Discover the Key to Cloud TCO

Have you been considering cutting the cable cord? Ads are everywhere, promoting amazing deals for new phones or video service providers that are hoping you’ll ditch your cable company. These deals sound great but when you consider the total cost of chasing down that great introductory offer, it generally isn’t the deal you were hoping for.

This is a classic Total Cost of Ownership (TCO) analysis for me. The phone or TV service provider offers a low acquisition cost, but the cost-benefit math doesn’t pencil out when all my adjacent costs are considered, such as any early termination penalty, time and hassle, new equipment, waiting for the installer, not to mention what happens to the rate with the introductory offer runs out.

TCO is critical when choosing a cloud instance for your applications. Recently, several cloud providers introduced instances based on competing platforms at prices that, at first look, seem like a great way to save some money. But let’s look at several considerations that should be part of your “full TCO” decision.

Instance Performance-per-Dollar

The hardware underpinning of your cloud instance makes a huge difference in the performance of your applications and services. The reduced acquisition cost with a cheaper instance is tempting, but it isn’t a good investment if performance declines by a greater percentage than you save. If you pay 10% less for the instance, but lose at least 20% performance, is that really a good investment when your customers and employees depend on your digital services? Intel Xeon processor-based platforms are the most widely deployed hardware in the cloud due to their high-performance cores, broad software optimizations and compatibility, and 20 year history of delivering leadership processors into the market. We encourage our customers to run proof-of-concepts or benchmark tests to see for themselves which instance delivers the better performance for their dollars invested. A recent press article compared Intel’s AWS instances to AMD’s and determined that our instances “offer both higher value and absolute performances across almost all cloud use cases.” Across every workload tested, Amazon’s instance powered by Intel Xeon Processors outperformed the AMD-based instance. When comparing performance per dollar, the Intel instance provided higher value across 11 out of the 11 workloads tested, with up to 4x with one HPC workload.1

Virtual Machine Portability

A cloud instance is usually a virtual machine (VM) that is, at least in theory, completely abstracted from dependencies on the underlying hardware, and thus portable across platforms. In practice, VMs are dependent on the hardware for performance-accelerating features and instructions. A VM running on Platform A cannot generally be transferred to Platform B without a time-consuming conversion process, and certainly can’t be live-migrated across platforms without service interruption. Your cloud TCO model should account for the cost of converting existing VMs on Intel platform to run on non-Intel instances, plus the cost or limitations if you want to move that workload back to Intel in your private cloud or another public provider.

Instance Availability

Several of the new instances on competing platforms are only available in certain countries or regional data centers. In the cloud, you can (again, theoretically), deliver global services from anywhere, but in reality, transit latency and data locality regulations make local presence in many geographies a must. To minimize TCO at worldwide scale, the best strategy is to replicate identical VMs in identical instances wherever a local presence is required to meet service level agreements or comply with applicable laws. The cost of creating and maintaining both Intel and non-Intel VM versions of the same service in non-uniform instances may offset any cost savings associated with instance price.

Software Optimizations and Compatibility

Intel engages the world’s cloud providers and software leaders to enable virtualization environments that expose the Intel hardware’s features directly to applications with minimal abstraction by the hypervisor. Consequently, application vendors tune their software for features like Intel AVX-512, Intel Turbo Boost, and Intel Software Guard Extenstions, even though they know they will operate inside a VM. Workloads running in Intel-based instances can take advantage of these features to increase performance and security, but those software-optimization gains may be lost if the VM is transferred to a non-Intel instance. The impact of foregoing software optimization gains should be included in any TCO analysis, as it may affect the performance-per-dollar ratio, and reduce service levels.

Deal, or Not So Much of a Deal?

As with cell phones and cable TV, that initial low-cost offer can sound great, but the full picture for most cloud users is more complex and nuanced than the marketing pitch. Competition is good because it keeps the vendors sharp and gives customers choices. As you consider your instance choices from your cloud providers, please consider all the factors that roll up into the full TCO to help you find the right platform and cloud type for your needs. Learn more about Intel’s support for CSPs at intel.com/csp.


1 Results as of January 4th, 2019

Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more complete information visit http://www.intel.com/benchmarks .

Intel's compilers may or may not optimize to the same degree for non-Intel microprocessors for optimizations that are not unique to Intel microprocessors. These optimizations include SSE2, SSE3, and SSSE3 instruction sets and other optimizations. Intel does not guarantee the availability, functionality, or effectiveness of any optimization on microprocessors not manufactured by Intel. Microprocessor-dependent optimizations in this product are intended for use with Intel microprocessors. Certain optimizations not specific to Intel microarchitecture are reserved for Intel microprocessors. Please refer to the applicable product User and Reference Guides for more information regarding the specific instruction sets covered by this notice.

