Sei sulla pagina 1di 16

SAP Software on Windows

Hardware Configurations for Optimal Performance


Diagrams to accompany Note 1612283

Applies to:
SAP NetWeaver and Stand Alone Components running on Windows

Summary
Introduction: SAP Note 1612283 contains a detailed and in depth explanation of SAP Software running on Windows and Intel servers.
These diagrams are designed to illustrate to customers and partners the concepts explained in Note 1612283.
Note 1612283 should be read carefully before reviewing this document.

Author(s):
Cameron Gardiner, Microsoft Corporation
Contact Person for questions and comments on this article:
cgardin@microsoft.com

Reviewer(s):
Karl-Heinz Hochmuth, SAP AG
Bernd Lober, SAP AG
Peter Simon, SAP AG
Jürgen Thomas, Microsoft Corporation

Company: SAP AG
Created on: May 2015
Config 1. Single SAP Application Server on Physical Server or Virtual Machine

2 Socket Intel Haswell CPU 2 Socket Intel Haswell CPU


32 thread / 16 core 32 thread / 16 core
2 NUMA Nodes 2 NUMA Nodes
384GB RAM 384GB RAM
10Gigabit NIC 10Gigabit NIC with SRIOv

Virtual Machine 1
ABAP Instance 1 8 vCPU
PHYS_MEMSIZE = not set – 32GB vRAM
SAP will try to use total physical
memory ABAP Instance 1
PHYS_MEMSIZE = 24000 –
recommend to set value on
VM

Configuration 1 Notes:
A single SAP instance on a powerful modern Intel or AMD server cannot leverage the CPU and memory resources available. Increasing the number of work
processes on a single instance above ~50 will not lead to a proportionate increase in throughput. Installing additional instances is recommended.

Physical Server Configuration (Left)


2 socket Intel E5v3 32 threads / 16 cores / 2 NUMA nodes / 384GB RAM (192GB RAM per NUMA node)
10 Gigabit Network
PHYS_MEMSIZE is not set therefore SAP defaults to the total amount of RAM (384GB)
Approximately 50 work processes. If more work processes are needed it is best to install another instance

Virtual Machine Configuration (Right)


Virtual Machine Configuration can be up to the total number of vCPU and vRAM or Hyper-Visor limit. Virtual Machines configuration must follow the
requirements of SAP Notes such as no overcommit for production systems, adequate network cards and enhanced monitoring
10 Gigabit Network with SRIOv is recommended to improve network performance
PHYS_MEMSIZE should be set on Virtual Machines
Approximately 50 work processes will give best performance. If more work processes are needed it is best to install another instance
Config 2. Multiple SAP Application Server on Physical Server or Virtual Machine

2 Socket AMD 2 Socket AMD


32 core 32 core
2 NUMA Nodes 2 NUMA Nodes
384GB RAM 384GB RAM
10Gigabit NIC 10Gigabit NIC with SRIOv
ABAP Instance 1 Virtual Machine 1
ECC App Server 1 16 vCPU
SAP System Number 00 128GB vRAM
PHYS_MEMSIZE = 32000 Hostname = vmhost1
ABAP Instance 1
ECC App Server 1
ABAP Instance 2
SAP System Number 00
ECC App Server 2
PHYS_MEMSIZE = 32000
SAP System Number 01
PHYS_MEMSIZE = 32000
ABAP Instance 2
ECC App Server 2
ABAP Instance 3 SAP System Number 01
BW App Server 1 PHYS_MEMSIZE = 32000
SAP System Number 02
PHYS_MEMSIZE = 32000 ABAP Instance 3
BW App Server 1
SAP System Number 02
ABAP Instance 4 PHYS_MEMSIZE = 32000
CRM App Server 1
SAP System Number 03
PHYS_MEMSIZE = 32000
Virtual Machine 2
16 vCPU
ABAP Instance 5 128GB vRAM
SRM App Server 1 Hostname = vmhost2
SAP System Number 04
PHYS_MEMSIZE = 32000 ABAP Instance 4
CRM App Server 1
SAP System Number 00
ABAP Instance 6 PHYS_MEMSIZE = 32000
Solman App Server 1
SAP System Number 05
PHYS_MEMSIZE = 32000 ABAP Instance 5
JVM Heap = 2048 SRM App Server 1
SAP System Number 01
Java Instance 7 PHYS_MEMSIZE = 32000
EP App Server 1
SAP System Number 06
JVM Heap = 2048
ABAP Instance 6
Solman App Server 1
ABAP Instance 8 SAP System Number 02
SCM App Server 1 PHYS_MEMSIZE = 32000
SAP System Number 07 JVM Heap = 2048
PHYS_MEMSIZE = 32000
Configuration 2 Notes:

It is recommended to consolidate multiple SAP application servers onto one powerful 2 socket server. Customers have successfully run 5-8 SAP instances on one
2 socket server. Real customer deployments have indicated that the limiting factor is RAM rather than CPU resources

Physical Server Configuration (Left)


2 socket AMD 32 thread / 32 core / 2 NUMA nodes / 384GB RAM (192GB RAM per NUMA node)
10 Gigabit Network
PHYS_MEMSIZE can be set to either a fixed amount of RAM or a percentage value. Example on a server with 8 instances and 384GB RAM 32GB can be allocated
to each ABAP instance. Alternatively the value 12.5% could be used for the profile parameter PHYS_MEMSIZE
Approximately 50 work processes per each ABAP instance. Performance testing has shown many smaller ABAP instances deliver better throughput than a few
large instances. If more work processes are needed it is best to install another instance

Virtual Machine Configuration (Right)


Virtual Machine Configuration can be up to the total number of vCPU and vRAM or hypervisor limit. Virtual Machines configuration must follow the
requirements of SAP Notes such as no overcommit for production systems, adequate network cards and enhanced monitoring
10 Gigabit Network with SRIOv is recommended to improve network performance
PHYS_MEMSIZE should be set on Virtual Machines
Approximately 50 work processes will give best performance. If more work processes are needed it is best to install another instance
Config 3. Multiple/Single SAP Multiple SAP Application Server
on Physical Server or Virtual Machine combined with DBMS
2 Socket Intel Haswell CPU 2 Socket Intel Haswell CPU
32 thread / 16 core 32 thread / 16 core
2 NUMA Nodes 2 NUMA Nodes
384GB RAM 384GB RAM
10Gigabit NIC 10Gigabit NIC with SRIOv

