SlideShare a Scribd company logo

Building Efficient Edge Nodes for Content Delivery Networks

Download as PPTX, PDF
R
Rebekah Rodriguez

The document discusses the collaboration among Varnish Software, Supermicro, and Intel to improve content delivery network (CDN) performance through efficient edge caching servers. It highlights the rising demands of live streaming and high-definition content, alongside advancements in Intel's Xeon D-1700 and D-2700 processors which provide enhanced performance and power efficiency for CDN applications. Additionally, it emphasizes the importance of optimizing software, hardware, and system design to achieve better sustainability and power utilization in edge computing environments.

Technology
Related topics:
Data Center Overview • Green Technology
1 of 31
Downloaded 23 times
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Building Efficient Edge Nodes for Content Delivery Networks Frank Miller, Chief Technology Officer, Varnish Software Raymond Chung, Director, Product Management, Supermicro Nash Kleppan, Cloud Solutions Engineer, Intel
Data Platforms Group 2 Notices & Disclaimers  Performance varies by use, configuration and other factors. Learn more on the Performance Index site.  Performance results are based on testing as of dates shown in configurations and may not reflect all publicly available ​updates. See backup for configuration details. No product or component can be absolutely secure.  Your costs and results may vary.  Intel technologies may require enabled hardware, software or service activation.  © Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.
Data Platforms Group 3 Agenda • Background • Varnish Software • Supermicro SuperServer System for CDN • Intel® Xeon® D-1700 and D-2700 • Test Environment • Power and Performance Testing • Sizing Considerations • Conclusions
Data Platforms Group 4 Background • There has been an explosion of live streaming consumption, estimated to be nearly 4 billion hours in 20201 • The growth of high definition (1080p) and ultra-high definition (4K) content increases the performance demands on a CDN • The deployment of 5G networks allows users of mobile devices to consume video at higher speed • Performance and optimization depend upon software efficiency and system design • The goal of this Varnish - Supermicro – Intel collaboration on CDN is to improve the performance per TCO and performance per Watt of edge caching servers 1: https://www.digitaltveurope.com/2020/05/14/almost-4-billion-hours-watched-as-live-streaming-industry-benefits-from-lockdown/
Data Platforms Group 5 Trends in Storage Systems • Increased demand for NVMe storage • 25GbE and 100GbE networking at the edge • High-performance workloads on broadly available servers • Workloads include: CDN and scale-out storage • Traditional high performance storage systems • Performance typically prioritized over other concerns • May not be suitable for reduced performance, power constrained deployments • Efficient storage systems • Require appropriately sized server and processor(s) running efficient software. • Must be appropriately sized to intended performance target for optimal power and performance • Watts turn into kilowatts – Need to keep sustainability goals and OpEx in mind.
Varnish Enterprise Frank Miller Chief Technology Officer
Varnish is an HTTP Cache Varnish is infrastructure software. HTTP HTTP Origin
The software core is kept lean and secure with only the necessary features to operate. Implemented in C and multi- threaded for linear scalability across CPU, RAM and I/O resources. A small privileged runtime minimizes the threat surface while worker threads, with limited privileges, perform the actual work with kernel and hardware awareness. Hardware Aware VMODs Varnish Enterprise (core) Varnish is fast with integration of hardware architectures and features (e.g. NUMA and TLS optimization into the Varnish runtime). Varnish is flexible with Varnish Modules (VMODS) for additional features. Implemented in Varnish Control Language (VCL) and compiled to a shared and dynamically linked library. 8 Varnish is Highly Performant, Flexible and Secure
There are key benefits in high performance in compute The Value of High Performance This requires the harmonization of software, processor and system Efficient solutions drive for lower power utilization for the same amount of unit work (e.g. Gbs/Watt, Flops/Watt, etc.) This should be table stakes in the future of compute at scale: • We can do more for less. • We can even show benefit on current hardware, with optimized software, to gain power savings with only a software upgrade. • There is a clear North Star … to Save the Planet
IoT/Embedded & Edge Computing Raymond Chung Director, Product Management raymondchung@supermicro.com Better Faster Greener™ © 2022 Supermicro
Why Supermicro IoT/Embedded & Edge Computing 10/3/2022 Better Faster Greener™ © 2022 Supermicro 11 Better Faster Greener Supermicro building blocks solutions is Better for IoT architecture design Supermicro Data Center Technology empower Faster IoT and Edge Computing Supermicro Edge Computing with Muti-Node System is Greener for environment Supermicro is a global technology leader committed to delivering first to market innovation for Enterprise, Cloud, AI, and 5GTelco/Edge IT Infrastructure.We are transforming into aTotal IT Solutions provider with environmentally-friendly and energy-saving server, AI, storage, IoT, and switch systems, software, and services while delivering advanced high-volume motherboard, power, and chassis products.
