based systems. With Supermicro’s next-generation architecture, the A+ servers take full advantage of 2x data

throughput of PCI-E 4.0 lanes that the new AMD processors provide, including the full spectrum of the latest components

storage, networking, accelerators like GPUs.

TABLE OF CONTENTS

Executive Summary

............................................

1

Benchmarking with Oracle Database 19C

......................

2

Hardware Configuration and Set Up

...........................

3

Hardware Linear Workload Scaling

............................

4

SUPERMICRO

Supermicro (Nasdaq: SMCI), the leading innovator in high-

performance, high-efficiency server and storage technology is a

premier provider of advanced server Building Block Solutions®

for Enterprise Data Center, Cloud Computing, Artificial

Intelligence, and Edge Computing Systems worldwide.

Supermicro is committed to protecting the environment

through its “We Keep IT Green®” initiative and provides

customers with the most energy-efficient, environmentally-

friendly solutions available on the market.

2

Solution Brief

Supermicro’s expertise and experience in optimizing systems for the AMD EPYC™ architecture has led to 27 world record

system performance benchmarks at launch, in August 2019, for 2nd Gen AMD EPYC 7002 series processors. World-record

performance benchmarks were set on Supermicro’s new H12 A+ Servers for TPCx-IoT and TPC-DS categories. For more

detailed information on AMD EPYC World Records, please see the URL (1) in the Footnotes. For TPCx-IoT, the performance of

472,200.88 IoTps was established on Supermicro's H12 TwinPro™ 2U 4-node server where faster IoT gateway data analytics

are critical for the coming explosion of IoT device numbers. This server delivers the highest performance and lowest

performance per dollar for a 10TB database with 64% higher QphDS throughput (Composite Query per Hour Metric) and cost

savings per QphDS over the previous world record holders, see the AMD EPYC 7002 Series Processors URL(2) in Footnotes.

Supermicro offers an industry-leading portfolio of AMD EPYC based systems based on Server Building Block Solutions®. From

single-socket mainstream and WIO servers to high-end Ultra server systems and multi-node systems, including BigTwin® and

TwinPro®, Supermicro enables customers to build application-optimized solutions with a wide range of configuration

possibilities.

AMD EPYC™ 7002 Series Processors Brings Flexibility, Performance, and Security Features

AMD EPYC 7002 Series processors offer up to 64 x86 cores per socket for high-density requirements and can dramatically

increase your system performance while running a number of concurrent and complex applications. With higher frequencies

than the other offerings, the AMD EPYC 7Fx2 processors are optimized for frequency sensitive and single-threaded workloads.

This efficiency and speed can help you to achieve significant savings in the total cost of ownership (TCO). Based on AMD

Infinity Guard, the new 2nd Gen AMD EPYC 7002 series processors also feature advanced security, including Secure

Memory Encryption and Secure Encrypted Virtualization.

AMD EPYC 7Fx2 processors bring high frequencies and very high ratios of cache per core to the 2nd Gen AMD EPYC family of

processors along with the large memory capacity, extreme memory bandwidth, and massive I/O, to deliver exceptional

performance.

The tested configuration is shown below in Table 1. The AMD EPYC 7Fx2 processors offer high frequency that is specifically

designed to optimize per-core performance for frequency sensitive workloads and core-based software licensing models.

AMD EPYC 7Fx2 Processors Demonstrate Superior Performance on Transactional Processing Workloads with

Oracle Database 19c

Oracle® Database delivers leading-edge innovations in relational database management systems (RDBMS) for on-premises,

cloud, and hybrid workloads with exceptional performance and ease-of-use. AMD’s internal test results show that AMD EPYC

7Fx2 processor-based systems deliver high performance and outstanding scaling for Online Transaction Processing (OLTP)

performance with Oracle Database 19c.

Supermicro AS -1014S-WTRT server, powered by AMD EPYC 7F72

For this benchmark with Oracle Database 19c, we are using the Supermicro AS -1014S-WTRT server powered by AMD EPYC

7F52 and AMD EPYC 7F72. This system is a single socket, as shown in the table below. With AMD EPYC’s core density, it is cost-

effective and ideal solution for database processing and enterprise application workloads.