ECC DBMS Instance 1 Virtual Machine 1


DBMS Max Memory =128GB – 16 vCPU
large data cache for DBMS will 192GB vRAM
greatly reduce the physical IO to Hostname = vmhost1
disk
ECC DBMS Instance 1
DBMS Max Memory =128GB – large data
Solman DBMS Instance 2 cache for DBMS will greatly reduce the
physical IO to disk
DBMS Max Memory = 16GB
Solman DBMS Instance 2
DBMS Max Memory = 16GB

CRM DBMS Instance 3


DBMS Max Memory = 32GB CRM DBMS Instance 3
DBMS Max Memory = 32GB

ABAP Instance 1
ECC App Server 1 Virtual Machine 2
16 vCPU
SAP System Number 00 128GB vRAM
Hostname = vmhost2
PHYS_MEMSIZE = 32000

ABAP Instance 1
ECC App Server 1
ABAP Instance 2 SAP System Number 00
ECC App Server 2 PHYS_MEMSIZE = 32000
SAP System Number 01
PHYS_MEMSIZE = 32000 ABAP Instance 2
ECC App Server 2
SAP System Number 01
PHYS_MEMSIZE = 32000
ABAP Instance 3
Solman App Server 1
SAP System Number 02 ABAP Instance 3
Solman App Ser ver 1
PHYS_MEMSIZE = 32000 SAP System Number 02
PHYS_MEMSIZE = 32000

ABAP Instance 4
ABAP Instance 4
CRM App Server 1 CRM App Server 1
SAP System Number 03 SAP System Number 03
PHYS_MEMSIZE = 32000
PHYS_MEMSIZE = 32000
Configuration 3 Notes:

Modern 2 socket servers deliver more than 90,000 SAPS. The total SAPS requirement for some customers is less than 90,000 SAPS. In such cases customers may
consolidate both DBMS and run 3-5 SAP instances on one 2 socket server. Real customer deployments have indicated that the limiting factor is RAM rather than
CPU or IO resources. Intel and AMD CPU are very powerful. Large DBMS data caches reduce physical IO to disk dramatically (in combination with DBMS
compression). IO performance can be scaled up by adding HBA, FCoE or 10 Gigabit NIC (for SMB 3.0 or iSCSI)
Physical Server Configuration (Left)
1. 2 socket Intel E5v3 32 threads / 16 cores / 2 NUMA nodes / 384GB RAM (192GB RAM per NUMA node)
2. 10 Gigabit Network
3. DBMS software should be configured to use a large data cache to reduce physical disk IO. Recommended values are 96GB, 192GB or more. DBMS
software is NUMA aware and can efficiently run across NUMA nodes.
4. DBMS software should be configured to use DB compression to further improve the DBMS cache hit ratio. Some databases such as SQL Server and
Sybase systems compress all tables and indexes by default. Consult the relevant SAP Notes discussing compression for each DBMS.
5. DBMS compression ratios of 40% to 85% have been observed on customer systems. If a 3TB system were compressed to 1TB (66% saving) and the
DBMS cache size set to 384GB there would be very minimal physical disk IO. Analysis of customer systems has shown when the DBMS cache size is more
than 25% of DB size in combination with compression very little (if any) physical disk READ IO will occur.
6. PHYS_MEMSIZE of all SAP ABAP instances, HEAP of all SAP Java instances and the total memory consumed by all DBMS software (data cache and any
other DBMS “system” type caches) should be less than the total physical memory of the server.
7. PHYS_MEMSIZE can be set to either a fixed amount of RAM or a percentage value.

Example: A single physical server with 384GB of RAM has three separate DBMS running with a total of 176GB of RAM allocated to DBMS software (192GB +
16GB + 32GB). The single physical server has four SAP instances each with a PHYS_MEMSIZE of 32GB. The total SAP memory is at least 192GB. In total the
DBMS + SAP memory is 304GB. The remaining 80GB of RAM can be used in case another instance or additional DBMS needs to be added or if the SAP systems
have inefficient custom ABAP code that consumes a lot of extended memory.

Approximately 50 work processes per each ABAP instance. Performance testing has shown many smaller ABAP instances deliver better throughput than a few
large instances. If more work processes are needed it is best to install another instance
Configuration and administration of multiple SAP application servers running on one or more physical or virtual hosts is greatly simplified by placing all profile
parameters into the default profile. This ensures identical configuration of instances, improves the operation of the SAP Logon Load Balancing mechanism and
eliminates complex configuration and administration.

Virtual Machine Configuration (Right)


Virtual Machine Configuration can be up to the total number of vCPU and vRAM or Hyper-Visor limit. Virtual Machines configuration must follow the
requirements of SAP Notes such as no overcommit for production systems, adequate network cards and enhanced monitoring
10 Gigabit Network with SRIOv is recommended to improve network performance
PHYS_MEMSIZE should be set on Virtual Machines
It is not recommended to use Dynamic Memory in conjunction with automatic DBMS memory management features.
Approximately 50 work processes
Config 4. Multi-SID DBMS + ASCS Cluster with SAP Multiple SAP
Application Server on Physical Server or Virtual Machine
Management Station Virtual
2 Socket Intel Haswell CPU 2 Socket Intel Haswell CPU 2 Socket Intel Haswell CPU Machine(s)
32 thread / 16 core 32 thread / 16 core 32 thread / 16 core 1 vCPU
384GB RAM 2 NUMA Nodes 384GB RAM 2 NUMA Nodes 384GB RAM 2 NUMA Nodes 8GB vRAM
Hostname = managevm1
10Gigabit NIC – Server LAN 10Gigabit NIC – Server LAN 10Gigabit NIC – Server LAN
1Gigabit NIC – User LAN 1Gigabit NIC – User LAN 1Gigabit NIC – User LAN
ECC DBMS Instance 1 BW DBMS Instance 4 SCM DBMS Instance 7
DBMS Max Memory = fixed DBMS Max Memory = fixed DBMS Max Memory = fixed
amount amount amount

Solman DBMS Instance 2 EP DBMS Instance 5 BusObj DBMS Instance 8


DBMS Max Memory = fixed DBMS Max Memory = fixed DBMS Max Memory = fixed
amount amount amount

CRM DBMS Instance 3 SRM DBMS Instance 6 BI Java DBMS Instance 9


DBMS Max Memory = fixed DBMS Max Memory = fixed DBMS Max Memory = fixed
amount amount amount