Edge Appliances SYS-220HE SYS-110P SYS-E5019A  Built for Net/Comms  2x workloads  Power Needs  Increased internal connections  Redundancy/HA  Expandability  Flexible  Rackmount or standalone  NEBS and harsher certifications  Cost Effective  Enterprise driven  Normalized Workloads  uCPE & Intel Select solutions  SMB capable  Lower power consumption Minimal Port / Throughput Maximum Port / Throughput SYS-E100/302 SYS-510D Better Faster Greener™ © 2021 Supermicro 12 10/3/2022 SYS-110D SYS-110A SYS-120U SYS-210SE SYS-E403 SYS-220U
10/3/2022 Better Faster Greener™ © 2022 Supermicro 13 SYS-110D-4C/8C/16C/20C-FR(A/D)N8TP Intel® Xeon D Compact 1U Embedded System https://www.supermicro.com/en/products/system/IoT/1U/SYS-110D-20C-FRDN8TP Processor Intel® Xeon D-2700 Processor, 4/8/14/16/20 Cores, 65W~125W Memory Capacity DDR4 3200MHz 1DPC, 4 slots RDIMM/LRDIMM, up to 256GB/512GB Expansion Slots 1x PCIe 4.0 x 16 1x M.2 M-Key SATA3.0/PCI-E 3.0 x4, 2242/2280 I/O Ports 1x IPMI dedicated LAN 4x GbE, 2x 10GbE, 2x 25G SFP28 2x USB 3.0 type A 1x VGA Drive Bays 2x Internal 2.5" drive bays Power Supply Redundant 800 W, 100-240 Vac Redundant 600 W, -48 Vdc System Cooling 5x 40x56mm Hot-swap Counter-rotation PWM Fans (one is optional) Motherboard: X12SDV-4/8/16/20C-SPT8F (7.5”x 8.5”) Chassis: 515M-R804/515M-R601 Dimension: W17.2” x D15.7” x H1.7” (437mm x 399mm x 43mm) Intel® Xeon D-2700 4/8/14/16/20 Cores Processor Up to DDR4 3200Mhz 4 channels, 4 slots RDIMM/LRDIMM up to 512GB 1x IPMI dedicated LAN; 4x GbE; 2x 10GbE; 2x 25G SFP28 2x USB 3.0 type A; 1x M.2 M-Key SATA/PCI-E 3.0 x4, 2280 2x SATA/U.2 2.5” drives Key features Applications 1x VGA PCIe 4.0 x16 M-Key M.2 Networking Application, Firewall, uCPE, Telecom Application Xeon D-2700 4/8/16/20C 4x GbE LAN, 2x 10GbE LAN 2x 25G SFP28 1 x IPMI 2x 2.5” drive bays 1 2 3 4 5 6 7
10/3/2022 Better Faster Greener™ © 2022 Supermicro 14 SYS-510D-4C/8C/10C-FN6P Intel® Xeon D Compact 1U Embedded System https://www.supermicro.com/en/products/system/iot/1u/sys-510d-10c-fn6p Processor Intel® Xeon D-1700 Processor, 4/8/10 Cores, 25W/50W/88W Memory Capacity 4 DIMM slots DDR4 2933MHz up to 256GB RDIMM/UDIMM ECC/Non-ECC Expansion Slots 1 x PCIe 4.0 x 16 1x M.2 M-Key SATA3.0/PCI-E 3.0 x4, 2280 1x M.2 E-Key PCI-E 3.0 x1, 2230 1x M.2 B-Key SATA/PCI-E 3.0 x2/USB 3.0, 2280/3042 I/O Ports 1x IPMI dedicated LAN 4x GbE, 2x 25G SFP28 2x USB 3.0 type A 1x VGA Drive Bays 1x Internal 3.5" or 4x Internal 2.5" drive bays Power Supply 200W AC-DC power supply System Cooling 3x 40x28mm 4-PIN PWM fans Motherboard: X12SDV-4C/8C/10C-SP6F (9.0” x 7.25” Flex ATX) Chassis: CSE-505-203B Dimension: W17.2” x D9.8” x H1.7” (437mm x 249mm x 43mm) Intel® Xeon D-1700 4/8/10 Cores Processor Up to 256GB RDIMM or 128GB UDIMM, DDR4 2933MHz in 4 Slots 1x IPMI dedicated LAN, 4x GbE,2x 25G SFP28, 2x USB 3.0 type A 1x M.2 M-Key SATA/PCI-E 3.0 x4, 2280, 1x M.2 E-Key PCI-E 3.0 x1, 2230 1x M.2 B-Key SATA3.0/PCI-E 3.0 x2/USB 3.0, 2280/3042 4x SATA drives Key features Applications 1x VGA PCIe 4.0 x16 M, B, E-Key M.2 Retail Application, Firewall, Gateway, uCPE Xeon D-1700 4/8/10C 4x GbE LAN, 2x 25G SFP28 1 x IPMI 1x 3.5” or 4x 2.5” drive bays 1 2 3 4 5 6 7
DISCLAIMER Super Micro Computer, Inc. may make changes to specifications and product descriptions at any time, without notice. The information presented in this document is for informational purposes only and may contain technical inaccuracies, omissions and typographical errors. Any performance tests and ratings are measured using systems that reflect the approximate performance of Super Micro Computer, Inc. products as measured by those tests. Any differences in software or hardware configuration may affect actual performance, and Super Micro Computer, Inc. does not control the design or implementation of third party benchmarks or websites referenced in this document. The information contained herein is subject to change and may be rendered inaccurate for many reasons, including but not limited to any changes in product and/or roadmap, component and hardware revision changes, new model and/or product releases, software changes, firmware changes, or the like. Super Micro Computer, Inc. assumes no obligation to update or otherwise correct or revise this information. SUPER MICRO COMPUTER, INC. MAKES NO REPRESENTATIONS OR WARRANTIES WITH RESPECT TO THE CONTENTS HEREOF AND ASSUMES NO RESPONSIBILITY FOR ANY INACCURACIES, ERRORS OR OMISSIONS THAT MAY APPEAR IN THIS INFORMATION. SUPER MICRO COMPUTER, INC. SPECIFICALLY DISCLAIMS ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR ANY PARTICULAR PURPOSE. IN NO EVENT WILL SUPER MICRO COMPUTER, INC. BE LIABLE TO ANY PERSON FOR ANY DIRECT, INDIRECT, SPECIAL OR OTHER CONSEQUENTIAL DAMAGES ARISING FROM THE USE OF ANY INFORMATION CONTAINED HEREIN, EVEN IF SUPER MICRO COMPUTER, Inc. IS EXPRESSLY ADVISED OF THE POSSIBILITY OF SUCH DAMAGES. 10/3/2022 Better Faster Greener™ © 2021 Supermicro 15
www.supermicro.com Thank You
Data Platforms Group 17 Intel® Xeon® D-1700 and D-2700 Nash Kleppan, Cloud Solutions Engineer, Intel
Data Platforms Group 18 Intel® Xeon® D-1700 • For CDN applications, these processors offer: • Integrated Ethernet up to 100 Gbps • Increased memory bandwidth (2 or 3 channels DDR4 2933 MT/s) • Support for 16 lanes PCIe Gen4 and 24 lanes HSIO • Increased core count and cache over prior gen. Up to 10 cores. • Intel® Software Guard Extensions (Intel® SGX) for more protected execution. • Intel® QAT Gen 2
Data Platforms Group 19 Intel® Xeon® D-2700 • For CDN applications, these processors offer: • Integrated Ethernet up to 100 Gbps • Increased memory bandwidth (4 channels DDR4 3200 MT/s 1DPC) • Support for 32 lanes PCIe Gen4 and 24 lanes HSIO • Increased core count and cache over prior gen. Up to 20 cores. • Intel® Software Guard Extensions (Intel® SGX) for more protected execution. • Intel® QAT Gen 3
Data Platforms Group 20 Test Configuration: Xeon® D-2700 Nash Kleppan, Cloud Solutions Engineer, Intel
