Is Power9 the best performing Linux server? Q&A With IBM Gold Partner, Midland Information Systems plus real world examples.

Q: What applications can I run on a IBM Power Linux Server?

A: Any x86 Linux server application running on Ubuntu Server, SUSE Linux Enterprise Server and RedHat Enterprise Linux is supported. In addition, the community versions of Linux server like Debian, openSUSE, CentOS, Fedora and others can also run on IBM Power Linux servers.

Q: Does an IBM Power9 Linux server cost more than an x86 server?

A: Yes, maybe and no. When comparing the upfront hardware costs of purchasing a IBM Power Linux server solution to a comparable x86 server solution, x86 hardware may be at times less expensive and sometimes not. However, after accounting for lower licensing costs, reliability, implementation time, scalability, performance, ongoing management and security, IBM Power9 Linux servers have a much lower cost of ownership compared to alternative x86 servers.

Q: Why does a IBM Power Linux server have a lower TCO compared to an x86 server?

A: IBM Power systems provide better performance per core, which means fewer cores needed to run applications, so software licensing costs will be lower. Fewer server environments require less system management, database administration and other labor costs, which lowers operating costs. Other cost saving benefits include smaller physical foot print and lower power consumption. Although an x86 server may have a lower upfront acquisition costs, industry testing and case study results prove this does not equate to a lower cost solution. On average, businesses can lower costs by 57% when migrating x86 workloads to a IBM Power9 Linux server.

Q: Which Linux server is the most reliable?

A: Without a doubt, it is the IBM Power Linux server line. IBM Power servers have a 99.9996% uptime score (2.1 minutes/server/annum unplanned downtime), and have proven to be the most reliable servers on the market for the past 10 years in a row. In addition, IBM Power Linux servers have the fastest shut down and boot up time. ITIC ranked IBM Power systems number 1 in every major reliability category for on premises infrastructure. Read the Linux server reliability testing results for yourself.

Q: Is application performance better on IBM Power9 Linux server versus x86 servers?

A: Yes, absolutely. Testing and business case studies consistently prove IBM Power9 Linux servers process data faster than x86 servers, using less cores and in a much smaller foot print. In the application server space, a 2-socket Power9 Linux server has 3.4x faster core performance and 2.4X the price-performance with two times more threads compared to an x86 server. IBM Power9 servers have up 5.6x more I/O, 9.5x I/O bandwidth, almost 5.6x CPU to GPU bandwidth, support up to 2x more RAM, 1.8x more memory bandwidth per socket, 5.6x data throughput 4x as many threads. Many of the Power9 server performance benefits also cascade to other areas of application performance and system resource utilization. IBM Power Linux server memory is larger and has much larger cache, which translates to lower latency and less I/O, faster shut down and startup. FYI, the world’s fastest supercomputers are running on a IBM Power9 processor.

Q: How many IBM Power9 Linux servers would I need to replace my x86 server farm?

A: Most environments would only need 1 IBM Power9 Linux server to replace an x86 cluster or server farm, but a workload analysis will determine the proper sizing of a Power9 replacement solution. A IBM Power9 server can have up to 192 cores. Below are some real world Power9 migration examples, illustrating the cost saving benefits by only calculating the initial server and software licensing costs over 3 years.

EXAMPLE #1 : Power9 Migration with MongoDB 700 user workload running 1000 transactions.

2-socket Intel Xeon Skylake Gold 6150 (2x18-core)

• 256 GB memory
• 2 x 480 GB SSD
• 3 x 10 Gb two-port network
• IBM Red Hat Linux RHEL 7.6
• KVM (2 VMs with 18-cores each) and 98 containers

would require $28,805 in upfront costs for the x86 server and $171,200 for licensing RHEL OS, Virtualization and OCP.

The equivalent IBM Power9 L922 (2x10-core)

• 256 GB memory
• 2x 388 GB SSD
• 2x 10 Gb two-port network
• IBM Red Hat Linux 7.6

with PowerVM (2 partitions with10-cores each) and 174 containers for $28,821 in upfront costs and $115,900 for licensing.

Results? $1,081 cost savings per container running on the IBM Power9 with additional significant savings for core software licenses.

The result is a $1,081 cost savings per container running on the IBM Power9 server, allowing 1.7x more containers (3.2x more containers per core) and 2.6 better price performance than a comparable x86 server. This company can now scale out MongoDB workload using half the number of x86 servers, providing significant savings for core based software licenses.