Central Services ECC Central Services BW Central Services SCM


vHostname = vHostname = vHostname =
SAP System Number = 00 SAP System Number = 04 SAP System Number = 07

Central Services Solman Central Services EP Central Services BI Java


vHostname = vHostname = vHostname =
SAP System Number = 01 ABAP/02 Java SAP System Number = 05 SAP System Number = 08

Central Services CRM Central Services SRM


vHostname = vHostname =
SAP System Number = 03 SAP System Number = 06

2 Socket Intel Haswell CPU 2 Socket Intel Haswell CPU 2 Socket Intel Haswell CPU 2 Socket Intel Haswell CPU
32 thread / 16 core 32 thread / 16 core 32 thread / 16 core 32 thread / 16 core
384GB RAM 2 NUMA Nodes 384GB RAM 2 NUMA Nodes 384GB RAM 2 NUMA Nodes 384GB RAM 2 NUMA Nodes
10Gigabit NIC – Server LAN 10Gigabit NIC – Server LAN 10Gigabit NIC – Server LAN 10Gigabit NIC – Server LAN
1Gigabit NIC – User LAN 1Gigabit NIC – User LAN 1Gigabit NIC – User LAN 1Gigabit NIC – User LAN
ABAP Instance 1 ABAP Instance 1 ABAP Instance 1 ABAP Instance 1
ECC App Server 1 ECC App Server 2 ECC App Server 3 ECC App Server 4
SAP System Number 00 SAP System Number 00 SAP System Number 00 SAP System Number 00
PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000

ABAP Instance 2 ABAP Instance 2 ABAP Instance 2 ABAP Instance 2


ECC App Server 5 ECC App Server 6 ECC App Server 7 ECC App Server 8
SAP System Number 01 SAP System Number 01 SAP System Number 01 SAP System Number 01
PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000

ABAP Instance 3 ABAP Instance 3 ABAP Instance 3 ABAP Instance 3


BW App Server 1 BW App Server 2 BW App Server 3 BW App Server 4
SAP System Number 02 SAP System Number 02 SAP System Number 02 SAP System Number 02
PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000

ABAP Instance 4 ABAP Instance 4 ABAP Instance 4 ABAP Instance 4


CRM App Server 1 CRM App Server 2 CRM App Server 3 CRM App Server 4
SAP System Number 03 SAP System Number 03 SAP System Number 03 SAP System Number 03
PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000

ABAP Instance 5 ABAP Instance 5 ABAP Instance 5 ABAP Instance 5


SRM App Server 1 SRM App Server 2 SRM App Server 3 SRM App Server 4
SAP System Number 04 SAP System Number 04 SAP System Number 04 SAP System Number 04
PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000

ABAP Instance 6 ABAP Instance 6 ABAP Instance 6 ABAP Instance 6


Solman App Server 1 Solman App Server 2 BI JAVA App Server 1 BI JAVA App Server 2
SAP System Number 05 SAP System Number 05 SAP System Number 05 SAP System Number 05
PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 JVM Heap = 2048 PHYS_MEMSIZE = 32000
JVM Heap = 2048 JVM Heap = 2048 JVM Heap = 2048

Java Instance 7 Java Instance 7 Java Instance 7 Java Instance 7


EP App Server 1 EP App Server 2 EP App Server 3 EP App Server 4
SAP System Number 06 SAP System Number 06 SAP System Number 06 SAP System Number 06
JVM Heap = 2048 JVM Heap = 2048 JVM Heap = 2048 JVM Heap = 2048

ABAP Instance 8 ABAP Instance 8 ABAP Instance 8 ABAP Instance 8


SCM App Server 1 SCM App Server 2 SCM App Server 3 SCM App Server 4
SAP System Number 07 SAP System Number 07 SAP System Number 07 SAP System Number 07
PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000
Configuration 4 Notes:

2 socket servers running in 3 tier configuration can deliver more than 400,000 SAPS for a single SAP component and more than 1,000,000 SAPS for an entire
production environment.
Approximately 20% or less of the SAPS of a system are consumed by the database and more than 80% can be scaled out on the application server layer.
Therefore a 1:4 ratio between DB and application server resources is normal. Experience has shown capacity can be scaled very effectively by adding application
servers. The bottleneck is typically not the DBMS servers. 10 Gigabit network and a separate dedicated internal SAP application server <-> DBMS network is
mandatory for large configurations. FusionIO or similar “in-server” SSD is strongly recommended for larger systems.

Physical Server Configuration (Left)


DBMS Server
1. 2 socket Intel E5v3 32 threads / 16 cores / 2 NUMA nodes / 384GB RAM (192GB RAM per NUMA node)
2. 10 Gigabit Network for SAP application server to DBMS server communication. A separate 1 Gigabit network for user traffic
3. DBMS software should be configured to use a large data cache to reduce physical disk IO. Recommended values are 96GB, 192GB or more. DBMS
software is NUMA aware and can efficiently run across NUMA nodes.
4. DBMS software should be configured to use DB compression to further improve the DBMS cache hit ratio. Some databases such as SQL Server and
Sybase systems compress all tables and indexes by default. Consult the relevant SAP Notes discussing compression for each DBMS.
5. DBMS compression ratios of 40% to 85% have been observed on customer systems. If a 3TB system were compressed to 1TB (66% saving) and the
DBMS cache size set to 384GB there would be very minimal physical disk IO. Analysis of customer systems has shown when the DBMS cache size is more
than 25% of DB size in combination with compression very little (if any) physical disk READ IO will occur.
6. Analysis of customer systems with very large DBMS cache sizes have shown almost zero physical read IO 6-12 hours after a system has been restarted.
Physical write will occur as most DBMS use write-ahead logging.
7. Some DBMS software requires manual configuration to enable NUMA and Large Memory Page Sizes. Consult the DBMS vendor and/or hardware
vendors best practice document for large scale up deployments

SAP application servers


1. 4 to 12 physical servers 2 socket Intel E5v3 32 threads / 16 cores / 2 NUMA nodes / 384GB RAM (192GB RAM per NUMA node)
2. 10 Gigabit Network
3. PHYS_MEMSIZE of all SAP ABAP instances and HEAP of all SAP Java instances should be less than the total physical memory of the server.
4. PHYS_MEMSIZE can be set to either a fixed amount of RAM or a percentage value.