Data Platforms Group 21 Xeon® D-2700 Test Configuration Edge Cache Node CDN Application CPU Content Origin Client (wrk) 100GbE L2 Switch • 1RU server • Sufficient drive fanout • NICs bonded • 50Gbps line rate • Supermicro SuperServer SYS- 110D-8C-FRAN8TP • 1x Intel® Xeon® D-2733NT • 4x 32GB DDR4 3200 (128GB) • Integrated 2x25GbE • 2x Intel® SSD D7-P5510 3.84TB NVMe 25GbE
Data Platforms Group 22 Xeon® D-2700 Test Configuration - Varnish Edge Cache Node Varnish CPU Content Origin Client (wrk) 100GbE L2 Switch • Supermicro SYS-110D-8C- FRAN8TP • 1x Intel® Xeon® D-2733NT • 4x 32GB DDR4 3200 (128GB) • Integrated 2x25GbE • 2x Intel® SSD D7-P5510 3.84TB NVMe • Ubuntu 22.04 • Varnish-Plus version 6.0.10r1 • OpenSSL 3.0.2 • Power consumption at both PSU inlets measured with Raritan PX3 PDU. 25GbE
Data Platforms Group 23 Benchmarking Results: Intel Xeon® D-2700 Nash Kleppan, Cloud Solutions Engineer, Intel
Data Platforms Group 24 Active-Idle Total System Power Measurements Power is measured at PDU outlets over 5 minutes. This measurement includes everything from the CPU to the LED status lights. See backup (slide 28) for workloads and configurations. Results may vary​. Edge Cache Node • Dedicated single socket server • Sufficient drive fanout • NICs bonded • 50Gbps line rate • Supermicro SYS-110D-8C-FRAN8TP • Ubuntu 22.04 • Using Intel_pstate CPU Scaling Driver • 90 watt average active-idle power in powersave mode • 113 watt average active-idle power in performance mode
Data Platforms Group 25 Xeon® D-2733NT Video on Demand Test Results Up to 368Mb/s per watt. Power is measured at PDU outlets over duration of test. This measurement includes everything from the CPU to the LED status lights. Video-on-Demand, 100% TLS, 100% Cache Hit Rate, uniform random distribution, 315GB dataset on NVMe See backup (slide 28) for workloads and configurations. Results may vary​. Edge Cache Node Varnish • Dedicated single socket server • Sufficient drive fanout • NICs bonded • 50Gbps line rate
Data Platforms Group 26 Xeon® D-2733NT Live-Linear Test Results Up to 386Mb/s per watt. Power is measured at PDU outlets over duration of test. This measurement includes everything from the CPU to the LED status lights. Live-Linear, 100% TLS, 93.3% Cache Hit Rate, uniform random distribution See backup (slide 28) for workloads and configurations. Results may vary​. Edge Cache Node Varnish • Dedicated single socket server • Sufficient drive fanout • NICs bonded • 50Gbps line rate
Data Platforms Group 27 Sizing Considerations • With a good understanding of the workload and software efficiency, you can build the CDN system around expected NIC throughput. • Once NIC throughput is determined, the system can be selected. • The CPU(s) can then be specified based upon the performance requirements and solution capabilities. • Additional components (memory, storage, etc) can also then be specified based on performance requirements and system capabilities.
Data Platforms Group 28 Conclusions • Solutions are systems of systems. Every component matters. • Determine the requirements and priorities • Software must be efficient and high performance • System must be appropriately sized for the workload • Processor(s) must also be appropriately sized for the workload and system constraints • Very low idle power, as low as 90 watts in powersave mode. • This solution provides excellent performance per watt. Up to 368Mb/s per watt VoD and 386Mb/s per watt for live-linear. • Efficient performance helps achieve both OpEx and sustainability goals
Data Platforms Group 29 Configuration Details (Intel® Xeon® D-2700) • As low as 90 watts active-idle power. Active-idle, intel_pstate in active mode, powersave governor. 1 node, Supermicro 110D-8C-FRAN8TP,1x Intel® Xeon® D-2733NT Processor, 8 core HT ON Turbo ON, Total Memory 128GB (4 slots/32GB/3200MT/s), BIOS 1.0 (ucode: 0x10001a0), 2x Intel® P5510 3.84TB, Integrated 2x25GbE ethernet, Ubuntu 22.04, kernel 55.15.0-40-generic, gcc 11.2.0- 19ubuntu1 compiler, openssl 3.0.2, Varnish-Plus version 6.0.10r1 revision 46f744d2ab5e7c30737dfcf8bbe823b01341efe8. Test by Intel as of 7/27/2022. • Up to 50Gbps network throughput and 368Mbps per watt. CDN Video On Demand with Intel NVMe: 1 node, Supermicro 110D-8C-FRAN8TP,1x Intel® Xeon® D-2733NT Processor, 8 core HT ON Turbo ON, Total Memory 128GB (4 slots/32GB/3200MT/s), BIOS 1.0 (ucode: 0x10001a0), 2x Intel® P5510 3.84TB, Integrated 2x25GbE ethernet, Ubuntu 22.04, kernel 55.15.0-40-generic, gcc 11.2.0- 19ubuntu1 compiler, openssl 3.0.2, Varnish-Plus version 6.0.10r1 revision 46f744d2ab5e7c30737dfcf8bbe823b01341efe8. 1 client, wrk master 4/17/2019 (keep alive on, 100,1000, and 5000 total connections). Test by Intel as of 7/27/2022. Throughput measured with 100% Transport Layer Security (TLS) traffic with 100% target cache hit ratio. • Up to 50Gbps network throughput and 386Mbps per watt. CDN Live-Linear: 1 node, Supermicro 110D-8C-FRAN8TP, 1x Intel® Xeon® D-2733NT Processor, 8 core HT ON Turbo ON, Total Memory 128GB (4 slots/32GB/3200MT/s), BIOS 1.0 (ucode: 0x10001a0), 2x Intel® P5510 3.84TB, Integrated 2x25GbE ethernet, Ubuntu 22.04, kernel 55.15.0-40-generic, gcc 11.2.0-19ubuntu1 compiler, openssl 3.0.2, Varnish-Plus version 6.0.10r1 revision 46f744d2ab5e7c30737dfcf8bbe823b01341efe8. 1 client, wrk master 4/17/2019 (keep alive on, 100,1000, and 5000 total connections). Test by Intel as of 7/27/2022. Throughput measured with 100% Transport Layer Security (TLS) traffic with 93.3% target cache hit ratio.
Data Platforms Group 30 To Learn More: • https://www.varnish-software.com/solutions/cdn/ • https://networkbuilders.intel.com/ecosystem/varnish-software • https://www.supermicro.com/en/products/system/IoT/1U/SYS- 110D-8C-FRAN8TP • https://www.intel.com/content/www/us/en/products/docs/processors /xeon-d/network-segments-product-brief.html • https://www.intel.com/content/www/us/en/products/docs/processors /xeon-d/intelligent-iot-edge-product-brief.html
31