A CPU intensive private cloud app environment achieves the best Java benchmark score

...with half the application server cores and a two thirds less database server cores running on:

IBM Power9 S924 2.9GHz

• 24-core and a S924 3.3GHz
• 16-core server with 256 GB of memory each

Delivering 2.4x better per core performance for a cost of $218,000.

Compared to:

Cisco UCS C240 M5 2xIntel Xeon Platinum 8168 (2.7GHz, 48-core)

• with 256 GB of memory
• and Cisco UCS C240 M5 2xIntel Xeon Platinum 8168 (2.7GHz, 48-core)
• with 1TB of memory for $358,000.

The Java test results show the IBM Power9 servers getting 34,517 WebjOPS (1,438 per core) and the Cisco server achieving only 28,860 (601 per core).

As you can see from the above results, IBM Power Servers thread count enable more containers per server. In CPU demanding workloads, x86 servers hits a maximum container count faster (72 containers) than Power9 (more than double x86) and has 15% less throughput than Power9. If these environments need additional cores, the x86 costs will continue to increase for the server to reach equivalent throughput of the Power9.

 EDB pgbench mark

Using the EDB pgbench mark, 853,709 transactions per second were achieved on a EnterpriseDB Postgres Advance Server 10 workload running on:

IBM Power9 L922 (2x10-core/2.9 GHz)

• 256 GB memory
• 2 x 300GB SATA
• 7.2K rpm LFF HDD
• 10 Gb two-port
• 1 x 16gbps FCA
• EDB Postgres Advanced Server 10
• RHEL 7.5 with PowerVM (4 partitions with 5-cores each)

cost $27,480 and $144,557 for IBM Red Hat Linux RHEL OS and Virtualization.

Compared to 611,118 transactions per second on:

2-socket Intel Xeon Skylake Gold 6148 (2x20-core/2.4 GHz

• 256 GB memory
• 2 x 300GB HDD
• 1 Gb two-port
• 1 x 16gbps FCA
• EDB Postgres Advanced Server 10
• IBM Red Hat Linux RHEL 7.5
• KVM (4 VMs with 10-cores each)

costing $29,247 and licensing of $243,166. The IBM Power9 server delivered 2.8x per core better performance, 1.4x system throughput and at 40% less cost. The IBM Power9 server delivered 2.4x better price performance than the x86 server.

 Cassandra on x86 versus ScyllaDB on Power9 stress test using Gaussian 9M

A Cassandra on x86 versus ScyllaDB on Power9 stress test using Gaussian 9M, 4.5M and 10K models with 80% reads and 20% writes shows IBM Power solution provided 216% more performance per system or a 3.9x price performance. The IBM Power9 solution used a LC922 (2x22-core/2.6 GHz/256 GB memory), with 2 x internal HDD, 40 GbE, 1 x 1.6TB NVMe adapter running Scylla Enterprise 2018.1.0 on IBM Red Hat Linux RHEL 7.5 LE cost $25,615. The Power9 solution delivered 906,463 operations per second performance or 35 Ops per second per $. The 2-socket Intel Xeon SP (Skylake) Gold 6140 (2x18-core/2.3 GHz/256 GB memory) using 2 x internal HDD, 40 GbE, 1 x 1.6TB NVMe adapter running Open Source Cassandra 3.11.2 on IBM Red Hat Linux RHEL 7.5 cost $31,373. The x86 solution delivered 286,627 operations per second performance or 9 Ops per second per $.

15 server rack environment with a VM image with containers bound to a socket including a MongoDB micro servic

A 15 server rack environment with a VM image with containers bound to a socket including a MongoDB micro service, RHEL OS + Virtualization + ICP Cloud Native VPC running on IBM Power9 L922 (2x8-core/3.4 GHz/256 GB memory), 2 x 600GB SATA 7.2K rpm LFF HDD, 10 Gb two-port, 1 x 16gbps FCA, EDB Postgres Advanced Server 10, IBM Red Hat Linux RHEL 7.4 with PowerVM (2 partitions with 8-cores each) cost $2,700,735, achieved 548,490 transactions per second using 2 VMs. The 2-socket Intel Xeon Skylake Gold 6130 (2x20-core/2.1 GHz/256 GB memory), 2 x 600GB SATA 7.2K rpm LFF HDD, 1 Gb two-port, 1 x 16gbps FCA, IBM Red Hat Linux RHEL 7.4, KVM (2 VMs with 16-cores each) cost $4,815,285, achieved 589,680 transactions per second using 2 VMs. The IBM Power Linux server solution delivers 80.6 more transactions per second per $1,000 spent, and 1.86x more container throughput per core compared to the Intel based x86 server.