Example:
A DBMS server with 384GB RAM is running DBMS software in a 3 node cluster. DBMS can run on any one of the 3 nodes.
A single physical server with 384GB of RAM has SAP instances each with a PHYS_MEMSIZE of 32GB plus a Java Instance with a HEAP size of 2GB. The total SAP
memory is at least 224GB. The remaining unused RAM can be used in case another instance needs to be added or if the SAP systems have inefficient custom
ABAP code that consumes a lot of extended memory.
Approximately 50 work processes per each ABAP instance. Performance testing has shown many smaller ABAP instances deliver better throughput than a few
large instances. If more work processes are needed it is best to install another instance
Configuration and administration of multiple SAP application servers running on one or more physical or virtual hosts is greatly simplified by placing all profile
parameters into the default profile. This ensures identical configuration of instances, improves the operation of the SAP Logon Load Balancing mechanism and
eliminates complex configuration and administration.

Virtual Machine Configuration (Right)


Virtual Machine Configuration can be up to the total number of vCPU and vRAM or Hyper-Visor limit. Virtual Machines configuration must follow the
requirements of SAP Notes such as no overcommit for production systems, adequate network cards and enhanced monitoring
10 Gigabit Network with SRIOv is recommended to improve network performance
PHYS_MEMSIZE should be set on Virtual Machines
It is not recommended to use Dynamic Memory in conjunction with automatic DBMS memory management features.

Cluster Configuration:
ASCS in 2 node cluster. The ASCS and ERS consume very little CPU, memory, network and IO. The impact of the ASCS on the sizing of the DBMS cluster is
negligible. Review the SAP Installation Guide for more information on Windows Clustering and Multi-SID clustering

Most DBMS software can run on more than 2 nodes. Example: Windows 2012 R2 & SQL Server 2014 supports 64 nodes

Multi-SID clusters generally require the use of mount points otherwise there will be insufficient drive letters.
Notes 1634991 - How to install an ASCS or SCS instance on more than 2 cluster nodes?
1678705 - Installation scenarios for a standalone ASCS instance
Config 5. Multi-SID DBMS Cluster, Separate SCS Cluster with SAP Multiple SAP
Application Server on Physical Server or Virtual Machine
2 Socket Intel Haswell CPU 2 Socket Intel Haswell CPU 2 Socket Intel Haswell CPU
32 thread / 16 core 32 thread / 16 core 32 thread / 16 core Virtual Machine 1 Virtual Machine 2
384GB RAM 2 NUMA Nodes 384GB RAM 2 NUMA Nodes 384GB RAM 2 NUMA Nodes 8 vCPU
32GB vRAM
8 vCPU
32GB vRAM
10Gigabit NIC – Server LAN 10Gigabit NIC – Server LAN 10Gigabit NIC – Server LAN Hostname = vmhost1 Hostname = vmhost1

1Gigabit NIC – User LAN 1Gigabit NIC – User LAN 1Gigabit NIC – User LAN Central Services ECC Central Services BW
vHostname = vHostname =
ECC DBMS Instance 1 BW DBMS Instance 4 SCM DBMS Instance 7 SAP System Number = 00 SAP System Number = 04
DBMS Max Memory = fixed DBMS Max Memory = fixed DBMS Max Memory = fixed
amount amount amount Central Services Solman Central Services EP
vHostname = vHostname =
SAP System Number = 01 ABAP/02 Java SAP System Number = 05

Solman DBMS Instance 2 EP DBMS Instance 5 BusObj DBMS Instance 8 Central Services CRM Central Services SRM
DBMS Max Memory = fixed DBMS Max Memory = fixed DBMS Max Memory = fixed vHostname = vHostname =
amount amount SAP System Number = 03 SAP System Number = 06
amount

Central Services SCM Central Services BI Java


vHostname = vHostname =
CRM DBMS Instance 3 SRM DBMS Instance 6 BI Java DBMS Instance 9 SAP System Number = 07 SAP System Number = 08
DBMS Max Memory = fixed DBMS Max Memory = fixed DBMS Max Memory = fixed
amount amount amount

Management Station Virtual


2 Socket Intel Haswell CPU 2 Socket Intel Haswell CPU 2 Socket Intel Haswell CPU 2 Socket Intel Haswell CPU Machine(s)
32 thread / 16 core 32 thread / 16 core 32 thread / 16 core 32 thread / 16 core 1 vCPU
384GB RAM 2 NUMA Nodes 384GB RAM 2 NUMA Nodes 384GB RAM 2 NUMA Nodes 384GB RAM 2 NUMA Nodes 8GB vRAM
Hostname = managevm1
10Gigabit NIC – Server LAN 10Gigabit NIC – Server LAN 10Gigabit NIC – Server LAN 10Gigabit NIC – Server LAN
1Gigabit NIC – User LAN 1Gigabit NIC – User LAN 1Gigabit NIC – User LAN 1Gigabit NIC – User LAN
ABAP Instance 1 ABAP Instance 1 ABAP Instance 1 ABAP Instance 1
ECC App Server 1 ECC App Server 2 ECC App Server 3 ECC App Server 4
SAP System Number 00 SAP System Number 00 SAP System Number 00 SAP System Number 00
PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000

ABAP Instance 2 ABAP Instance 2 ABAP Instance 2 ABAP Instance 2


ECC App Server 5 ECC App Server 6 ECC App Server 7 ECC App Server 8
SAP System Number 01 SAP System Number 01 SAP System Number 01 SAP System Number 01
PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000

ABAP Instance 3 ABAP Instance 3 ABAP Instance 3 ABAP Instance 3


BW App Server 1 BW App Server 2 BW App Server 3 BW App Server 4
SAP System Number 02 SAP System Number 02 SAP System Number 02 SAP System Number 02
PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000

ABAP Instance 4 ABAP Instance 4 ABAP Instance 4 ABAP Instance 4


CRM App Server 1 CRM App Server 2 CRM App Server 3 CRM App Server 4
SAP System Number 03 SAP System Number 03 SAP System Number 03 SAP System Number 03
PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000

ABAP Instance 5 ABAP Instance 5 ABAP Instance 5 ABAP Instance 5


SRM App Server 1 SRM App Server 2 SRM App Server 3 SRM App Server 4
SAP System Number 04 SAP System Number 04 SAP System Number 04 SAP System Number 04
PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000

ABAP Instance 6 ABAP Instance 6 ABAP Instance 6 ABAP Instance 6


Solman App Server 1 Solman App Server 2 BI JAVA App Server 1 BI JAVA App Server 2
SAP System Number 05 SAP System Number 05 SAP System Number 05 SAP System Number 05
PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 JVM Heap = 2048 PHYS_MEMSIZE = 32000
JVM Heap = 2048 JVM Heap = 2048 JVM Heap = 2048