More Related Content

PDF
Best practices for DB2 for z/OS log based recovery
Florence Dubois
 
PDF
Teradata sql-tuning-top-10
Ducan Janson
 
PPTX
CXL Fabric Management Standards
Memory Fabric Forum
 
PDF
Using pySpark with Google Colab & Spark 3.0 preview
Mario Cartia
 
PPTX
HDFS Erasure Coding in Action
DataWorks Summit/Hadoop Summit
 
PPTX
Hyper-Converged Infrastructure Vx Rail
Jürgen Ambrosi
 
PDF
Block I/O Layer Tracing: blktrace
Babak Farrokhi
 
PDF
Oracle RAC on Engineered Systems
Markus Michalewicz
 
Best practices for DB2 for z/OS log based recovery
Florence Dubois
 
Teradata sql-tuning-top-10
Ducan Janson
 
CXL Fabric Management Standards
Memory Fabric Forum
 
Using pySpark with Google Colab & Spark 3.0 preview
Mario Cartia
 
HDFS Erasure Coding in Action
DataWorks Summit/Hadoop Summit
 
Hyper-Converged Infrastructure Vx Rail
Jürgen Ambrosi
 
Block I/O Layer Tracing: blktrace
Babak Farrokhi
 
Oracle RAC on Engineered Systems
Markus Michalewicz
 

What's hot (20)

