Many SAP customers are switching to SAP HANA to take advantage of its real-time analytics functionality and its ability to perform transactional processing at lightning speed. These capabilities open up an array of opportunities for the business, but to fully benefit, customers must correctly calculate the sizing requirements for SAP HANA to support the in-memory technique that drives its powerful performance, and to?minimize future maintenance costs. This article provides SAP IT and business users with an overview of how to properly size for SAP HANA. It starts with an explanation of the key performance indicators (KPIs) used for SAP HANA sizing, and how sizing for SAP HANA is different from sizing for traditional databases.

Dell and SAP collaborated to create an optimally configured SAP HANA solution that includes a hardware appliance. Dell SAP HANA Solution Overview and Sizing Guide.

It then outlines the two different sizing approaches — initial sizing (for new solution implementations) and delta sizing (for changes to existing solution implementations) — supported by SAP, and how to apply these approaches when sizing for SAP HANA-based implementations. Note: This article primarily focuses on SAP Business Suite and SAP BW, but there are many other products that are also powered by SAP HANA.

SAP offers specific sizing guidelines and recommendations for these products. Key Performance Indicators for SAP HANA Sizing Sizing calculations for determining a system’s, solution’s, or database’s requirements are based on certain KPIs. The three main KPIs used to size for SAP HANA are memory space, CPU processing performance, and disk size.

Memory While traditional sizing approaches focus on CPU performance, the main driver for SAP HANA sizing is memory. Because SAP HANA is a main memory database, essentially all business data (e.g., master and transactional data) resides in the main memory, which leads to a higher memory footprint compared to traditional databases. In addition to the main memory required for storing the business data, temporary memory space is needed to operate the database management system — to support complex queries or data that is needed for buffers and caches, for example. CPU Sizing for SAP HANA includes unique requirements for CPU processing performance. The CPU behaves differently with SAP HANA compared to traditional databases.

The processing engine for SAP HANA is optimized to operate very complex queries at maximum speed, which means that many of these queries are processed internally and in parallel, and most of the data is stored in a column-based format. This architecture not only might lead to a higher CPU demand compared to traditional databases, it also requires planning for a lower average utilization to ensure that there is enough headroom for the database to process queries sufficiently fast. Disk Size An in-memory database still requires disk storage space — to preserve database information if the system shuts down, either intentionally or due to a power loss, for instance. Data changes in the database must be periodically copied to disk to ensure a full image of the business data on disk, and to preserve the current state of the database and all of the data entered in the persistence layer.

In addition, a logging mechanism is required to log changes and enable system recovery. To accommodate these requirements, there always must be enough space on disk to save data and logs. Let’s now take a look at the two main approaches that use these KPIs for sizing?SAP environments: initial sizing and delta sizing. Initial Sizing vs. Delta Sizing SAP distinguishes between two types of sizing: initial sizing and delta sizing.

With an initial sizing, the sizing is being.

The following example is provided to help you and Gateway customer to understand the delta sizing calculations for the SAP backend systems using flows called through SAP NetWeaver Gateway. It illustrates initial sizing for backend components like standard SAP services or customer own code that is called remotely via SAP NetWeaver Gateway. In addition, the example provides some recommendations on how to deal with, or avoid overloading of the backend systems. More information about Gateway sizing can be found in the Gateway sizing guide document located in → → → (1) or (2) →. Data preparation for sizing calculations • Customer A identified the peak time period of one hour for their entire business flows. During this peak hour, the end users that consumed Gateway services triggered the most number of calls.

For example, it can be Monday morning between 9:00 and 10:00 AM. • Customer A identified the most commonly used flows, out of all possible flows that consume Gateway. Usually, 20% of the flows are called in 80% of the cases. Customer A also identified several flows out of the most commonly used flows in the existing CRM system, and some flows in the existing ERP system through Gateway. The ERP and CRM systems are already in use by non-Gateway scenarios, and customer A has collected information about the current CPU and memory utilization during a working day and required peak hour for application and database (DB) layers.

This information is essential in order to find whether current ERP and CRM systems can handle the additional load initiated by Gateway. • Customer A extrapolated the maximum number of calls that will be called by every end user per each of the flows found in step 2 above. The number of calls extrapolation can be also done according to averages, but for more precise sizing, it is better to take the maximum. For example, the following information was identified: Maximum operations per peak hour table. Flow description Maximum Number of operations in peak hour per end user Backend system Query Sales Activities for 1 day period 3 CRM Query Sales Activities for 1 week period 1 CRM Query contacts 3 CRM Query products according to sales organization and customer Id.

10 ERP Retrieve pricing of material according to sales organization, customer Id and material ID. 10 ERP In this example, total number of operations per end user per peak hour equals to 27.