Java Instance 7 Java Instance 7 Java Instance 7 Java Instance 7


EP App Server 1 EP App Server 2 EP App Server 3 EP App Server 4
SAP System Number 06 SAP System Number 06 SAP System Number 06 SAP System Number 06
JVM Heap = 2048 JVM Heap = 2048 JVM Heap = 2048 JVM Heap = 2048

ABAP Instance 8 ABAP Instance 8 ABAP Instance 8 ABAP Instance 8


SCM App Server 1 SCM App Server 2 SCM App Server 3 SCM App Server 4
SAP System Number 07 SAP System Number 07 SAP System Number 07 SAP System Number 07
PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000
Configuration 5 Notes:

2 socket servers running in 3 tier configuration can deliver more than 200,000 SAPS for a single SAP component and more than 600,000 SAPS for an entire
production environment.
Approximately 20% or less of the SAPS of a system are consumed by the database and more than 80% can be scaled out on the application server layer.
Therefore a 1:4 ratio between DB and application server resources is normal. Experience has shown capacity can be scaled very effectively by adding application
servers. The bottleneck is typically not the DBMS servers. 10 Gigabit network and a separate dedicated internal SAP application server <-> DBMS network is
mandatory for large configurations.

Physical Server Configuration (Left)


DBMS Server
1. 2 socket Intel E5v3 32 threads / 16 cores / 2 NUMA nodes / 384GB RAM (192GB RAM per NUMA node)
2. 10 Gigabit Network for SAP application server to DBMS server communication. A separate 1 Gigabit network for user traffic
3. DBMS software should be configured to use a large data cache to reduce physical disk IO. Recommended values are 96GB, 192GB or more. DBMS
software is NUMA aware and can efficiently run across NUMA nodes.
4. DBMS software should be configured to use DB compression to further improve the DBMS cache hit ratio. Some databases such as SQL Server and
Sybase systems compress all tables and indexes by default. Consult the relevant SAP Notes discussing compression for each DBMS.
5. DBMS compression ratios of 40% to 85% have been observed on customer systems. If a 3TB system were compressed to 1TB (66% saving) and the
DBMS cache size set to 384GB there would be very minimal physical disk IO. Analysis of customer systems has shown when the DBMS cache size is more
than 25% of DB size in combination with compression very little (if any) physical disk READ IO will occur.
6. Analysis of customer systems with very large DBMS cache sizes have shown almost zero physical read IO 6-12 hours after a system has been restarted.
Physical write will occur as most DBMS use write ahead logging.
7. Some DBMS software requires manual configuration to enable NUMA and Large Memory Page Sizes. Consult the DBMS vendor and/or hardware
vendors best practice document for large scale up deployments.

SAP application servers


1. 4 to 12 physical servers 2 socket Intel E5v3 32 threads / 16 cores / 2 NUMA nodes / 384GB RAM (192GB RAM per NUMA node)
2. 10 Gigabit Network
3. PHYS_MEMSIZE of all SAP ABAP instances and HEAP of all SAP Java instances should be less than the total physical memory of the server.
4. PHYS_MEMSIZE can be set to either a fixed amount of RAM or a percentage value.

Example:
A DBMS server with 384GB RAM is running DBMS software in a 3 node cluster. DBMS can run on any one of the 3 nodes.
A single physical server with 384GB of RAM has SAP instances each with a PHYS_MEMSIZE of 32GB plus a Java Instance with a HEAP size of 2GB. The total SAP
memory is at least 224GB. The remaining unused RAM can be used in case another instance needs to be added or if the SAP systems have inefficient custom
ABAP code that consumes a lot of extended memory.
Approximately 50 work processes per each ABAP instance. Performance testing has shown many smaller ABAP instances deliver better throughput than a few
large instances. If more work processes are needed it is best to install another instance
Configuration and administration of multiple SAP application servers running on one or more physical or virtual hosts is greatly simplified by placing all profile
parameters into the default profile. This ensures identical configuration of instances, improves the operation of the SAP Logon Load Balancing mechanism and
eliminates complex configuration and administration.

Virtual Machine Configuration (Right)

1. Virtual Machine Configuration (Right)can be up to the total number of vCPU and vRAM or Hyper-Visor limit. Virtual Machines configuration must follow
the requirements of SAP Notes such as no overcommit for production systems, adequate network cards and enhanced monitoring
2. 10 Gigabit Network with SRIOv is recommended to improve network performance
3. PHYS_MEMSIZE should be set on Virtual Machines
4. It is not recommended to use Dynamic Memory in conjunction with automatic DBMS memory management features.

Cluster Configuration:
ASCS in 2 node cluster running on two small VMs. The ASCS and ERS consume very little CPU, memory, network and IO.
Most DBMS software can run on more than 2 nodes. Example: Windows 2012 & SQL Server 2012 supports 64 nodes

Multi-SID clusters generally require the use of mount points otherwise there will be insufficient drive letters.
Notes 1634991 - How to install an ASCS or SCS instance on more than 2 cluster nodes?
1678705 - Installation scenarios for a standalone ASCS instance
Config 6. Extra Large Multi-SID DBMS Cluster, Separate SCS Cluster
with SAP Multiple SAP Application Server on Physical Server or Virtual Machine
8 Socket Intel CPU 240 8 Socket Intel CPU 240 8 Socket Intel CPU 240
thread / 120 core thread / 120 core thread / 120 core Virtual Machine 1 Virtual Machine 2 Management Station Virtual
4TB RAM 8 NUMA Nodes 4TB RAM 8 NUMA Nodes 4TB RAM 8 NUMA Nodes 8 vCPU
32GB vRAM
8 vCPU
32GB vRAM Machine(s)
10Gigabit NIC – Server LAN 10Gigabit NIC – Server LAN 10Gigabit NIC – Server LAN Hostname = vmhost1 Hostname = vmhost1 1 vCPU
8GB vRAM
1Gigabit NIC – User LAN 1Gigabit NIC – User LAN 1Gigabit NIC – User LAN Central Services ECC Central Services BW Hostname = managevm1
vHostname = vHostname =
ECC DBMS Instance 1 BW DBMS Instance 4 SCM DBMS Instance 7 SAP System Number = 00 SAP System Number = 04
DBMS Max Memory = fixed DBMS Max Memory = fixed DBMS Max Memory = fixed
amount amount amount Central Services Solman Central Services EP
vHostname = vHostname =
SAP System Number = 01 ABAP/02 Java SAP System Number = 05