PDF
Tegra 186のu-boot & Linux
Mr. Vengineer
 
PPTX
Supermicro Servers with Micron DDR5 & SSDs: Accelerating Real World Workloads
Rebekah Rodriguez
 
DOCX
Solid State Drives - Seminar Report for Semester 6 Computer Engineering - VIT...
ravipbhat
 
PDF
MySQL Replication Performance in the Cloud
Vitor Oliveira
 
PPTX
ECS/Cloud Object Storage - DevOps Day
Bob Sokol
 
PPTX
VxRail Appliance - Modernize your infrastructure and accelerate IT transforma...
Maichino Sepede
 
PDF
VSAN – Architettura e Design
VMUG IT
 
PPTX
Ceph Intro and Architectural Overview by Ross Turk
buildacloud
 
PPTX
eMMC Embedded Multimedia Card overview
VijayGESYS
 
PDF
AMD and the new “Zen” High Performance x86 Core at Hot Chips 28
AMD
 
PDF
B35 Inside rac by Julian Dyke
Insight Technology, Inc.
 
PPTX
Past Present and Future of CXL
Memory Fabric Forum
 
PDF
Hardware & Software Platforms for HPC, AI and ML
inside-BigData.com
 
PDF
Raid
Pradeep Kumar
 
PDF
z/OS Communications Server: z/OS Resolver
zOSCommserver
 
PDF
Delivering the Future of High-Performance Computing
AMD
 
PDF
Gcp data engineer
Narendranath Reddy T
 
PPTX
Q1 Memory Fabric Forum: Building Fast and Secure Chips with CXL IP
Memory Fabric Forum
 
PDF
How should I monitor my idaa
Cuneyt Goksu
 
PPTX
Cynthia Thurlow - Metabolic Health is Wealth
DwightVillos
 
Tegra 186のu-boot & Linux
Mr. Vengineer
 
Supermicro Servers with Micron DDR5 & SSDs: Accelerating Real World Workloads
Rebekah Rodriguez
 
Solid State Drives - Seminar Report for Semester 6 Computer Engineering - VIT...
ravipbhat
 
MySQL Replication Performance in the Cloud
Vitor Oliveira
 
ECS/Cloud Object Storage - DevOps Day
Bob Sokol
 
VxRail Appliance - Modernize your infrastructure and accelerate IT transforma...
Maichino Sepede
 
VSAN – Architettura e Design
VMUG IT
 
Ceph Intro and Architectural Overview by Ross Turk
buildacloud
 
eMMC Embedded Multimedia Card overview
VijayGESYS
 
AMD and the new “Zen” High Performance x86 Core at Hot Chips 28
AMD
 
B35 Inside rac by Julian Dyke
Insight Technology, Inc.
 
Past Present and Future of CXL
Memory Fabric Forum
 
Hardware & Software Platforms for HPC, AI and ML
inside-BigData.com
 
Raid
Pradeep Kumar
 
z/OS Communications Server: z/OS Resolver
zOSCommserver
 
Delivering the Future of High-Performance Computing
AMD
 
Gcp data engineer
Narendranath Reddy T
 
Q1 Memory Fabric Forum: Building Fast and Secure Chips with CXL IP
Memory Fabric Forum
 
How should I monitor my idaa
Cuneyt Goksu
 
Cynthia Thurlow - Metabolic Health is Wealth
DwightVillos
 
Ad

Similar to Building Efficient Edge Nodes for Content Delivery Networks (20)

PDF
Introduction to container networking in K8s - SDN/NFV London meetup
Haidee McMahon
 
PDF
The Intel Xeon Scalable Processor and IoT
Advantech Industrial Automation Group
 
PPTX
Tackling Retail Technology Management Challenges at the Edge
Rebekah Rodriguez
 
PDF
Re-Imagining the Data Center with Intel
Intel IT Center
 
PPTX
Yashi dealer meeting settembre 2016 tecnologie xeon intel italia
Yashi Italia
 
PDF
Inside story on Intel Data Center @ IDF 2013
Intel IT Center
 
PPTX
Webinář: Dell VRTX - datacentrum vše-v-jednom za skvělou cenu / 7.10.2013
Jaroslav Prodelal
 
PDF
NVMe_Infrastructure_final1.pdf
IrfanBroadband
 
PPTX
Accelerating Innovation from Edge to Cloud
Rebekah Rodriguez
 
PDF
5G Multi-Access Edge Compute
Michelle Holley
 
PPTX
Webinář: Provozujte datacentrum v kanceláři (Dell VRTX) / 5.9.2013
Jaroslav Prodelal
 
PPTX
Acceleration_and_Security_draft_v2
Srinivasa Addepalli
 
PDF
Intel® Xeon® Scalable Processors Enabled Applications Marketing Guide
Intel IT Center
 
PDF
“Smarter Manufacturing with Intel’s Deep Learning-Based Machine Vision,” a Pr...
Edge AI and Vision Alliance
 
