IBM Systems and Technology Computer Services Technical White Paper Migrating Linux applications from x86 to IBM Power Systems A discussion of best practices Executive summary Contents With IBM® Power Systems™ servers and blades based on the IBM POWER® processor architecture, IBM has significantly elevated 1 Executive summary system performance, throughput and resiliency. These servers are capable of running your choice of operating systems (Linux, IBM AIX® and 2 Best practices for workload IBM i) in whatever combination you choose. Most importantly, the new selection architecture also integrates IBM PowerVM® virtualization to provide 6 Best practices during extreme scalability, flexibility and robustness.1 As a result, enterprise workload migration workloads deployed in PowerVM virtual machines (VMs) not only run faster on POWER-based platforms, but they can also scale further and 6 Capabilities be optimized more efficiently. 6 Valuable intellectual property PowerVM is a complete virtualization solution that is integrated and 10 Conclusion packaged with Power Systems. This is a very robust implementation of virtualization developed by IBM, based on best practices learned over the course of four decades of experience with the IBM mainframe. With each new Power Systems generation, IBM continues to grow its virtualization offerings beyond just the hypervisor, using features such as Live Partition Mobility and Active Memory Sharing. Reducing costs, improving service and managing risk are three focus areas virtualization customers are interested in. Deploying virtualization can maximize scalability to reduce IT costs. IBM Systems and Technology Computer Services Technical White Paper While all Power Systems servers are capable of running Linux, 2 Best practices for workload selection new Linux-only servers —announced in 2012—are optimized Selecting Linux on x86 workloads to migrate to Power Systems for specific Linux-based workloads such as big data analytics, requires an understanding of the value you gain when using a industry application solutions, and open source infrastructure Power System. services. These new servers offer lower acquisition costs, while simultaneously providing more value than solutions deployed 2.1 Linux consistency across platforms on commodity x86 hardware. Best practice: Leverage Linux on x86 compatible with Linux on Power If you manage complex and energy-inefficient x86-based IBM works with Red Hat and SUSE to provide Linux distribu- server farms with each server dedicated to a single application tions that are consistent between x86 and POWER platforms. or operating environment, you can consolidate workloads, Versions of Red Hat Enterprise Linux (RHEL) and SUSE whether dedicated or virtualized, on Linux on x86. Migrating Linux Enterprise Server (SLES) are available on POWER these workloads to Linux on Power servers can help reduce simultaneously with x86, with the same package and driver costs throughout your infrastructure and improve your ability levels. IBM also provides significant differentiation with Linux to meet changing processing demands. on Power using PowerVM virtualization, extensions to Linux reliability, availability, serviceability (RAS), and performance 2 This paper details the IBM recommendations for best practices optimization. in the selection and migration of Linux on x86 workloads and 2.2 Improve performance for virtualized workloads applications to Power Systems servers. Best practices discussed Best practice: Select Linux on x86 virtualized workloads include: and migrate to PowerVM A May 2013 study from the Edison Group3 comparing ●● Optimizing performance and scalability with virtualization on PowerVM and VMware vSphere (4.1 & 5.0) PowerVM-based virtualization. of virtualized benchmark workloads shows Power Systems ●● Choosing when to migrate applications. PowerVM-integrated virtualization technology leads ●● Maximizing costs savings associated with energy, administra- x86-based add-on virtualization, such as VMware vSphere 4 in tion, licensing and maintenance, and capacity on demand. performance. Key findings from the study conclude that ●● Using IBM offerings such as the IBM Migration Factory PowerVM: and IBM Power Systems Rewards program. ●● Developing applications. ●● Performs up to 126 percent better than VMware. ●● Scales to four times more virtual CPUs than VMware The following sections provide numerous best practices and in a VM. references to additional material for your consideration when ●● Scales linearly to use all CPUs, while VMware does not. selecting and migrating Linux on x86 workloads to Linux on Power. 2 IBM Systems and Technology Computer Services Technical White Paper It is clear from the study’s published benchmark results that Many data centers today have small, single-purpose Linux PowerVM on POWER-based platforms not only offers supe- servers. In addition to high energy costs and management rior scalability than VMware vSphere on Intel x86-based challenges, these servers are frequently underutilized yet servers, but it also makes more efficient use of system resources under-perform at peak loads. This creates both efficiency and and imposes a negligible impact on performance. Many of these user satisfaction issues. In the end, these issues cost money. advantages relate to the fact that PowerVM is built directly into Maximize the return on your IT investments (including floor the firmware of all Power Systems servers, as opposed to space and operational costs) by moving workloads from under- x86-based virtualization products, such as VMware vSphere, utilized systems onto a single, larger system. You can signifi- which are typically third-party software add-ons that are sold cantly lower operational costs with Power servers running all and installed separately. Linux applications, while others are running Linux applications alongside IBM AIX and IBM i applications. Some clients are Selecting existing workloads running on x86 servers using increasing utilization by more than 60 percent,5 eliminating the VMware for virtualization can realize significant performance need for multiple, small, single-purpose servers by running improvements and scaling with PowerVM. Up to 65 percent Linux applications on Power. performance improvement, 32 times increase in virtual proces- sors and higher CPU utilization. In some cases, Power Systems can achieve an average invest- ment payback period of 6.3 months6 through reduction of As you will see in the next section, when software licenses and various costs including energy, cooling and management. maintenance costs are based on number of processors assigned Realizing long-term cost savings is the result of the advanced to VMs, savings can be achieved by using PowerVM to increase features of Power Systems. Power is designed to cut costs and processor utilization with fewer processors. For example, a soft- increase system utilization while ensuring applications get the ware savings of up to 50 percent could occur when doubling resources they need—when they need it. Additional cost savings processor utilization on half the processors. can be achieved by maximizing the use of physical resources by sharing processors, memory and I/O between clients’ logical 2.3 Maximize system utilization to reduce partitions within the server. At peak time, Power Systems can system costs even borrow extra capacity from development or test partitions Best practice: Select workloads from underutilized to effectively meet demand. IBM EnergyScale™ technology x86 systems provides functions that help the user understand and control Select a Linux on x86 workload running on multiple servers IBM server power and cooling use. This enables better facility that are underutilized. In many cases, workloads running across planning, provides energy and cost savings, enables peak energy multiple x86-based servers leave the system less than 35 percent utilized.4 Consolidating these workloads to a single Power Systems server can reduce total cost by leveraging the ability of a Power Systems server to run at higher utilization levels. 3 IBM Systems and Technology Computer Services Technical White Paper usage control and increases system availability. The client can 2.5 Learn from experience of others use EnergyScale capabilities to customize the power consump- Best practice: Select high ROI workloads based on tion of their Power processor-based system and tailor it to their experiences of others particular data center needs. IDC published results from research data and analysis demon- strating the high return on investment (ROI) of running Linux 2.4 Improve reliability, availability, serviceability workloads on IBM servers. Customers involved with this study Best practice: Select workloads based on RAS reported a total annual value of USD30,000 per 100 users for requirements benefits migrating to Power Systems and IBM System z® An IDC study of Linux workloads states “…All of these work- servers. Major contributors to the savings were: IT staff loads have high uptime requirements because any downtime productivity increase (14 percent); user productivity increase would impact user productivity and therefore business produc- (25.4 percent); and IT infrastructure cost reduction tivity.”7 We understand that many Linux on x86 workloads are (60.6 percent). Based on this study, selection of workloads considered mission-critical and migrating these workloads takes with high costs in these areas could result in substantial ROI.8 advantage of the high reliability,