3

Solution Brief

Server

Form Factor

System

Memory

Drive Bays

Network

Controllers

AS-1014S-

WTRT

1U Rackmount

8 x DDR4 slots up

to

3200MHZ

4 x hot swap

3.5” or 2.5”

SATA3 or NVMe

drives

Dual Broadcom 10G

Base -TLAN ports

Table 1: Benchmarks CPU Model

Supermicro’s AS -1014S-WTRT server’s design philosophy was to provide a high density, single-socket server in a compact

form factor (1U, 25.6”deep) that is powerful and robust enough to handle the most demanding enterprise applications while

also maintaining cost efficiency. The AS -1014S-WTRT is a single socket system, in which customers may choose among AMD’s

8, 16, 24, 32, and 64 cores, P variant CPU SKUs, which offer unique pricing discounts over dual socket capable AMD CPUs.

We specifically selected this single-socket system for benchmarking with Oracle Database 19c because it offers a unique and

strong value proposition, capable of predictable workload performance scaling. With AMD’s 2

nd

generation cores, what used to

require dual or multi-socket systems, can now be consolidated into a single scale-up server.

AS -1014S-WTRT – Oracle Database 19c

AMD EPYC 7Fx2 Processors Demonstrate Superior Performance on Transactional Processing

Workloads with Oracle Database 19c

Oracle® Database delivers leading-edge innovations in relational database management systems (RDBMS) for on-premises,

cloud, and hybrid workloads with exceptional performance and ease-of-use. AMD’s internal test results show that AMD EPYC™

7Fx2 processor-based systems deliver high performance and exceptional scaling for Online Transaction Processing (OLTP)

performance with Oracle Database 19c.

4

Solution Brief

Server

Base

Frequency

Boost

Frequency

(up to)*

Core

Processor

Memory

Channels

Maximum

Memory/Socket(DDR4-

3200)

PCIe®

Gen4

Lanes /

System

AMD

EPYC

7F52

3.5 GHz

3.9 GHz

16

8

4 TB

128

AMD

EPYC

7F72

3.2 GHz

3.7 GHz

24

8

4 TB

128

Table 2: Figure CPU's deployed

Benchmarking with Oracle Database 19c and AMD EPYC 7Fx2 Processors

These CPUs are optimized for higher frequency as in Table 2; they excel at single-threaded workloads where core count

limitation is preferred.

This workload models an order fulfillment system where the database receives requests for data, adds new data, and

makes multiple changes to the data from a large number of users. The results show sustained transaction throughput and

predictable scaling across 16-core and 24-core processors allowing one to right-size the compute power to the

application needs and help lower total cost of ownership by only paying for cores needed to optimize core-based

software licensing model costs.

Table 3: TPC - C Benchmarks

5

Solution Brief

This OLTP Workload, used for this benchmark, is derived from the TPC-C Benchmark, and as such, is not comparable to

published TPC-C Benchmark results, as the OLTP workload results do not comply with the TPC-C Benchmarks in the above

Table 3.

Hardware Configuration and Setup

6

Solution Brief

Supermicro AS -1014S-WTRT server is a 1U Rackmount Server supporting Single Socket with AMD EPYC™ 7002 Series

Processors. Two different CPU models were used in this benchmark study to estimate a scaling efficiency due to the different

number of cores contained.

The same tests have been applied to the same server configuration, with the only difference being the CPU, per the two

options listed below in Table 2.

SuperMicro engineers executed these benchmarks with the Industry Standard HammerDB Benchmark Tool on a Supermicro

AS -1014S-WTRT Server using 16-core AMD EPYC 7F52 and 24-core AMD EPYC 7F72 processors as displayed in Table 2.

The same configuration used for the TPC-C Benchmark systems has been applied to both systems.

The creation scripts for the database schema and execution scripts can be found in this document Appendix section.

The tested Oracle Database version was 19.3.0.0.0.

An Oracle Restart Instance was configured for the tests.

o 4 Datacenter Class 1.8T NVMe SSDs were used for Oracle storage.

o The 4 NVMe disks had been added to an Oracle ASM instance for use by Oracle Database.

▪ 2 x 1.8 TB = DATA Disk Group

Firmware/BIOS configuration

The Test configuration had been designed to get the most efficiency out of the Memory channels and interleaving on all 8

DIMMs, as shown in Figure 1: Firmware / Bios configuration.

