IBM LinuxONE III Brings Performance Advantages to the Temenos Transact Banking Solution
To examine the efficacy of IBM LinuxONE III for the Temenos Transact R19.08 banking solution, IBM and Temenos conducted benchmark testing
By Dr. J C Yao11/17/2020
To examine the efficacy of IBM LinuxONE III for the Temenos Transact R19.08 banking solution, IBM and Temenos conducted benchmark testing2 to examine LinuxONE functional, performance and platform specific capabilities.
Transactions per CoreLab testing found that Temenos Transact R19 on LinuxONE provides up to 100 TPS per core as shown in Figure 1 below. This offers increased transaction throughput and scalability for growing workloads as compared to 28 TPS per core on compared x86 servers or 8.6 TPS per core in x86 public cloud environments3.
Figure 1. Transactions per Second (TPS) per Core
Response TimeLinuxONE is able to maintain low latency, enabling swift transaction response time. Temenos Transact R19 on LinuxONE enables completion of transactions (58 milliseconds) in one third of the time of compared x86 servers (160 milliseconds) and one fourth of the time of x86 public cloud environments (240 milliseconds)4 as shown in Figure 2.
Figure 2. Average Response Time
Database RequirementsLow latency can improve database performance and can decrease database capacity requirements. The testing found that Temenos Transact R19 on LinuxONE uses one fifth of the capacity of its application servers for its database capacity versus one third of the capacity of its application servers for the same database capacity on compared x86 servers.5
Significant Capacity Reduction for LinuxONEHigher transaction throughput per core, lower latency and less required database capacity combined to give IBM LinuxONE significant performance advantages over x86 servers, as shown in Figure 3. This can reduce infrastructure, maintenance efforts and costs.
Figure 3. Application and database capacity comparison
Temenos Transact on LinuxONE Financial BenefitsIBM and Temenos examined the financial impact of running the Temenos Transact R19.08 banking solution on IBM LinuxONE III versus an x86 server environment. To analyze costs, a three-year TCO model was built for each of the five TPS requirement use cases using LinuxONE III and x86 servers. Each TCO model included acquisition and maintenance hardware costs, software, floor space, energy, networking and disaster recovery costs based on data from client environments in IBM IT Economics assessments with the following assumptions:
- All scenarios assumed 15% year-to-year production capacity growth based on peak TPS.
- The models included an additional 100% of production capacity to estimate non-production environments (development, test and quality assurance).
- All use case scenarios included a disaster recovery environment that required an additional 100% of production environment resources.
- All use cases were modeled with the smaller LinuxONE III LT2 system since TPS requirements over three years could be satisfied within the maximum number of available Integrated Facility for Linux (IFLs) in the smaller server.
TCO Model Findings
- LinuxONE delivered better price/performance ratios in all TPS use case scenarios. Depending on the infrastructure required for different client transaction throughput requirements from 100 to 2,000
- TPS, TCO savings over three years with LinuxONE ranged from 7% to 60% versus the compared x86 environments6 as shown in Figure 4.
- While 16-core x86 servers can support small transaction throughput, larger 32-core servers are required when peak transaction volume reaches 1,000 TPS. Similarly, 80-core or larger x86 servers may be required to support database servers with higher peak TPS beyond 2,000 TPS. The x86 server configuration changes requires significant effort and cost in servers acquisition, installation, and maintenance.
- Our TCO models show LinuxONE III LT2 provides seamless scale up with a single production server for use cases requiring up to 3,200 TPS and, with a maximum capacity of 190 cores in a LinuxONE III LT1, a dual server configuration with two LinuxONE III LT1 servers in the main data center can potentially support up to 8,000 TPS.
Figure 4. Three Year TCO Comparison
Why LinuxONE Reduces CostIBM LinuxONE is a centralized, 100% Linux®, enterprise server designed for large data serving and Transactional applications7. It is designed for the highest level of HSM security (FIPS 140-2 Level 4, EAL 5+ certification) to protect against cyberattacks and other security threats.