Solman DBMS Instance 2 EP DBMS Instance 5 BusObj DBMS Instance 8 Central Services CRM Central Services SRM
DBMS Max Memory = fixed DBMS Max Memory = fixed DBMS Max Memory = fixed vHostname = vHostname =
amount amount SAP System Number = 03 SAP System Number = 06
amount

Central Services SCM Central Services BI Java


vHostname = vHostname =
CRM DBMS Instance 3 SRM DBMS Instance 6 BI Java DBMS Instance 9 SAP System Number = 07 SAP System Number = 08
DBMS Max Memory = fixed DBMS Max Memory = fixed DBMS Max Memory = fixed
amount amount amount

2 Socket Intel Haswel lC PU32 thr ead / 16 core 2 Socket Intel Haswel lC PU32 thr ead/ 16 core 2 Socket Intel Haswel lC PU32 thr ead/ 16 core 2 Socket Intel Haswel lC PU 32 thr ead / 16 core
384GB RAM 2 NUM AN od es 384GB RAM 2 NUM AN od es 384GB RAM 2 NUM AN od es 384GB RAM 2 NUM AN od es

2 Socket Intel Haswell CPU 2 Socket Intel Haswell CPU 2 Socket Intel Haswell CPU 2 Socket Intel Haswell CPU 10Gi gab i t NIC – Ser v
1Gi gab
er LAN
i t NIC – User LAN
10Gi gab i t NIC – Ser v
1Gi gab
er LAN
i t NIC – User LAN
10Gi gab i t NIC – Ser v
1Gi gab
er LAN
i t NIC – User LAN
10Gi gab i t NIC – Ser v
1Gi gab
er LAN
i t NIC – User LAN

32 thread / 16 core 32 thread / 16 core 32 thread / 16 core 32 thread / 16 core ABA PI nstance 1
ECC AppS e r ver 1
SAP System Number 00
ABA PI nstance 1
ECC AppS e r ver 2
SAP System Number 00
ABA PI nstance 1
ECC AppS e r ver 3
SAP System Number 00
ABA PI nstance 1
ECC AppS e r ver 4
SAP System Number 00
PHYS_ME MSIZE = 32000 PHYS _ME MSIZE = 32000 PHYS_ME MSIZE = 32000 PHYS _ME MSIZE = 32000

384GB RAM 2 NUMA Nodes 384GB RAM 2 NUMA Nodes 384GB RAM 2 NUMA Nodes 384GB RAM 2 NUMA Nodes
10Gigabit NIC – Server LAN 10Gigabit NIC – Server LAN 10Gigabit NIC – Server LAN 10Gigabit NIC – Server LAN
ABA PI nstance 2 ABA PI nstance 2 ABA PI nstance 2 ABA PI nstance 2
ECC AppS e r ver 5 ECC AppS e r ver 6 ECC AppS e r ver 7 ECC AppS e r ver 8
SAP System Number 01 SAP System Number 01 SAP System Number 01 SAP System Number 01
PHYS_ME MSIZE = 32000 PHYS _ME MSIZE = 32000 PHYS_ME MSIZE = 32000 PHYS _ME MSIZE = 32000

1Gigabit NIC – User LAN 1Gigabit NIC – User LAN 1Gigabit NIC – User LAN 1Gigabit NIC – User LAN ABA PI nstance 3
BW App Server 1
SAP System Number 02
PHYS_ME MSIZE = 32000
ABA PI nstance 3
BW App Server 2
SAP System Number 02
PHYS _ME MSIZE = 32000
ABA PI nstance 3
BW App Server 3
SAP System Number 02
PHYS_ME MSIZE = 32000
ABA PI nstance 3
BW App Server 4
SAP System Number 02
PHYS _ME MSIZE = 32000

ABAP Instance 1 ABAP Instance 1 ABAP Instance 1 ABAP Instance 1 ABA PI nstance 4
CRM App Server 1
SAP System Number 03
PHYS_ME MSIZE = 32000
ABA PI nstance 4
CRM App Server 2
SAP System Number 03
PHYS _ME MSIZE = 32000
ABA PI nstance 4
CRM App Server 3
SAP System Number 03
PHYS_ME MSIZE = 32000
ABA PI nstance 4
CRM App Server 4
SAP System Number 03
PHYS _ME MSIZE = 32000

ECC App Server 1 ECC App Server 2 ECC App Server 3 ECC App Server 4
SAP System Number 00 SAP System Number 00 SAP System Number 00 SAP System Number 00 ABA PI nstance 5 ABA PI nstance 5 ABA PI nstance 5 ABA PI nstance 5
SRM App Server 1 SRM App Server 2 SRM App Server 3 SRM App Server 4

PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 SAP System Number 04
PHYS_ME MSIZE = 32000
SAP System Number 04
PHYS _ME MSIZE = 32000
SAP System Number 04
PHYS_ME MSIZE = 32000
SAP System Number 04
PHYS _ME MSIZE = 32000

ABA PI nstance 6 ABA PI nstance 6 ABA PI nstance 6 ABA PI nstance 6


Solm anApp Server 1 Solm anApp Server 2 BI JAVA AppSer ver1 BI JAVA AppSer ver 2
SAP System Number 05 SAP System Number 05 SAP System Number 05 SAP System Number 05
PHYS_ME MSIZE = 32000 PHYS _ME MSIZE = 32000 JVM Heap = 2048 PHYS _ME MSIZE = 32000
JVM Heap = 2048 JVM Heap = 2048 JVM Heap = 2048

ABAP Instance 2 ABAP Instance 2 ABAP Instance 2 ABAP Instance 2 JavaI nstance 7
EP AppSer ver 1
JavaI nstance 7
EP AppSer ver 2
JavaI nstance 7
EP AppSer ver 3
JavaI nstance 7
EP AppSer ver 4

ECC App Server 5 ECC App Server 6 ECC App Server 7 ECC App Server 8
SAP System Number 06 SAP System Number 06 SAP System Number 06 SAP System Number 06
JVM Heap = 2048 JVM Heap = 2048 JVM Heap = 2048 JVM Heap = 2048

