Used IBM Servers | New Power 10 Systems | QRadar SIEM Security
8 minutes reading time
(1661 words)
IBM Linux server hardware performance and software license costs
IBM Linux server hardware performance and software license costs
The best Linux server hardware for enterprise open-source ecosystems will be the platform that delivers the best processor, memory and disk performance per core. Judging by the number of discussions in customer support communities, technical articles and blog posts on various Linux server performance troubleshooting, there is clearly a need for a better performing Linux server for both scale up and scale out infrastructures. This article will explain the importance of selecting the best performing hardware for your Linux server farm and how it affects software license costs.
IBM POWER10 and IBM POWER9 Linux Server Hardware models:
9080-HEX E1080 1-4 Processors, 40-240 Cores, 16 TB Memory
9009-22G S9221-2 Processors, 8-22 Cores, 4TB Memory
9008-22L S922 1-2 Processors, 8-12 Cores, 4TB Memory
9009-22G S9221-2 Processors, 8-22 Cores, 4TB Memory
9008-22L S922 1-2 Processors, 8-12 Cores, 4TB Memory
9183-22X IC922 1-2 Processors, 12-20 Cores, 2TB Memory
9006-12P LC921 1-2 Processors, 16-40 Cores, 1TB Memory
9006-22P LC922 1-2 Processors, 16-44 Cores, 1TB Memory
9009-41G S914 1 Processor, 4-8 Cores, 1TB Memory
9009-41A S914 1 Processor, 4-8 Cores, 1TB Memory
9009-42G S924 1-2 Processors, 8-24 Cores, 4TB Memory
9009-42A S924 1-2 Processors, 8-24 Cores, 4TB Memory
9009-22A S922 1-2 Processors, 4-10 Cores, 4TB Memory
9006-12P LC921 1-2 Processors, 16-40 Cores, 1TB Memory
9006-22P LC922 1-2 Processors, 16-44 Cores, 1TB Memory
9009-41G S914 1 Processor, 4-8 Cores, 1TB Memory
9009-41A S914 1 Processor, 4-8 Cores, 1TB Memory
9009-42G S924 1-2 Processors, 8-24 Cores, 4TB Memory
9009-42A S924 1-2 Processors, 8-24 Cores, 4TB Memory
9009-22A S922 1-2 Processors, 4-10 Cores, 4TB Memory
Should I scale up or scale out my Linux server farm?
A scale up Linux server environment enables economies and performance of scale for all levels of the IT infrastructure, by reducing the number of physical systems and underlying ecosystem components that will require administration and support. Scaling up means less human resources needed, less software licenses, less hardware, less failures, less power, less vulnerabilities and less everything else that negatively effects costs. In addition, a scale up environment can have power and resources available to share much more efficiently than a scale out infrastructure, which must distribute workloads across multiple servers, will increase software licenses, as well as infrastructure and system administration costs at every level. Almost every virtualized Linux server environment can benefit from a scale up server approach to save money and improve performance.
IBM Power9 Linux server performance results have consistently shown it is the most cost effective and fastest midrange Linux platform to run database, ERP, big data analytics, Java, video, cloud, Nutanix, web, file system, DevOps. Any CPU, I/O intensive or memory demanding Linux server workload running NoSQL like MongoDB, Relational DBMS like MySQL, PostgreSQL, MariaDB and Oracle, or Hadoop and Spark big data platforms running on Cloudera will achieve significant per core cost savings and performance gains by reducing latency and throughput using a Power9 Linux server.
Add up the potential Power9 performance and price savings
Up to 5.6X I/O bandwidth versus x86
1.8X more data bandwidth: 230 GB/sec per Power9 socket versus to 128GB/sec per socket for x86
2X data throughput: 31.5 GB/s 4th Generation PCIe versus 15.8 GB/s 3rd Generation PCIe Gen 3 for x86 from 2010
1.3X more memory capacity: 4TB per system versus 3 TB per x86 server
2X more threads per core than x86
= 2X better price performance
1.8X more data bandwidth: 230 GB/sec per Power9 socket versus to 128GB/sec per socket for x86
2X data throughput: 31.5 GB/s 4th Generation PCIe versus 15.8 GB/s 3rd Generation PCIe Gen 3 for x86 from 2010
1.3X more memory capacity: 4TB per system versus 3 TB per x86 server
2X more threads per core than x86
= 2X better price performance
IBM Linux servers should be the preferred hardware for any enterprise open-source ecosystem for both performance and reducing software licensing costs. Why does the Linux server performance matter? The larger the workload a core or socket can support, the lower the software licenses costs per node will be. If your Linux software is priced per core or socket, the best performing hardware will result in a lower cost of ownership. However, better performing Linux server hardware will also lower other operating costs such as availability, management, integration and scalability, which will also be addressed in this article. Other factors that will contribute to lower costs of ownership and should be included in your calculations are reboot time, patching and time spent trouble shooting performance issues.