PDF
DAOS - Scale-Out Software-Defined Storage for HPC/Big Data/AI Convergence
inside-BigData.com
 
PPTX
Performance out of the box developers
Michelle Holley
 
PDF
IBM InterConnect 2013 Expert Integrated Systems Keynote: Sotiropoulos & Wieck
IBM Events
 
PPTX
Simplify Data Management and Go Green with Supermicro & Qumulo
Rebekah Rodriguez
 
PDF
Edge Computing and 5G - SDN/NFV London meetup
Haidee McMahon
 
PDF
ONS 2018 LA - Intel Tutorial: Cloud Native to NFV - Alon Bernstein, Cisco & K...
Kuralamudhan Ramakrishnan
 
Introduction to container networking in K8s - SDN/NFV London meetup
Haidee McMahon
 
The Intel Xeon Scalable Processor and IoT
Advantech Industrial Automation Group
 
Tackling Retail Technology Management Challenges at the Edge
Rebekah Rodriguez
 
Re-Imagining the Data Center with Intel
Intel IT Center
 
Yashi dealer meeting settembre 2016 tecnologie xeon intel italia
Yashi Italia
 
Inside story on Intel Data Center @ IDF 2013
Intel IT Center
 
Webinář: Dell VRTX - datacentrum vše-v-jednom za skvělou cenu / 7.10.2013
Jaroslav Prodelal
 
NVMe_Infrastructure_final1.pdf
IrfanBroadband
 
Accelerating Innovation from Edge to Cloud
Rebekah Rodriguez
 
5G Multi-Access Edge Compute
Michelle Holley
 
Webinář: Provozujte datacentrum v kanceláři (Dell VRTX) / 5.9.2013
Jaroslav Prodelal
 
Acceleration_and_Security_draft_v2
Srinivasa Addepalli
 
Intel® Xeon® Scalable Processors Enabled Applications Marketing Guide
Intel IT Center
 
“Smarter Manufacturing with Intel’s Deep Learning-Based Machine Vision,” a Pr...
Edge AI and Vision Alliance
 
DAOS - Scale-Out Software-Defined Storage for HPC/Big Data/AI Convergence
inside-BigData.com
 
Performance out of the box developers
Michelle Holley
 
IBM InterConnect 2013 Expert Integrated Systems Keynote: Sotiropoulos & Wieck
IBM Events
 
Simplify Data Management and Go Green with Supermicro & Qumulo
Rebekah Rodriguez
 
Edge Computing and 5G - SDN/NFV London meetup
Haidee McMahon
 
ONS 2018 LA - Intel Tutorial: Cloud Native to NFV - Alon Bernstein, Cisco & K...
Kuralamudhan Ramakrishnan
 
Ad

More from Rebekah Rodriguez (20)

PPTX
Delivering Supermicro Software Defined Storage Solutions with OSNexus QuantaStor
Rebekah Rodriguez
 
PPTX
MWC Roundtable: Accelerating Innovation from the Intelligent Edge to Cloud
Rebekah Rodriguez
 
PPTX
Supermicro and The Green Grid (TGG)
Rebekah Rodriguez
 
PPTX
X13 Products + Intel® Xeon® CPU Max Series–An Applications & Performance View
Rebekah Rodriguez
 
PPTX
X13 Products + Intel® Xeon® CPU Max Series–An Applications & Performance View
Rebekah Rodriguez
 
PPTX
Supermicro AI Pod that’s Super Simple, Super Scalable, and Super Affordable
Rebekah Rodriguez
 
PDF
X13 Pre-Release Update featuring 4th Gen Intel® Xeon® Scalable Processors
Rebekah Rodriguez
 
PPTX
The Power of HPC with Next Generation Supermicro Systems
Rebekah Rodriguez
 
PPTX
New Accelerated Compute Infrastructure Solutions from Supermicro
Rebekah Rodriguez
 
PPTX
Innovative Solutions for Cloud Gaming, Media, Transcoding, & AI Inferencing
Rebekah Rodriguez
 
PPTX
Zero Trust for Private 5G and Edge
Rebekah Rodriguez
 
PPTX
Benefits of Operating an On-Premises Infrastructure
Rebekah Rodriguez
 
PPTX
Emerging Cloud Storage Trends for Enterprises
Rebekah Rodriguez
 
PPTX
Drive Data Center Efficiency with SuperBlade, Powered by AMD EPYC™ and Instinct™
Rebekah Rodriguez
 
PPTX
Supermicro’s Universal GPU: Modular, Standards Based and Built for the Future
Rebekah Rodriguez
 
PPTX
Optimize Content Delivery with Multi-Access Edge Computing
Rebekah Rodriguez
 
PPTX
Delivering Breakthrough Performance Per Core with AMD EPYC
Rebekah Rodriguez
 
PPTX
Delivering Breakthrough Performance Per Core with AMD EPYC
Rebekah Rodriguez
 
PPTX
High-Density Top-Loading Storage for Cloud Scale Applications
Rebekah Rodriguez
 
PPTX
Consumption Based On-Demand Private Cloud in a Box
Rebekah Rodriguez
 
Delivering Supermicro Software Defined Storage Solutions with OSNexus QuantaStor
Rebekah Rodriguez
 