Cost reduction scenarios described are intended as examples of how a given Intel-based product, in the specified circumstances and configurations, may affect future costs and provide cost savings. Circumstances will vary. Intel does not guarantee any costs or cost reduction.

Intel® Advanced Vector Extensions (Intel® AVX)* provides higher throughput to certain processor operations. Due to varying processor power characteristics, utilizing AVX instructions may cause a) some parts to operate at less than the rated frequency and b) some parts with Intel® Turbo Boost Technology 2.0 to not achieve any or maximum turbo frequencies. Performance varies depending on hardware, software, and system configuration and you can learn more at http://www.intel.com/go/turbo.

Intel does not control or audit third-party benchmark data or the web sites referenced in this document. You should visit the referenced web site and confirm whether referenced data are accurate.

© 2019 Intel Corporation.

 Intel, the Intel logo, and Intel Xeon are trademarks of Intel Corporation in the U.S. and/or other countries.

*Other names and brands may be claimed as property of others.

Performance testing done on AWS EC2 M and R instances (https://aws.amazon.com/ec2/instance-types/). Database Benchmarks show R5 vs. R5a due to higher memory capacity.

Memory capacity for instances: m5.24xlarge and m5a.24xlarge: 384GB; r5.24xlarge and r5a.24xlarge: 768GB.

For details on EC2 instance protections for various vulnerabilities including side-channel, please refer to Amazon security bulletins: https://aws.amazon.com/security/security-bulletins/

For more complete information about performance and benchmark results, visit www.intel.com/benchmarks. Intel does not control or audit the design or implementation of third party benchmark data or Web sites referenced in this document. Intel encourages all of its customers to visit the referenced Web sites or others where similar performance benchmark data are reported and confirm whether the referenced benchmark data are accurate and reflect performance of systems available for purchase. Any differences in your system hardware, software or configuration may affect your actual performance.

Estimated SPEC*rate2017_int_base (higher is better):

  • AWS M5.4xlarge (Intel) Instance, SPECrate2017_int_base CPU1.0.2, Intel ICC Version 18.0.2 20180210, RedHat Enterprise Linux 7.5, Kernel 3.10.0-862.el7.x86_64, 16 copies, Estimated Score 46, measured by Intel on 12/6/18
  • AWS M5a.4xlarge (AMD) Instance, SPECrate2017_int_base CPU1.0.2, AOCC1.0/LLVM, RedHat Enterprise Linux 7.5, Kernel 3.10.0-862.el7.x86_64, 16 copies, Estimated Score 30.5, measured by Intel on 12/6/18

Estimated SPEC*rate2017_int_base 1 copy (higher is better):

  • AWS M5.4xlarge (Intel) Instance, SPECrate2017_int_base CPU1.0.2, Intel ICC Version 18.0.2 20180210, RedHat Enterprise Linux 7.5, Kernel 3.10.0-862.el7.x86_64, 1 copy, Estimated Score 5.23, measured by Intel on 12/6/18
  • AWS M5a.4xlarge (AMD) Instance, SPECrate2017_int_base CPU1.0.2, AOCC1.0/LLVM, RedHat Enterprise Linux 7.5, Kernel 3.10.0-862.el7.x86_64, 1 copy, Estimated Score 4.06, measured by Intel on 12/6/18

Estimated SPEC*rate2017_fp_base (higher is better):

  • AWS M5.4xlarge (Intel) Instance, SPECrate2017_fp_base CPU1.0.2, Intel ICC Version 18.0.2 20180210, RedHat Enterprise Linux 7.5, Kernel 3.10.0-862.el7.x86_64, 16 copies, Estimated Score 57.7, measured by Intel on 12/6/18
  • AWS M5a.4xlarge (AMD) Instance, SPECrate2017_fp_base CPU1.0.2, AOCC1.0/LLVM, RedHat Enterprise Linux 7.5, Kernel 3.10.0-862.el7.x86_64, 16 copies, Estimated Score 27.7, measured by Intel on 12/6/18

Estimated SPEC*rate2017_fp_base 1 copy (higher is better):

  • AWS M5.4xlarge (Intel) Instance, SPECrate2017_fp_base CPU1.0.2, Intel ICC Version 18.0.2 20180210, RedHat Enterprise Linux 7.5, Kernel 3.10.0-862.el7.x86_64, 1 copy, Estimated Score 7.65, measured by Intel on 12/6/18
  • AWS M5a.4xlarge (AMD) Instance, SPECrate2017_fp_base CPU1.0.2, AOCC1.0/LLVM, RedHat Enterprise Linux 7.5, Kernel 3.10.0-862.el7.x86_64, 1 copy, Estimated Score 4.88, measured by Intel on 12/6/18