LinuxONE servers are engineered to deliver essential business-critical quality of services such as high availability, security, scalability and reliability through the following capabilities:
- Security with on-chip cryptographic performance to provide faster encryption of data in-
- flight and at-rest and pervasive encryption8
- Pervasive encryption features with HSM-based key management9 and Secure Service Containers10 to reduce security risks and facilitates compliance with security standards such as General Data Protection Regulation (GDPR)
- Large-scale server workload consolidation11
- Seamless scaling to meet changing business requirements
- Java pause-less garbage collection capabilities that reduce pause duration and avoid potential impacts to Transaction response time
- Simultaneous multi-threading (SMT) capabilities providing significant performance advantages12
- Reduced planned and unplanned downtime13
- Virtualization and system management tools to simplify and expedite administration14
- Lower administration costs compared to distributed server environments15
- GDPS® for robust disaster recovery processing16
- Average high sustained CPU utilization that enables consolidation of many workloads onto a single server17
1. Temenos Transact was formerly called Temenos T24 https://www.temenos.com/us/news/2020/03/06/t24-is-now-temenos-transact/
3. Results were achieved using JMeter to drive transactions in a simulated client banking transaction environment using MQ to measure average transactions per second (TPS) and response time. Temenos Transact testing was performed at IBM on a LinuxONE III LT1 system with 52 IFLs to reach more than 3,000 TPS. Temenos testing was performed on Sandy Bridge x86 servers in both cloud and x86 on-prem environments. x86 results were extrapolated for newer x86 chipsets. TPS results
4. Results were achieved using JMeter to drive transactions in a simulated client banking transaction environment using MQ to measure average transactions per second (TPS) and response time. Temenos Transact testing was performed at IBM on a LinuxONE III LT1 system with 52 IFLs to reach more than 3,000 TPS. Temenos testing was performed on Sandy Bridge x86 servers in both cloud and x86 on-prem environments. x86 results were extrapolated for newer x86 chipsets. TPS results were obtained under laboratory conditions, not in an actual customer environment. Results were obtained under laboratory conditions, not in an actual customer environment. Results will vary.
5. Results were achieved using JMeter to drive transactions in a simulated client banking transaction environment using MQ to measure average transactions per second (TPS) and response time. Temenos Transact testing was performed at IBM on a LinuxONE III LT1 system with 52 IFLs to reach more than 3,000 TPS. Utilized application cores versus database cores data showed that LinuxONE used one fifth of application cores compared to database cores, while x86 data showed that utilized database capacity correlates to one third of application capacity. Temenos testing was performed on Sandy Bridge x86 servers in both cloud and x86 on-prem environments. x86 results were extrapolated for newer x86 chipsets. TPS results were obtained under laboratory conditions, not in an actual customer environment. Results will vary.
6. An IBM IT Economics total cost of ownership three-year model was used to examine LinuxONE and x86 hardware, software, floor space, energy, networking and disaster recovery costs. Five use cases based on transaction throughput requirements, measured in transactions per second (TPS), were analyzed using LinuxONE III LT2 servers with 9 IFLs versus 42 utilized x86 cores on 11 16-core Skylake x86 servers for the 100 TPS use case, 31 IFLs versus 180 utilized x86 cores on 17 16-core Skylake x86 servers for the 500 TPS use case, 59 IFLs versus 360 utilized x86 cores on 17 32-core Skylake x86 servers for the 1,000 TPS use case, 90 IFLs versus 540 utilized x86 cores on 23 32-core. Skylake x86 servers for the 1,500 TPS use case, and 124 IFLs and 717 utilized x86 cores on 15 80-core Cascade Lake x86 servers for the 2,000 TPS use case. Both environments included RHEL, a web application server, and a relational DB with high availability support. Prices, where applicable, are based on U.S. prices as of February 2020 from client data in IT Economics assessments with 40% discounting. x86 hardware pricing is based on IBM analysis of U.S. prices as of 06/10/2020 from IDC. The cost model assumes 15% annual growth of peak TPS to project production capacity. An additional 100% of production capacity was projected to provision development, test, and QA environments, and an additional 100% of production capacity was projected for a remote site for disaster recovery. LinuxONE and x86 TPS sizing requirements used in the cost model were extrapolated for LinuxONE III LT2 based on results obtained under laboratory conditions on LinuxONE III LT1, not in an actual customer environment. Results will vary.
13. ITIC server reports https://itic-corp.com
17. IBM internal tests show that application servers and databases exhibiting high-peak-to-average levels of workload demand in a distributed architecture running on average at 10% - 20% CPU utilization can run on LinuxONE III LT1 at the same throughput while maintaining a higher sustained average CPU utilization of 89% reducing wasted CPU resource. This claim is designed to replicate a mix of application servers and database workloads on x86 and IFLs on IBM Z using JMeter and HammerDB. Workloads were driven to emulate real customer environments running under a service level agreement of 130,400 TPS within 500 milliseconds 95% of the time. For the x86 environment workload activity drivers were configured to represent a population of servers with a variety of workloads exhibiting varying levels of demand for computing resources similar to that observed in 15 client environments for development, test, quality assurance, and production levels of CPU and throughput. The test measured average CPU utilization that varied between 10% and 20% for x86 workloads running on 65 16-core, 24-core and 28-core x86 servers with 1,476 cores total. For the IFL environment the same workload activity drivers were used on 120 LinuxONE III LT1 IFLs and driven at 89% CPU utilization achieving the same required transactions per second and response time as were observed on the x86 servers. LinuxONE III LT1 throughput was projected from actual LinuxONE Emperor II test results. Results may vary.
Dr. J C Yao is the Chief Strategist of IBM IT Economics team. J C’s expertise includes computer engineering applications, numerical modeling and analysis, middleware, software architecture, solution framework, IT financial analysis and software/hardware comparative positioning to help clients address IT concerns.
Sponsored ContentAchieve Compliance Without Impacting Productivity
Post a Comment
Note: Comments are moderated and will not appear until approvedcomments powered by Disqus