MWC Roundtable: Accelerating Innovation from the Intelligent Edge to Cloud
Rebekah Rodriguez
 
Supermicro and The Green Grid (TGG)
Rebekah Rodriguez
 
X13 Products + Intel® Xeon® CPU Max Series–An Applications & Performance View
Rebekah Rodriguez
 
X13 Products + Intel® Xeon® CPU Max Series–An Applications & Performance View
Rebekah Rodriguez
 
Supermicro AI Pod that’s Super Simple, Super Scalable, and Super Affordable
Rebekah Rodriguez
 
X13 Pre-Release Update featuring 4th Gen Intel® Xeon® Scalable Processors
Rebekah Rodriguez
 
The Power of HPC with Next Generation Supermicro Systems
Rebekah Rodriguez
 
New Accelerated Compute Infrastructure Solutions from Supermicro
Rebekah Rodriguez
 
Innovative Solutions for Cloud Gaming, Media, Transcoding, & AI Inferencing
Rebekah Rodriguez
 
Zero Trust for Private 5G and Edge
Rebekah Rodriguez
 
Benefits of Operating an On-Premises Infrastructure
Rebekah Rodriguez
 
Emerging Cloud Storage Trends for Enterprises
Rebekah Rodriguez
 
Drive Data Center Efficiency with SuperBlade, Powered by AMD EPYC™ and Instinct™
Rebekah Rodriguez
 
Supermicro’s Universal GPU: Modular, Standards Based and Built for the Future
Rebekah Rodriguez
 
Optimize Content Delivery with Multi-Access Edge Computing
Rebekah Rodriguez
 
Delivering Breakthrough Performance Per Core with AMD EPYC
Rebekah Rodriguez
 
Delivering Breakthrough Performance Per Core with AMD EPYC
Rebekah Rodriguez
 
High-Density Top-Loading Storage for Cloud Scale Applications
Rebekah Rodriguez
 
Consumption Based On-Demand Private Cloud in a Box
Rebekah Rodriguez
 

Recently uploaded (20)

PDF
7 ChatGPT Prompts to Help You Define Your Ideal Customer Profile.pdf
SOFTTECHHUB
 
PDF
Network Security Unit 5.pdf for BCA BBA.
Serpent6
 
PDF
Unlocking AI with Model Context Protocol (MCP)
Brian McKeiver
 
PDF
Diabetes mellitus diagnosis method based random forest with bat algorithm
IAESIJAI
 
PDF
Dropbox Q2 2025 Financial Results & Investor Presentation
Dropbox
 
PDF
Chapter 3 Spatial Domain Image Processing.pdf
Getnet Tigabie Askale -(GM)
 
PDF
KodekX | Application Modernization Development
KodekX
 
PDF
Advanced IT Governance
University of Hertfordshire
 
DOCX
The AUB Centre for AI in Media Proposal.docx
GAONE9
 
PDF
Reach Out and Touch Someone: Haptics and Empathic Computing
Mark Billinghurst
 
PDF
Approach and Philosophy of On baking technology
30anthuong17
 
PPTX
VMware vSphere Foundation How to Sell Presentation-Ver1.4-2-14-2024.pptx
blackmambaettijean
 
PDF
Optimiser vos workloads AI/ML sur Amazon EC2 et AWS Graviton
Julien SIMON
 
PDF
Peak of Data & AI Encore- AI for Metadata and Smarter Workflows
Safe Software
 
PPTX
breach-and-attack-simulation-cybersecurity-india-chennai-defenderrabbit-2025....
defencerabbit Team
 
PPT
“AI and Expert System Decision Support & Business Intelligence Systems”
ZubinRadhakrishnan
 
PDF
Spectral efficient network and resource selection model in 5G networks
IAESIJAI
 
PDF
solutions_manual_-_materials___processing_in_manufacturing__demargo_.pdf
AbdullahSani29
 
PDF
How UI/UX Design Impacts User Retention in Mobile Apps.pdf
SynapseIndia
 
PPTX
Cloud computing and distributed systems.
kkkkkhan
 
7 ChatGPT Prompts to Help You Define Your Ideal Customer Profile.pdf
SOFTTECHHUB
 
Network Security Unit 5.pdf for BCA BBA.
Serpent6
 
Unlocking AI with Model Context Protocol (MCP)
Brian McKeiver
 
Diabetes mellitus diagnosis method based random forest with bat algorithm
IAESIJAI
 
Dropbox Q2 2025 Financial Results & Investor Presentation
Dropbox
 
Chapter 3 Spatial Domain Image Processing.pdf
Getnet Tigabie Askale -(GM)
 
KodekX | Application Modernization Development
KodekX
 
Advanced IT Governance
University of Hertfordshire
 
The AUB Centre for AI in Media Proposal.docx
GAONE9
 
Reach Out and Touch Someone: Haptics and Empathic Computing
Mark Billinghurst
 
Approach and Philosophy of On baking technology
30anthuong17
 
VMware vSphere Foundation How to Sell Presentation-Ver1.4-2-14-2024.pptx
blackmambaettijean
 