There are 7 operations in the CRM system and 20 operations in the ERP system. For example, current CRM and ERP systems utilization (before Gateway) during the peak hour: SAP systems utilization during peak hour table. ERP system utilization CRM system utilization Application layer DB layer Application layer DB layer CPU Memory CPU Memory CPU Memory CPU Memory 40% 30% 30% 20% 30% 40% 30% 20% The assumption is that the peak hour of the backend systems is not shifted by Gateway usage. Another assumption is that peaks for backend and Gateway systems are at the same peak hour. Customer A must find the number of the end users that will trigger the calls toward Gateway during peak hour. (For example 10,000 users) 5.

Customer A asked their hardware partner to provide the SAPS of ERP and CRM systems. If the customer’s hardware partner does not inform him about the SAPS of his systems, he can get similar systems at: For example, the following SAPS numbers and memory were provided to a customer by the hardware partner. SAP systems SAPS & Memory table. ERP system CRM system Layer Application layer DB layer Application layer DB layer SAPS 150,000 30,000 100,000 30,000 Memory (GB) 400 100 300 100 For more information about SAPS go to: Prerequisites for sizing calculations To start the sizing calculations, make sure that the following requirements are met: • Single user test requirements. Customer has to verify that every flow specified by him for sizing calculations meet his performance Key Performance Indicators (KPIs).

Sometimes the customer has to implement Gateway interfaces, and has to verify that his implementation (especially in backend systems) meet his KPIs. The verification can be done by using STAD transaction by monitoring response time, CPU and memory consumption for every flow on your Gateway and backend systems. • The flow has scalable behavior. The customer has to verify at least data scalability dimension. Example for non linear behavior: the CPU consumption for retrieving 10 objects is 100 ms, but retrieving 100 objects it is 2 seconds. The expected CPU consumption for 100 objects retrieval is under 1 second.

• The customer can verify also user scalability dimension by having parallel users as incremental factor and monitoring system resources. Sizing calculations Once Customer A completed the data preparation and prerequisites steps, customer A started with sizing calculation process in order to find SAPS values of his Gateway and backend systems. First step: Initially, customer A found CPU consumption of the Application and DB layers for Gateway and backend systems of every flow participated in the sizing guide by using STAD transaction. For example, the following are the CPU and DB consumption numbers for the single user flows: Single operation consumption table. Flow description Gateway consumption (ms) ERP Backend system consumption (ms) CRM Backend system consumption (ms) CPU DB CPU DB CPU DB Query Sales Activities for 1 day period 30 5 N/R N/R 500 150 Query Sales Activities for 1 week period 50 5 N/R N/R 800 200 Query contacts 40 5 N/R N/R 200 100 Query products according to sales organization and customer Id. 50 5 300 150 N/R N/R Retrieve pricing of material according to sales organization, customer Id and material ID.

30 5 150 100 N/R N/R Second step: Now, customer A has to identify the required consumption on application and DB layers for all operations during peak hour per each flow and for all users. In order to do so, every value in the table “Single operation consumption” that was found in step 1, will be multiplied with “ Maximum operations per peak hour ” table (see the section, Data preparation for sizing calculations) according to the flow description.

In addition, each value will be multiplied with the end users number (see the section, Data preparation for sizing calculations) which will trigger the calls toward Gateway during peak hour (10,000 users). Microsoft Autoroute Euro 2013 Full Lz0275a00582501 on this page. For example, Gateway total required consumption on application layer for “Query contacts” flow will be calculated as following. Flow description Maximum Number of operations in peak hour per end user Gateway total consumption (sec) per hour ERP Backend system total consumption (sec) per hour CRM Backend system total consumption (sec) per hour CPU DB CPU DB CPU DB Query Sales Activities for 1 day period 3 900 150 N/R N/R Query Sales Activities for 1 week period 1 500 50 N/R N/R 8000 2000 Query contacts 3 1200 150 N/R N/R 6000 3000 Query products according to sales organization and customer Id. 10 5000 500 0 N/R N/R Retrieve pricing of material according to sales organization, customer Id and material ID. 10 3000 500 0 N/R N/R Third step: Since the operations will be executed during one hour, customer A calculated the number of single computing units (SCUs), it is needed to complete the required processes for Gateway and backend systems of application and DB layers per each flow.

Customer A divided every value in table “Total consumption” by 3,600 per hour (1 hour = 60 minutes*60 seconds). Number of SCUs required completing the required processes table. Flow description GW – number of SCUs ERP Backend system – number of SCUs ERP Backend system – number of SCUs Layer Application DB Application DB Application DB Query Sales Activities for 1 day period 0.25 0.041667 N/R N/R 4.16666667 1.25 Query Sales Activities for 1 week period 0.13888889 0.013889 N/R N/R 2.22222222 0.555556 Query contacts 0.33333333 0.041667 N/R N/R 1.66666667 0.833333 Query products according to sales organization and customer Id. 1.38888889 0.138889 8.33333333 4.16667 N/R N/R Retrieve pricing of material according to sales organization, customer Id and material ID.

