Datapower Capacity Planning for X3 and Virtual Form Factors

Datapower Capacity Planning for X3 and Virtual Form Factors

1. TechCon 2023 Session #332 DataPower Capacity Planning for X3 and Virtual Form Factors IBM Marketing / October 2022 / © 2022 IBM Corporation 1 Ivan Heninger DataPower and API Connect Performance Engineering

2. 2 Integration Technical Conference / © 2020 IBM Corporation Ivan Heninger worked 10 years prior to college in manufacturing and deployment of antenna systems, 3 MHz to 40 GHz. This included mechanical assembly, machine shop, tower climbing, and customer service assignments. Ivan entered college as an adult and received a BSEE from the State University of New York at Binghamton. After several years as an antenna Test and Measurement Engineer, Ivan moved to Software Engineering. Currently, Ivan is a Sr Engineer at IBM leading the Performance Engineering efforts on the DataPower and API Connect product development teams. Ivan has more than 20 patents from his 25 years of work in Software Engineering. Ivan enjoys spending time with Jackie his wife, their grandchildren (and their parents), and large garden in North Carolina USA. Session 332 Speaker bio IBM Marketing / October 2022 / © 2022 IBM Corporation

3. What’s new in 2023 DataPower and APIGateway from performance perspective IDG X3 CPU, memory, disk, network 10.5.0 LTS firmware A few X3 benchmark results: TC1 soap request validate , sort, base64 encode TC2 soap request validate, sign, verify TC3 rest request validate, gws xform run on response APIC Simple-invoke and Complex-map including: rate-limit enforcement analytics capture and send clientID security model “Lift and shift” sizing guide Use your current deployment’s resource utilization data to estimate resource requirements for other DP form factors running the same workloads Map IDG from today’s production to a planned equivalent Kubernetes or OVA cluster Chart on the left is a Smith Chart used in impedance matching of RF networks Queuing software networks also benefit from impedance matching of software nodes IBM Marketing / October 2022 / © 2022 IBM Corporation Session 332 Agenda

4. What’s new in 2023 DataPower and APIGateway IDG X3 Physical Appliance 10.5.0 LTS firmware, APIC life-cycle event processing Config-sequence refactor Batch and de-dup commit list of objects referenced by updated object 9006/8436 IDG 9007/8441 IDG X2 9008/8496 IDG X3 Processor 2*Intel Xeon (2.8GHz) 20 Cores/40 Threads 2*Intel Xeon (2.6GHz) 24 Cores/48 Threads 2*Intel Xeon (2.9Ghz) 32 Cores/64 Threads Memory 192 GB 192 GB 256 GB Network 2 x 10Gbps 8 x 1Gbps 6 x 10Gbps 8 x 1 Gbps 2 x 40Gbps 4 x 10Gbps 8 x 1Gbps Disk 2 x 1.2 TB HDD 2 x 1.2 TB HDD 2 x 1.6 TB - SSD IBM Marketing / October 2022 / © 2022 IBM Corporation

5. TechCon 2023 Agenda A few X3 benchmark results: Testbed topology Rest request validate, gws xform run on response X2 and X3 CPU saturation S oap request validate , sort, base64 encode X2 network saturation Soap request validate, sign, verify X2 network saturation APIC including: rate-limit enforcement analytics capture and send clientID security model Simple-invoke X2 network saturation Complex-map X2 and X3 CPU saturation IBM Marketing / October 2022 / © 2022 IBM Corporation

6. DP749 DP747 DP614 Test Topology: Drivers, DP Under Test, Backend DP811 X3 DP809 X3 V10.5.0.2 DP617 DP618 DP640 DP641 DP750 X2 DP746 X2 DP748 X2 dpcentoscloud dpserver44 10 Gbps 40 Gbps switch dpclone801 DataPowers under test Load Drivers Idealized Backend cluster 40 Gbps ingress 10 Gbps DPESX11 VMWare 6.7.0 Update 3 DPVIRT5b OVA Fleet DPSERVER46 K8S CLUSTER v1.24.3 132 CPU/822 GB Mem dpserver41 dpserver42 dpserver44 dpserver45 dpserver48 dpserverNN IBM Marketing / October 2022 / © 2022 IBM Corporation

7. X3/X2 10.5.0.x benchmark results “How to” read the tables IBM Marketing / October 2022 / © 2022 IBM Corporation 2 appliances, X3 and X2 Minimum DataPower latency recorded for each payload X3 and X2 DataPower latency recorded at maximum TPS for each payload X3 and X2 Maximum TPS for each payload X3 and X2

8. X3/X2 10.5.0.x benchmark results json validate+gws response transform 1 Request content validation against local schema Response transformation using gateway-script X3 tops 34K TPS with under 10 mS latency IBM Marketing / October 2022 / © 2022 IBM Corporation

9. X3 10.5.0.x benchmark results json validate+gws response transform 2 Includes 50 mS backend latency Raw data: TPS, Latency and DataPower CPU Each run is increasing payload 4KB, 8KB, 16KB, 128KB, 256KB, 1MB IBM Marketing / October 2022 / © 2022 IBM Corporation

10. X3/X2 10.5.0.x benchmark results soap validate+sort+base64 encode 1 1,000,000 KBps * 8 = 8Gbps X2 10Gbps Network Saturated See raw results network chart IBM Marketing / October 2022 / © 2022 IBM Corporation Request content validation against local schema and sort list within request Base 64 encode XML node within response For the 3 larger payload cases on X2, the 10Gbps network is saturated preventing CPU saturation X3 981 TPS with 1024KB payload drives 16.8 Gbps of network bandwidth X3 topline 44K TPS 4KB payload