Memory Bandwidth – Stream Triad (higher is better):

  • AWS M5.4xlarge (Intel) Instance, McCalpin Stream (OMP version), (Source: https://www.cs.virginia.edu/stream/FTP/Code/stream.c); Intel ICC 18.0.3 20180410 with AVX512, -qopt-zmm-usage=high, -DSTREAM_ARRAY_SIZE=134217728 -DNTIMES=100 -DOFFSET=0 –qopenmp, -qopt-streaming-stores always -o $OUT stream.c, RedHat Enterprise Linux 7.5, Kernel 3.10.0-862.el7.x86_64, OMP_NUM_THREADS : 8, KMP_AFFINITY : proclist=[0-7:1], granularity=thread, explicit, Score 81216.7 MB/s, measured by Intel on 12/6/18
  • AWS M5a.4xlarge (AMD) Instance, McCalpin Stream (OMP version), (Source: https://www.cs.virginia.edu/stream/FTP/Code/stream.c); Intel ICC 18.0.3 20180410 with AVX2, -DSTREAM_ARRAY_SIZE=134217728, -DNTIMES=100 -DOFFSET=0 -qopenmp -qopt-streaming-stores always -o $OUT stream.c, RedHat Enterprise Linux 7.5, Kernel 3.10.0-862.el7.x86_64, OMP_NUM_THREADS : 8, KMP_AFFINITY : proclist=[0-7:1], granularity=thread,explicit, Score 32154.4 MB/s, measured by Intel on 12/6/18

HPC Materials Science – LAMMPS (higher is better):

  • AWS M5.4xlarge (Intel) Instance, LAMMPS version: 2018-08-22 (Code: https://lammps.sandia.gov/download.html), Workload: Water – 512K Particles, Intel ICC 18.0.3.20180410, Intel(R) MPI Library for Linux* OS, Version 2018 Update 3 Build 20180411, 8 MPI Ranks, RedHat Enterprise Linux 7.5, Kernel 3.10.0-862.el7.x86_64, OMP_NUM_THREADS=2, Score 25.603 timesteps/sec, measured by Intel on 12/6/18
  • AWS M5a.4xlarge (AMD) Instance, LAMMPS version: 2018-08-22 (Code: https://lammps.sandia.gov/download.html), Workload: Water – 512K Particles, Intel ICC 18.0.3.20180410, Intel(R) MPI Library for Linux* OS, Version 2018 Update 3 Build 20180411, 8 MPI Ranks, RedHat Enterprise Linux 7.5, Kernel 3.10.0-862.el7.x86_64, OMP_NUM_THREADS=2, Score 9.185 timesteps/sec, measured by Intel on 12/6/18

Changes for AMD to support AVX2 (AMD only supports AVX2, so these changes were necessary):

               sed -i 's/-xHost/-xCORE-AVX2/g' Makefile.intel_cpu_intelmpi

               sed -i 's/-qopt-zmm-usage=high/-xCORE-AVX2/g' Makefile.intel_cpu_intelmpi

HPC Linpack (higher is better):

  • AWS M5.4xlarge (Intel) Instance, HP Linpack Version 2.2 (https://intel.com/en-us/articles/intel-mkl-benchmarks-suite Directory: benchmarks_2018.3.222/linux/mkl/benchmarks/mp_linpack/bin_intel/intel64), Intel ICC 18.0.3.20180410 with AVX512, Intel(R) MPI Library for Linux* OS, Version 2018 Update 3 Build 20180411, RedHat Enterprise Linux 7.5, Kernel 3.10.0-862.el7.x86_64, OMP_NUM_THREADS=8, Score 566.2 GB/s, measured by Intel on 12/6/18
  • AWS M5a.4xlarge (AMD) Instance, HP Linpack Version 2.2, (HPL Source: http://netlib.org/benchmark/hpl/hpl-2.2.tar.gz; Version 2.2; icc (ICC) 18.0.2 20180210 used to compile and link to BLIS library version 0.4.0; https://github.com/flame/blis; Addt’l Compiler flags: -O3 -funroll-loops -W -Wall –qopenmp; make arch=zen OMP_NUM_THREADS=8; 6 MPI processes.), Intel ICC 18.0.3.20180410 with AVX2, Intel(R) MPI Library for Linux* OS, Version 2018 Update 3 Build 20180411, RedHat Enterprise Linux 7.5, Kernel 3.10.0-862.el7.x86_64, OMP_NUM_THREADS=8, Score 123.9 GB/s, measured by Intel on 12/6/18