SAP System Number 01 SAP System Number 01 SAP System Number 01 SAP System Number 01
PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 ABA PI nstance 8
SCM App Server 1
SAP System Number 07
PHYS_ME MSIZE = 32000
ABA PI nstance 8
SCM App Server 2
SAP System Number 07
PHYS _ME MSIZE = 32000
ABA PI nstance 8
SCM App Server 3
SAP System Number 07
PHYS_ME MSIZE = 32000
ABA PI nstance 8
SCM App Server 4
SAP System Number 07
PHYS _ME MSIZE = 32000

ABAP Instance 3 ABAP Instance 3 ABAP Instance 3 ABAP Instance 3 2 Socket Intel Haswel lC PU32 thr ead / 16 core
384GB RAM 2 NUM AN od
10Gi gab i t NIC – Ser v
1Gi gab
es
er LAN
i t NIC – User LAN
2 Socket Intel Haswel lC PU32 thr ead/ 16 core
384GB RAM 2 NUM AN od
10Gi gab i t NIC – Ser v
1Gi gab
es
er LAN
i t NIC – User LAN
2 Socket Intel Haswel lC PU32 thr ead/ 16 core
384GB RAM 2 NUM AN od
10Gi gab i t NIC – Ser v
1Gi gab
es
er LAN
i t NIC – User LAN
2 Socket Intel Haswel lC PU 32 thr ead / 16 core
384GB RAM 2 NUM AN od
10Gi gab i t NIC – Ser v
1Gi gab
es
er LAN
i t NIC – User LAN

BW App Server 1 BW App Server 2 BW App Server 3 BW App Server 4


SAP System Number 02 SAP System Number 02 SAP System Number 02 SAP System Number 02
PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 ABA PI nstance 1
ECC AppS e r ver 1
ABA PI nstance 1
ECC AppS e r ver 2
ABA PI nstance 1
ECC AppS e r ver 3
ABA PI nstance 1
ECC AppS e r ver 4
SAP System Number 00 SAP System Number 00 SAP System Number 00 SAP System Number 00
PHYS_ME MSIZE = 32000 PHYS _ME MSIZE = 32000 PHYS_ME MSIZE = 32000 PHYS _ME MSIZE = 32000

ABA PI nstance 2 ABA PI nstance 2 ABA PI nstance 2 ABA PI nstance 2


ECC AppS e r ver 5 ECC AppS e r ver 6 ECC AppS e r ver 7 ECC AppS e r ver 8
SAP System Number 01 SAP System Number 01 SAP System Number 01 SAP System Number 01
PHYS_ME MSIZE = 32000 PHYS _ME MSIZE = 32000 PHYS_ME MSIZE = 32000 PHYS _ME MSIZE = 32000

ABAP Instance 4 ABAP Instance 4 ABAP Instance 4 ABAP Instance 4


CRM App Server 1 CRM App Server 2 CRM App Server 3 CRM App Server 4 ABA PI nstance 3
BW App Server 1
ABA PI nstance 3
BW App Server 2
ABA PI nstance 3
BW App Server 3
ABA PI nstance 3
BW App Server 4
SAP System Number 02 SAP System Number 02 SAP System Number 02 SAP System Number 02

SAP System Number 03 SAP System Number 03 SAP System Number 03 SAP System Number 03 PHYS_ME MSIZE = 32000 PHYS _ME MSIZE = 32000 PHYS_ME MSIZE = 32000 PHYS _ME MSIZE = 32000

PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000


ABA PI nstance 4 ABA PI nstance 4 ABA PI nstance 4 ABA PI nstance 4
CRM App Server 1 CRM App Server 2 CRM App Server 3 CRM App Server 4
SAP System Number 03 SAP System Number 03 SAP System Number 03 SAP System Number 03
PHYS_ME MSIZE = 32000 PHYS _ME MSIZE = 32000 PHYS_ME MSIZE = 32000 PHYS _ME MSIZE = 32000

ABA PI nstance 5 ABA PI nstance 5 ABA PI nstance 5 ABA PI nstance 5


SRM App Server 1 SRM App Server 2 SRM App Server 3 SRM App Server 4

ABAP Instance 5 ABAP Instance 5 ABAP Instance 5 ABAP Instance 5


SAP System Number 04 SAP System Number 04 SAP System Number 04 SAP System Number 04
PHYS_ME MSIZE = 32000 PHYS _ME MSIZE = 32000 PHYS_ME MSIZE = 32000 PHYS _ME MSIZE = 32000

SRM App Server 1 SRM App Server 2 SRM App Server 3 SRM App Server 4 ABA PI nstance 6 ABA PI nstance 6 ABA PI nstance 6 ABA PI nstance 6

SAP System Number 04 SAP System Number 04 SAP System Number 04 SAP System Number 04 Solm anApp Server 1
SAP System Number 05
PHYS_ME MSIZE = 32000
JVM Heap = 2048
Solm anApp Server 2
SAP System Number 05
PHYS _ME MSIZE = 32000
JVM Heap = 2048
BI JAVA AppSer ver1
SAP System Number 05
JVM Heap = 2048
BI JAVA AppSer ver 2
SAP System Number 05
PHYS _ME MSIZE = 32000
JVM Heap = 2048

PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000


JavaI nstance 7 JavaI nstance 7 JavaI nstance 7 JavaI nstance 7
EP AppSer ver 1 EP AppSer ver 2 EP AppSer ver 3 EP AppSer ver 4
SAP System Number 06 SAP System Number 06 SAP System Number 06 SAP System Number 06
JVM Heap = 2048 JVM Heap = 2048 JVM Heap = 2048 JVM Heap = 2048

ABA PI nstance 8 ABA PI nstance 8 ABA PI nstance 8 ABA PI nstance 8

ABAP Instance 6 ABAP Instance 6 ABAP Instance 6 ABAP Instance 6 SCM App Server 1
SAP System Number 07
PHYS_ME MSIZE = 32000
SCM App Server 2
SAP System Number 07
PHYS _ME MSIZE = 32000
SCM App Server 3
SAP System Number 07
PHYS_ME MSIZE = 32000
SCM App Server 4
SAP System Number 07
PHYS _ME MSIZE = 32000

Solman App Server 1 Solman App Server 2 BI JAVA App Server 1 BI JAVA App Server 2
SAP System Number 05 SAP System Number 05 SAP System Number 05 SAP System Number 05
PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 JVM Heap = 2048 PHYS_MEMSIZE = 32000
JVM Heap = 2048 JVM Heap = 2048 JVM Heap = 2048