Read about more business case reasons to switch to Power9 for Linux server workloads at Power9 Linux Server case studies. Contact us for specific benchmark or workload type you would like to see performance and sizing information on.

Q: Are IBM Power Linux servers more secure than x86 servers?

A: IBM Power Systems are built with security at every level, from processor to the OS, and deliver end-to-end security out of the box. All IBM Power9 Linux servers are delivered with pre-loaded firmware and operating system security patches that mitigate known vulnerabilities and threats such as Meltdown and Spectre, which include all supported OS types: Linux, AIX and IBM i OS400. The IBM Power server holistic security measures include the hardware, hypervisor, firmware, operating system, applications, network resources and security software.

Q: Are IBM Power Linux servers more scalable than x86 servers?

A: Yes, there is no contest. Please see the chart of the IBM Power Linux Server line, and let us know if you would like to compare server pricing for an x86 replacement: Power9 Linux Server Models Specifications

Q: Is migrating to IBM Power Linux Server difficult?

A: In most cases, no. After recompiling your x86 code, you will be ready to go without making any additional changes. Some customers have claimed over 8x performance improvement immediately upon migration. Below are some real life examples and case studies.

Linux to IBM Power Migration Case Studies

Earth Signal Processing LTD applications are written in very simple C and C++, composed of approximately two million lines of code (800 programs and about 150 libraries), and it took about an hour to recompile on IBM Power.

The software immediately worked without any other changes, and saw a 100 percent performance improvement right out of the box, took 1/10 the x86 rack space, and saw over 40 percent decrease in power consumption.

LiveMon’s application is written in FPC (Free Pascal Compiler) and assembly language, composed of approximately 200,000 lines of code, and recompiled in under two hours and did not require modifications to run on IBM Power. Their application immediately saw a 2x performance improvement over the x86 server environment, and got an additional 8x performance improvement after performing some optimization.

Vuble runs web services on Ubuntu Linux and uses a basic LAMP (Linux, Apache, MySQL, PHP) stack that has precompiled packages available for PPC64 Linux distributions, so they did not have to recompile any code. All they had to do was move files from the x86 server to the IBM Power server. Vuble reduced their server foot print by 5x, achieved a 15% reduction in response times and performs 94% less server administration.

A retail chain with a hybrid database environment MongoDB, Redis and DB2 and a batch application that consumed all available cycles on the system needed to dramatically improve application response times. After the databases and application was migrated to the IBM Power server, it executed 12X faster than the x86 Linux server, utilizing only 10% of the processor, while taking the x86 batch job down from 24 hours to less than one hour on the IBM Power server. The database migration to the IBM Power Server involved using the MongoDB dump/restore program, which took one business day to complete.

Q: Is there a guide to help migrate x86 workloads to a IBM Power Linux server?

A: Yes. Below are a few links to help you migrate your x86 applications to IBM Power-based platforms:

• https://www.ibm.com/support/knowledgecenter/linuxonibm/liabw/liabwp9qsg_8335.htm
• https://developer.ibm.com/linuxonpower/porting-guide/
• https://developer.ibm.com/tutorials/l-porting-tuning-tools/
• https://developer.ibm.com/linuxonpower/sdk/
• https://developer.ibm.com/linuxonpower/tutorials/sdk_linux_on_power/

Send Us Your Server Workload For Analysis.

A Small company may save thousands per year.
While a larger company can save millions of dollars annually.

If you have not seen enough evidence to build your business case to switch to a IBM Power Linux server, please request from us supporting documentation that reflects your specific workloads.

Midland also provides a detailed workload analysis, illustrating how your applications will run on a comparable Power system, whether it be for Linux, AIX (Unix) or on an IBM i system. The bigger your environment, the more money your company can save by switching to IBM Power servers. Small companies may save 10s of thousands per year, while larger companies can save millions per year.

IBM Power Linux Server Financing

We offer very flexible and affordable monthly and quarterly payment options, as well as financing options as low as 0% for 12 months and long term leasing options providing both $1 buyout and fair market value buyout.
ibm-power-linux-server-financing