Web Front End WordPress (higher is better):

  • AWS M5.4xlarge (Intel) Instance, oss-performance/wordpress Ver 4.2.0; Ver 10.2.19-MariaDB-1:10.2.19+maria~bionic; Workload Version': u'4.2.0; Client Threads: 200; PHP 7.2.12-1; perfkitbenchmarker_version="v1.12.0-944-g82392cc; Ubuntu 18.04, Kernel Linux 4.15.0-1025-aws, Score 683.68 TPS, measured by Intel on 12/7/18
  • AWS M5a.4xlarge (AMD) Instance, oss-performance/wordpress Ver 4.2.0; Ver 10.2.19-MariaDB-1:10.2.19+maria~bionic; Workload Version': u'4.2.0; Client Threads: 200; PHP 7.2.12-1; perfkitbenchmarker_version="v1.12.0-944-g82392cc; Ubuntu 18.04, Kernel Linux 4.15.0-1025-aws, Score 466.71 TPS, measured by Intel on 12/7/18

Server Side Java (higher is better):

  • AWS M5.4xlarge (Intel) Instance, Java Server Benchmark No NUMA binding, 1JVM, OpenJDK 10.0.1, RedHat Enterprise Linux 7.6, Kernel 3.10.0-957.el7.x86_64, Score 19152 Transactions/sec, measured by Intel on 12/7/18
  • AWS M5a.4xlarge (AMD) Instance, Java Server Benchmark No NUMA binding, 1JVM, OpenJDK 10.0.1, RedHat Enterprise Linux 7.6, Kernel 3.10.0-957.el7.x86_64, Score 14911 Transactions/sec, measured by Intel on 12/7/18

Database: HammerDB – PostgreSQL (higher is better):

  • AWS R5.4xlarge (Intel) Instance, HammerDB 3.0 PostgreSQL 10.2, Memory: 128GB, Hypervisor: KVM; Storage Type: EBS io1, Disk Volume 200GB, Total Storage 200GB, Docker version: 18.06.1-ce, RedHat Enterprise Linux 7.6, 3.10.0-957.el7.x86_64, 6400MB shared_buffer, 256 warehouses, 64 users. Score “NOPM” 207192, measured by Intel on 12/7/18
  • AWS R5a.4xlarge (AMD) Instance, HammerDB 3.0 PostgreSQL 10.2, Memory: 128GB, Hypervisor: KVM; Storage Type: EBS io1, Disk Volume 200GB, Total Storage 200GB, Docker version: 18.06.1-ce, RedHat Enterprise Linux 7.6, 3.10.0-957.el7.x86_64, 6400MB shared_buffer, 256 warehouses, 64 users. Score “NOPM” 142494, measured by Intel on 12/7/18

Database: MongoDB (higher is better):

  • AWS R5.4xlarge (Intel) Instance, MongoDB v4.0, journal disabled, sync to filesystem disabled, wiredTigeCache=27GB, maxPoolSize = 256; 7 MongoDB instances, 14 client VMs, 1 YCSB client per VM, 96 threads per YCSB client, RedHat Enterprise Linux 7.5, Kernel 3.10.0-862.el7.x86_64, Score 284014.51 ops/sec, measured by Intel on 12/13/18
  • AWS R5a.4xlarge (AMD) Instance, MongoDB v4.0, journal disabled, sync to filesystem disabled, wiredTigeCache=27GB, maxPoolSize = 256; 7 MongoDB instances, 14 client VMs, 1 YCSB client per VM, 96 threads per YCSB client, RedHat Enterprise Linux 7.5, Kernel 3.10.0-862.el7.x86_64, Score 142169.17 ops/sec, measured by Intel on 12/13/18
Published on Categories Cloud ComputingTags , , , , ,
Jeff Wittich

About Jeff Wittich

Jeff is Intel's Director of Cloud Service Provider (CSP) Business Strategy and Product Enabling team in the Data Center Group and is responsible for setting global strategic initiatives in order to accelerate cloud growth and deliver innovative platforms to CSPs. Over the last 14 years, Jeff has held a wide range of roles at Intel across engineering, management, and leadership, including product development for 5 generations of Intel® Xeon® processors. He holds a Bachelor of Science degree in Electrical Engineering from the University of Notre Dame and a Master in Science degree in Electrical and Computer Engineering from the University of California, Santa Barbara. This unique background has given him a wealth of knowledge which he now leverages to drive Intel's platform and business strategy for the fast growing CSP market.