• Possible NUMA Nodes / Socket configuration was set as NPS1

• SMT was enabled

7

Solution Brief

Figure 1: Firmware / Bios configuration

Hardware Linear Workload Scaling

A crucial step when building your solution is to calculate the total cost of ownership (TCO). Oracle Processor Licensing is

calculated by multiplying the total number of cores on each processor by a licensing factor defined by the Oracle Processor

Core Factor Table. In addition to the Core license, Oracle typically charges an additional 20% for service & maintenance. With

this license model, customers can reduce license costs by selecting low core count CPUs with the highest frequencies, such as

the AMD EPYC 7F52 and 7F72.

Oracle Database Standard Edition licensing is based on a per-socket license model. For multi-CPU servers, such as dual and

quad-socket systems, Oracle Database Standard Edition requires an individual license for each socket in the system. This

makes the Supermicro AS-1014S-WTRT an ideal cost-optimized solution for small to mid-size Oracle Database deployments

with linear CPU scaling.

8

Solution Brief

APPENDIX A – Oracle Linux Configuration

The recommended way to configure Oracle Linux for Oracle Database is by installing the Oracle Preinstallation RPM.

The procedure can be found in Oracle’s “Database Installation Guide for Linux.”, see the following regarding Oracle Database

19c.

• https://docs.oracle.com/en/database/oracle/oracle-database/19/ladbi/overview-of-oracle-linux-

configuration-with-oracle-rpms.html#GUID-693599D4-BD32-4E6A-9689-FA7D1CD75653

• Install Oracle Linux 7.6 with UEK 4 (4.14.35.1902.0.18)

• Register your Linux distribution through Oracle’s Unbreakable Linux Network (ULN) and download and configure the

yum repository for your system using the Oracle Linux yum server for your Oracle Linux release.

• Install the Oracle Preinstallation RPM with the RPM for your Oracle Grid Infrastructure and Oracle Database releases

and update your Linux release.

•

Make sure that after the Oracle binaries are extracted to ORACLE_HOME and after executing root.sh, you have the

following entries in your/etc/sysctl.conf file. Please add the vm.nr Hugepages at the end of /etc/syssctl conf based on

your available DRAM, In our test case, the total DRAM was 512 GB, and we allowed the Hugepages size to occupy

around 460GB. Please refer to Oracle Tuning SGA with Hugepages in the link below (Chapter A.7)

•

See https://docs.oracle.com/en/database/oracle/oracle-database/19/unxar/administering-oracle-