0.83333333 0.138889 4.16666667 2.77778 N/R N/R Forth step: Customer A identified the required SAPS values of Application and DB layers for Gateway and backend systems. In order to complete the calculations, customer A identified the “Single computing Unit” SAPS value of the Gateway, ERP and CRM systems in which the measurements done. For more information about “Single computing Unit” please read SAP Note number 1501701, or contact your hardware partner.

For example, customer A identified that “Single computing Unit” SAPS value of the Gateway, ERP and CRM systems equals to 1,000 SAPS. In addition, customer A took into account that target CPU utilization on all layers for all machines should not exceed 65%. In order to get the required SAPS value for application and DB layers on GW, EPR and CRM machines, customer A multiplied “Single computing Unit” (1,000 SAPS) by every value in “Number of SCUs required to complete the required processes” table found in step 3 and divided each value by factor 0.65. Final SAPS values table. Flow description GW system Required SAPS ERP Backend system Required SAPS CRM Backend system Required SAPS Layer Application DB Application DB Application DB Query Sales Activities for 1 day period 385 64 N/R N/R 6410 1923 Query Sales Activities for 1 week period 214 21 N/R N/R 3419 855 Query contacts 513 64 N/R N/R 2564 1282 Query products according to sales organization and customer Id. 2137 214 N/R N/R Retrieve pricing of material according to sales organization, customer Id and material ID.

1282 214 6410 4274 N/R N/R Total 4530 577 4. The Total SAPS numbers represent the required SAPS for application or DB layers on Gateway, ERP and CRM machines for processing Gateway flows. Customer A also identified the memory consumption required for Gateway flows for application and DB layers on gateway, ERP and CRM systems as following: Gateway application & DB layers: memory consumption formula is 1,000 SAPS = 1.5 GB memory. ERP & CRM application layer (only valid for Gateway scenarios): memory consumption formula is 1,000 SAPS = 1.5 GB memory. ERP & CRM DB layer (only valid for Gateway scenarios): memory consumption formula is 1,000 SAPS = 3 GB memory.

Flow description GW system Required Memory(GB) ERP Backend system Required Memory (GB) CRM Backend system Required Memory (GB) Layer Application DB Application DB Application DB Total SAPS 4530 577 4 Total memory 7 GB 1 GB 30 GB 32 GB 20 GB 12 GB Fifth step: Customer A identified the hardware required for Gateway system, and whether their ERP and CRM systems with current hardware can serve Gateway flows in addition to the current systems utilization. Customer A contacted their hardware partner and requested for a Gateway server hardware according to the SAPS found in forth step. As already mentioned in the section, Data preparation for sizing calculations, the conversion of SAPS to potential hardware is done by SAP technology partners. In addition, customer A identified the percentage of ERP and CRM SAPS as found in forth step, out of the total amount of SAPS specified in the section, Data preparation for sizing calculations. Percentage of required SAPS out of total SAPS table.

ERP system SAPS CRM system SAPS Layer Application layer DB layer Application layer DB layer Total SAPS (already found) 150,000 30,000 100,000 30,000 Additional SAPS required for Gateway flows (as found in forth step) 4 Percentage of SAPS required for Gateway flows out of total 13% 36% 13% 14% Customer A identified the percentage of ERP and CRM memory consumption found in forth step out of total memory consumption specified in “Data preparation for sizing calculations” st ep 5. ERP system utilization CRM system utilization Application layer DB layer Application layer DB layer CPU Memory CPU Memory CPU Memory CPU Memory Current load without Gateway 40% 30% 30% 20% 30% 40% 30% 20% Target utilization including factor of 65% 62% N/R 46% N/R 46% N/R 46% N/R Additional load caused by Gateway flows 13% 8% 36% 32% 13% 7% 14% 12% Total load (Current + Gateway) 75% 38% 82% 52% 59% 47% 60% 32% Customer A identified the ERP and CRM systems can handle additional load caused by Gateway flows since all the numbers are below 100%. In case you identify that one of the total load values in percentage is over 100% you can consider the following: • Change the value target utilization to be more than 65%.

Highly recommended for big systems and DB systems. • Consult your hardware partner regarding additional hardware resources. Add more Hardware to support additional load caused by SAP NetWeaver Gateway flows. • In case it is not possible to add more hardware for your existing SAP systems (especially DB layer), consider implementing the throttling BADI provided within SAP NetWeaver Gateway 2.0 SP5 version.

Customer A successfully implemented SAP NetWeaver Gateway solution and they are live with a production system.