11. Excludes 50 mS backend latency IBM Marketing / October 2022 / © 2022 IBM Corporation Raw data: TPS and Latency in “Little’s Law” form comparing X2 to X3 Each run is increasing payload 4KB, 8KB, 16KB, 128KB, 256KB, 1MB X3/X2 10.5.0.x benchmark results soap validate+sort+base64 encode 2 X3 X2

12. 1,000,000 KBps * 8 = 8Gbps Network Saturated 2,100,000 KBps * 8 = 16.8Gbps X2 X3 IBM Marketing / October 2022 / © 2022 IBM Corporation X3/X2 10.5.0.x benchmark results soap validate+sort+base64 encode 3

13. X3/X2 10.5.0.x benchmark results soap validate+sign+verify 1 1,000,000 KBps * 8 = 8Gbps X2 10 Gbps Network is Saturated IBM Marketing / October 2022 / © 2022 IBM Corporation 1,800,000 KBps * 8 = 14.4Gbps X3 40 Gbps network not saturated Request content validation, sign request content and verify signature For the 3 larger payload cases on X2, the 10Gbps network is saturated preventing CPU saturation X3 862 TPS with 1024KB payload drives 14.4 Gbps of network bandwidth

14. X3/X2 10.5.0.x benchmark results soap validate+sign+verify 2 IBM Marketing / October 2022 / © 2022 IBM Corporation Raw data: X2 network saturation, X3 40 Gbps capability Each run is increasing payload 4KB, 8KB, 16KB, 128KB, 256KB, 1MB 1,000,000 KBps * 8 = 8Gbps Network Saturated 1,800,000 KBps * 8 = 14.4Gbps X2 X3

15. X3/X2 10.5.0.x benchmark results APIC 10.0.5.2 Assemblies APIGW using v2.0.0 actions, native mode Simple-invoke Complex-map Options included : rate-limit enforcement analytics capture and send (activity level capture) clientID security model IBM Marketing / October 2022 / © 2022 IBM Corporation

16. Minimal API with full function analytics, rate limiting and client-ID based security Iterate on request payload size Small response IBM Marketing / October 2022 / © 2022 IBM Corporation X3/X2 10.5.0.x benchmark results APIC 10.0.5.2 Assemblies, Simple Invoke X2 10 Gbps Network is Saturated

17. JSON request payload mapped to XML and response payload mapped to JSON 100 % of request and response payload is mapped Iterate on request payload IBM Marketing / October 2022 / © 2022 IBM Corporation X3/X2 10.5.0.x benchmark results APIC 10.0.5.2 Assemblies, Complex Map

18. “Lift and shift” sizing guide Use your current deployment’s resource utilization data to estimate resource requirements for other DP form factors running the same workloads Map IDG from today’s production to a planned equivalent Kubernetes cluster IBM Marketing / October 2022 / © 2022 IBM Corporation TechCon 2023 Agenda

19. Estimate Required CPU Count to Replicate Existing Deployment Using a Different Form Factor Physical form has the largest dynamic range of capacity per instance Highest TPS with minimum machine count Lowest ethernet port count with maximum bandwidth Virtual and container forms have reduced dynamic range of capacity per instance IBM Marketing / October 2022 / © 2022 IBM Corporation

20. Estimate Required CPU Count to Replicate Existing Deployment Using a Different Form Factor Peak CPU utilization for the existing service? You know your business! Hourly, busiest hour of day Daily, busiest day of week Weekly, busiest week of month Monthly, busiest month Event peak: annual enrollment Cyber-Monday end of horse race at Cheltenham Festival, or … IBM Marketing / October 2022 / © 2022 IBM Corporation

21. Estimate Required CPU Count to Replicate Existing Deployment Using a Different Form Factor Picking the right peak as design point for capacity in the new deployment is a choice Hourly or daily peak is good starting point For weekly, monthly, and event peaks, temporary additional capacity can be added Designing capacity to meet spikes that last only a few seconds is not recommended It is also important to consider capacity required for management-plane deployments of application content onto the DataPower The previous discussion of CPU peak is focused on data-plane workload DataPower (non APIC) policy updates Planned and unplanned outages, high availability event “reserve capacity” to absorb workload during outages API Connect catalog life cycle, product publish and application subscribe, events IBM Marketing / October 2022 / © 2022 IBM Corporation

22. Estimate Required CPU Count First, we normalize the actual CPU % to an equivalent number of CPUs at operating 100% IDG has 40 execution threads, 20 CPUs with hyper threading enabled For example: 40% of 40 vCPUs is 16 equivalent vCPUs operating at 100% Next, we add the overhead costs of virtualization and K8s, ~20% is typical and you can pick the value you believe is correct for your environment 16 * 1.2 = 19.2 vCPUs at 100% Last, we de-rate the 100% utilization to the desired maximum CPU utilization, say 60% 19.2 / .6 = 32 CPUs This is to replace 1 IDG so * 3 is 96 CPUs to replace 3 IDGs We also need to decide on how many CPUs we want to use per DP instance 8 CPU per instance is a common choice and would result in 12 instances See spread sheet on next slide IBM Marketing / October 2022 / © 2022 IBM Corporation

23. Estimate Required CPU Count This spread sheet can be used to facilitate “what if” design review IBM Marketing / October 2022 / © 2022 IBM Corporation Number of CPUs in 1 instance of current solution Virtualization and k8s overhead, 20% Existing deployment’s peak CPU Target CPU utilization on new solution Number of CPUs per instance in new solution

24. Thank you. Please take the survey in the chat box! Want more? Join the IBM Integration Community Attend Proof of Technology workshops in Phoenix and Atlanta Meet us in Vegas Sept 11-14 at TechXchange IBM Marketing / October 2022 / © 2022 IBM Corporation 25