database-on-linux.html#GUID-76C03D99-6025-41F2-8BE3-F6DCDB1DCEE0

9

Solution Brief

# cat /etc/sysctl.conf

# sysctl settings are defined through files in

# /usr/lib/sysctl.d/, /run/sysctl.d/, and /etc/sysctl.d/.

#

# Vendors settings live in /usr/lib/sysctl.d/.

# To override a whole file, create a new file with the same in

# /etc/sysctl.d/ and put new settings there. To override

# only specific settings, add a file with a lexically later

# name in /etc/sysctl.d/ and put new settings there.

#

# For more information, see sysctl.conf(5) and sysctl.d(5).

# oracle-database-preinstall-19c setting for fs.file-max is 6815744

fs.file-max = 6815744

# oracle-database-preinstall-19c setting for kernel.sem is '250 32000 100 128'

kernel.sem = 250 32000 100 128

# oracle-database-preinstall-19c setting for kernel.shmmni is 4096

kernel.shmmni = 4096

# oracle-database-preinstall-19c setting for kernel.shmall is 1073741824 on x86_64

kernel.shmall = 173741824

# oracle-database-preinstall-19c setting for kernel.shmmax is 4398046511104 on

x86_64

kernel.shmmax = 4398046511104

# oracle-database-preinstall-19c setting for kernel.panic_on_oops is 1 per Orabug

19212317

kernel.panic_on_oops = 1

# oracle-database-preinstall-19c setting for net.core.rmem_default is 262144

net.core.rmem_default = 262144

# oracle-database-preinstall-19c setting for net.core.rmem_max is 4194304

net.core.rmem_max = 4194304

# oracle-database-preinstall-19c setting for net.core.wmem_default is 262144

net.core.wmem_default = 262144

# oracle-database-preinstall-19c setting for net.core.wmem_max is 1048576

net.core.wmem_max = 1048576

# oracle-database-preinstall-19c setting for net.ipv4.conf.all.rp_filter is 2

net.ipv4.conf.all.rp_filter = 2

# oracle-database-preinstall-19c setting for net.ipv4.conf.default.rp_filter is 2

net.ipv4.conf.default.rp_filter = 2

# oracle-database-preinstall-19c setting for fs.aio-max-nr is 1048576

fs.aio-max-nr = 1048576

# oracle-database-preinstall-19c setting for net.ipv4.ip_local_port_range is 9000 65500

net.ipv4.ip_local_port_range = 9000 65500

vm.nr_hugepages=236000 ( reduce it to memory available ) NR huge pages x 2MB /

1024 = 460 GB (In our test case our DRAM capacity used was 512 GB)

10

Solution Brief

APPENDIX B – Oracle Database 19c Configuration

• Oracle Database is tuned for this Benchmark using the profile parameters below:

$ cat /u01/app/oracle/product/19.0.0/dbhome_1/dbs/initROMETPCC.ora

11

Solution Brief

12

Solution Brief

Preparing Benchmark Database

Benchmark Database Creation

Benchmark database has been created using the SQL Scripts below

Tablespace creation

Temporary Tablespace creation

Create Users and grants

13

Solution Brief

APPENDIX C – Configuration Validation for HammerDB

Preparing Benchmark Client

a. Install Oracle Database 19c client on Oracle Linux

Install Oracle client software on the client machine and configure as per the below instructions

https://www.hammerdb.com/docs/ch04s02.html#d0e699

Install Oracle Database 19c client software on the client machine and configure as per the below instructions

https://www.hammerdb.com/docs/ch01s06.html#d0e382

HammerDB In Linux user’s .bashrc , include the following line:

export LD_LIBRARY_PATH= $ORACLE_HOME/lib:$LD_LIBRARY_PATH

b. Install HammerDB on HammerDB Linux user

The Installation instructions for HammerDB can be found on HammerDB website :

“Section 5 , Installing and Starting HammerDB on Linux”

https://www.hammerdb.com/docs/ch01s05.html

Install HammerDB on the Local Test System (LTS) from: https://www.hammerdb.com/docs/ch01s05.html

c. Benchmark Schema TPCC800 Data Creation

Benchmark database had been created using HammerDB CLI ( command line interface). Commands used for schema

creation are as below

14

Solution Brief

15

Solution Brief

Executing Benchmark runs

Here is a sample script for HammerDB benchmark run with 256 Warehouses and 100 Virtual users.

Conclusion

Oracle Database Standard Edition licensing is based on a per-socket license model. For multi-CPU servers, such as dual and

quad-socket systems, Oracle Database Standard Edition requires an individual license for each socket in the system. This

makes the Supermicro AS-1014S-WTRT an ideal cost-optimized solution for small to mid-size Oracle Database deployments

with linear CPU scaling.

16

Solution Brief

Footnotes

Performance measured with Oracle Database 19c on Oracle Linux 7.6 with UEK 4 (4.14.35.1902.0.18) using HammerDB version

22. Max boost for AMD EPYC processors is the maximum frequency achievable by any single core on the processor under

normal operating conditions for server systems.

1. AMD EPYC™ World Records: : AMD EPYC™ Processor World Records | AMD

2. AMD EPYC™ 7002 series Processors: EPYC-7002-TPC-DS-10TB-World-Record.pdf

*Max Boost reference: Max boost for AMD EPYC processors is the maximum frequency achievable by any single core on the

processor under normal operating conditions for server systems.

About Super Micro Computer, Inc.

Supermicro (Nasdaq: SMCI), the leading innovator in high-performance, high-efficiency server and storage technology is a

premier provider of advanced Server Building Block Solutions® for Enterprise Data Center, Cloud Computing, Artificial

Intelligence, and Edge Computing Systems worldwide. Supermicro is committed to protecting the environment through its

“We Keep IT Green®” initiative and provides customers with the most energy-efficient, environmentally-friendly solutions

available on the market.

Supermicro, Server Building Block Solutions, and We Keep IT Green are trademarks and/or registered trademarks of Super

Micro Computer, Inc.

All other brands, names, and trademarks are the property of their respective owners.

“AMD, the AMD logo, EPYC, and combinations thereof are trademarks of Advanced Micro Devices, Inc.”