Java Instance 7 Java Instance 7 Java Instance 7 Java Instance 7


EP App Server 1 EP App Server 2 EP App Server 3 EP App Server 4
SAP System Number 06 SAP System Number 06 SAP System Number 06 SAP System Number 06
JVM Heap = 2048 JVM Heap = 2048 JVM Heap = 2048 JVM Heap = 2048

ABAP Instance 8 ABAP Instance 8 ABAP Instance 8 ABAP Instance 8


SCM App Server 1 SCM App Server 2 SCM App Server 3 SCM App Server 4
SAP System Number 07 SAP System Number 07 SAP System Number 07 SAP System Number 07
PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000 PHYS_MEMSIZE = 32000
Configuration 6 Notes:

8 socket servers running in 3 tier configuration can deliver >1,000,000 SAPS for a single SAP component. SAP customers with more than 20 application servers
have demonstrated near linear scaling with the addition of application servers.
Approximately 20% or less of the SAPS of a system are consumed by the database and more than 80% can be scaled out on the application server layer.
Therefore a 1:4 or 1:5 or greater ratio between DB and application server resources is normal for large systems.
Experience from real customer deployments has shown capacity can be scaled very effectively by adding application servers.
The bottleneck is typically not the DBMS server hardware.
10 Gigabit network and a separate dedicated internal SAP application server <-> DBMS network is mandatory for such systems.
Real customer deployments of very large systems with > 25,000,000 dialog steps per day are sometimes constrained by locking/blocking on critical tables such as
NRIV. Application logic and/or business processes must be changed to alleviate such issues. The addition of hardware resources will show little or no benefit

Physical Server Configuration (Left)


DBMS Server
1. 8 socket Intel server with 3TB to 4TB RAM (6TB or more possible with Windows Server 10)
2. DBMS software should be configured to use a large data cache to reduce physical disk IO. SAP customers deployed on 1TB 8 socket systems are running
with DBMS cache size of 940GB (nearly all available RAM is allocated to buffer). DBMS software is NUMA aware and can efficiently run across NUMA
nodes.
3. DBMS software should be configured to use DB compression to further improve the DBMS cache hit ratio. Some databases such as SQL Server and
Sybase systems compress all tables and indexes by default. Consult the relevant SAP Notes discussing compression for each DBMS.
4. DBMS compression ratios of 40% to 85% have been observed on customer systems. If a 15TB system were compressed to 5TB (66% saving) and the
DBMS cache size set to 1TB there would be very minimal physical disk IO. Analysis of customer systems has shown when the DBMS cache size is more
than 25% of DB size in combination with compression very little (if any) physical disk READ IO will occur.
5. Analysis of customer systems with very large DBMS cache sizes have shown almost zero physical read IO 6-12 hours after a system has been restarted.
Physical write will occur as most DBMS use write ahead logging.
6. SSD disk can be used for temporary DB storage spaces and for transaction logs if these physical write operations to disk become critical
7. Some DBMS software requires manual configuration to enable NUMA and Large Memory Page Sizes. Consult the DBMS vendor and/or hardware
vendors best practice document for large scale up deployments.

SAP application servers


1. Between 6 to 30 physical servers 2 socket Intel E5v3 32 threads / 16 cores / 2 NUMA nodes / 384GB RAM (192GB RAM per NUMA node)
2. 10 Gigabit Network for SAP application server to DBMS server communication. A separate 1 Gigabit network for user traffic
3. PHYS_MEMSIZE of all SAP ABAP instances and HEAP of all SAP Java instances should be less than the total physical memory of the server.
4. PHYS_MEMSIZE can be set to either a fixed amount of RAM or a percentage value.
Example:
A DBMS server with 1TB RAM is running DBMS software in a 3 node cluster. DBMS can run on any one of the 3 nodes.
Physical or virtual servers as required – between 6 to 30 balanced with SAP Load Balancing.
Approximately 50 work processes per each ABAP instance. Performance testing has shown many smaller ABAP instances deliver better throughput than a few
large instances. If more work processes are needed it is best to install another instance
Configuration and administration of multiple SAP application servers running on one or more physical or virtual hosts is greatly simplified by placing all profile
parameters into the default profile. This ensures identical configuration of instances, improves the operation of the SAP Logon Load Balancing mechanism and
eliminates complex configuration and administration.

Virtual Machine Configuration (Right)


The amount of RAM and threads required exceeds that currently offered by both VMWare and Microsoft Hyper-V. DBMS software would need to run on
physical servers. 8 socket systems offer 160-240 threads and 4TB RAM which is well in excess of any Hypervisor currently available in May 2015 (no Hypervisor
can create a Guest with 240 threads).

SAP application servers


1. Virtual Machine Configuration can be up to the total number of vCPU and vRAM or Hyper-Visor limit. Virtual Machines configuration must follow the
requirements of SAP Notes such as no overcommit for production systems, adequate network cards and enhanced monitoring
2. 10 Gigabit Network with SRIOv is recommended to improve network performance
3. PHYS_MEMSIZE should be set on Virtual Machines
4. It is not recommended to use Dynamic Memory in conjunction with automatic DBMS memory management features.

Cluster Configuration:
ASCS in 2 node cluster running on two small VMs. The ASCS and ERS consume very little CPU, memory, network and IO.
Most DBMS software can run on more than 2 nodes. Example: Windows 2012 & SQL Server 2012 supports 64 nodes

Multi-SID clusters generally require the use of mount points otherwise there will be insufficient drive letters.
Notes 1634991 - How to install an ASCS or SCS instance on more than 2 cluster nodes
1678705 - Installation scenarios for a standalone ASCS instance
Network Configuration

Cluster Node #1 Cluster Node #2


4 Westmere EX 4 Westmere EX
0.5-1TB RAM 0.5-1TB RAM
~75,000 SAPS ~75,000 SAPS

HEARTBEAT HEARTBEAT

1Gig Team 1Gig Team

10Gig Team 10Gig Team

10Gig Switch 10Gig Switch

Main Core Switch(s)


with HA

10Gig Team 10Gig Team 10Gig Team 10Gig Team

2 CPU 2 CPU 2 CPU 2 CPU


192GB RAM 192GB RAM 192GB RAM 192GB RAM
SandyBridge E5 SandyBridge E5 SandyBridge E5 SandyBridge E5
42,000 SAPS 42,000 SAPS 42,000 SAPS 42,000 SAPS
Site A Site A Site A Site A

1Gig Team 1Gig Team 1Gig Team 1Gig Team

Review this blog: How to Setup a Dedicated SAP Application Server to DB Server Network

Potrebbero piacerti anche