IBM Power9 Linux servers do not just outperform x86 servers by a small margin or in a few categories. Power9 Linux server performance benchmarks for IBM Red Hat server (RHEL), Ubuntu server and SuSE Linux Enterprise server (SLES) distributions win across the board for in every key category that affect I/O latency, bandwidth and throughput of interactive and batch processes by a very large margin. Independent testing results show running Linux server workloads on a Power9 Linux server can deliver 75% or more cost savings over the first 5 years in apples to apples comparisons. Power9 servers can run the same x86 Linux server workload with less cores, database licenses and less memory than a x86 Linux server.
Power9 vs x86 cost comparison for Linux server workloads
Power9 per core performance increases about 35% with each new processor generation, whereas x86 performance per core has made very little performance improvements with each new generation. If your company will save money by reducing the number of cores or sockets required to support a Linux server workload, the solution is clear. Every application and workload will vary, but it is not uncommon to replace 10 or more x86 server cores with a single Power9 core. Simple math can be done in your head to realize the core-based licensing, support and administration savings. You will likely need a spreadsheet to calculate the power and footprint savings, but the economies of scale will certainly contribute to higher cost savings in the millions for larger the environments.
A Linux server’s main memory speed and bandwidth have a significant impact on application response times, however x86 servers have not made any significant advancements in this area. Processor manufactures attempt to offset this hurdle is to use various levels of cache of a higher speed than the memory to lower latency. Level 1 cache will typically have the most significant impact on application performance, which is limited to 64 bytes for x86 processors. A Power9 Linux server support up to 2x more RAM (4 TB) and will have 1.8x more memory bandwidth per socket than an x86 server. L1 cache for a Power9 processors is 128 bytes, which is twice that of an x86 server.
Most newer and modern applications support multi-threading (SMT), but Linux server applications running on x86 systems can only support up to two threads per core, where as an IBM Power9 can support up eight threads per core. What number of threads does your application support and perform optimally? Although Power9 servers allows customers to assign every core in the system to run single-threaded, there are significant performance benefits to be gained by taking advantage of multi-threading for latency and throughput. When applications can run multiple threads within programs on the same core, the processor and related resources are better utilized and perform more efficiently. The typical workload will generally have higher throughput per core when running in a multithreaded mode.
On average, companies refresh x86 servers on a three to four-year cycle due to unplanned and planned downtime events or administration costs associated with older hardware. In comparison, IBM POWER for Linux customers refresh their system(s) every four to five years on average, which means a 30% reduction in refresh cycles. Now, factor in the downtime costs for replacing each server.
Today, reliability and security can in some ways use a single line item for assessing costs. On average, a Power9 Linux server will suffer a maximum of two minutes of unplanned downtime per year, compared to x86 servers which range from two minutes to 47 minutes per year. Security is another important factor that must be considered. The Power9 Linux server uses PowerVM as its virtualization engine and has had zero reported security vulnerabilities. The Power9 Linux server hardware and entire framework was designed and built with security and reliability, and ships fully secured and ready for deployment as soon as you plug it in. Read about IBM Power Linux Server reliability test results.
How many security vulnerabilities have been reported for your hypervisor? According to NIST’s National Vulnerability Database, some of the most common hypervisors have reported well over 100 vulnerabilities, see your yourself: https://nvd.nist.gov/vuln/search After you have completed that search, investigate all the other hardware and layers of vulnerabilities that exist within your x86 server that need to be patched regularly. The Power9 hypervisor is a firmware-based virtualization layer and has had ZERO reported vulnerabilities, whereas VMware, x86 server’s most popular virtualization software has had 100s of reported vulnerabilities. Power9 processors have accelerated encryption built into the chip that helps protect data at rest and motion. Read more about IBM Linux server security architecture.
Speaking of security vulnerabilities, keeping up with patching your database is critical, but can be exhausting and expensive. Have you analyzed the number the times you had to apply an Oracle, MySQL or MS SQL database security patch? https://www.oracle.com/security-alerts/
How often do you have to reboot your Linux server? Companies have to reboot their x86 Linux servers regularly due to patching, performance issues and other maintenance tasks, which can add up to hours of unplanned downtime per year. How much downtime have your x86 Linux servers cost your company? IBM Power9 Linux servers deliver 99.9996% uptime, which his roughly 2 minutes of unplanned downtime per year.
IBM Power9 and Power8 servers are available for both scale up and scale out environments, and support the most commonly Linux distributions used by corporations: IBM Red Hat Linux Enterprise Server (RHEL), Ubuntu, SuSE Linux Enterprise server (SLES) and others. Customer’s may choose to purchase their desired OS and support from any Linux distribution vendor directly or via their IBM Business Partner for turnkey maintenance and support services. IBM Power System customer can also benefit from the speed and simplicity of using the Red Hat OpenShift Container platform for all cloud deployments, regardless of their Linux server hardware vendor or cloud deployment type. You can also check to see if your open source package is available on Power Linux Servers using IBM’s Open Source POWER Availability Tool (OSPAT), BioBuilds, Jenkins, Docker Hub, rpmfind or GitHub. View the IBM Power9 Linux Server Line.