Optimiser vos workloads AI/ML sur Amazon EC2 et AWS Graviton
Julien SIMON
 
Peak of Data & AI Encore- AI for Metadata and Smarter Workflows
Safe Software
 
breach-and-attack-simulation-cybersecurity-india-chennai-defenderrabbit-2025....
defencerabbit Team
 
“AI and Expert System Decision Support & Business Intelligence Systems”
ZubinRadhakrishnan
 
Spectral efficient network and resource selection model in 5G networks
IAESIJAI
 
solutions_manual_-_materials___processing_in_manufacturing__demargo_.pdf
AbdullahSani29
 
How UI/UX Design Impacts User Retention in Mobile Apps.pdf
SynapseIndia
 
Cloud computing and distributed systems.
kkkkkhan
 

Building Efficient Edge Nodes for Content Delivery Networks

  • 1. Building Efficient Edge Nodes for Content Delivery Networks Frank Miller, Chief Technology Officer, Varnish Software Raymond Chung, Director, Product Management, Supermicro Nash Kleppan, Cloud Solutions Engineer, Intel
  • 2. Data Platforms Group 2 Notices & Disclaimers  Performance varies by use, configuration and other factors. Learn more on the Performance Index site.  Performance results are based on testing as of dates shown in configurations and may not reflect all publicly available ​updates. See backup for configuration details. No product or component can be absolutely secure.  Your costs and results may vary.  Intel technologies may require enabled hardware, software or service activation.  © Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.
  • 3. Data Platforms Group 3 Agenda • Background • Varnish Software • Supermicro SuperServer System for CDN • Intel® Xeon® D-1700 and D-2700 • Test Environment • Power and Performance Testing • Sizing Considerations • Conclusions
  • 4. Data Platforms Group 4 Background • There has been an explosion of live streaming consumption, estimated to be nearly 4 billion hours in 20201 • The growth of high definition (1080p) and ultra-high definition (4K) content increases the performance demands on a CDN • The deployment of 5G networks allows users of mobile devices to consume video at higher speed • Performance and optimization depend upon software efficiency and system design • The goal of this Varnish - Supermicro – Intel collaboration on CDN is to improve the performance per TCO and performance per Watt of edge caching servers 1: https://www.digitaltveurope.com/2020/05/14/almost-4-billion-hours-watched-as-live-streaming-industry-benefits-from-lockdown/
  • 5. Data Platforms Group 5 Trends in Storage Systems • Increased demand for NVMe storage • 25GbE and 100GbE networking at the edge • High-performance workloads on broadly available servers • Workloads include: CDN and scale-out storage • Traditional high performance storage systems • Performance typically prioritized over other concerns • May not be suitable for reduced performance, power constrained deployments • Efficient storage systems • Require appropriately sized server and processor(s) running efficient software. • Must be appropriately sized to intended performance target for optimal power and performance • Watts turn into kilowatts – Need to keep sustainability goals and OpEx in mind.
  • 7. Varnish is an HTTP Cache Varnish is infrastructure software. HTTP HTTP Origin
  • 8. The software core is kept lean and secure with only the necessary features to operate. Implemented in C and multi- threaded for linear scalability across CPU, RAM and I/O resources. A small privileged runtime minimizes the threat surface while worker threads, with limited privileges, perform the actual work with kernel and hardware awareness. Hardware Aware VMODs Varnish Enterprise (core) Varnish is fast with integration of hardware architectures and features (e.g. NUMA and TLS optimization into the Varnish runtime). Varnish is flexible with Varnish Modules (VMODS) for additional features. Implemented in Varnish Control Language (VCL) and compiled to a shared and dynamically linked library. 8 Varnish is Highly Performant, Flexible and Secure
  • 9. There are key benefits in high performance in compute The Value of High Performance This requires the harmonization of software, processor and system Efficient solutions drive for lower power utilization for the same amount of unit work (e.g. Gbs/Watt, Flops/Watt, etc.) This should be table stakes in the future of compute at scale: • We can do more for less. • We can even show benefit on current hardware, with optimized software, to gain power savings with only a software upgrade. • There is a clear North Star … to Save the Planet
  • 10. IoT/Embedded & Edge Computing Raymond Chung Director, Product Management raymondchung@supermicro.com Better Faster Greener™ © 2022 Supermicro
  • 11. Why Supermicro IoT/Embedded & Edge Computing 10/3/2022 Better Faster Greener™ © 2022 Supermicro 11 Better Faster Greener Supermicro building blocks solutions is Better for IoT architecture design Supermicro Data Center Technology empower Faster IoT and Edge Computing Supermicro Edge Computing with Muti-Node System is Greener for environment Supermicro is a global technology leader committed to delivering first to market innovation for Enterprise, Cloud, AI, and 5GTelco/Edge IT Infrastructure.We are transforming into aTotal IT Solutions provider with environmentally-friendly and energy-saving server, AI, storage, IoT, and switch systems, software, and services while delivering advanced high-volume motherboard, power, and chassis products.
  • 12. Edge Appliances SYS-220HE SYS-110P SYS-E5019A  Built for Net/Comms  2x workloads  Power Needs  Increased internal connections  Redundancy/HA  Expandability  Flexible  Rackmount or standalone  NEBS and harsher certifications  Cost Effective  Enterprise driven  Normalized Workloads  uCPE & Intel Select solutions  SMB capable  Lower power consumption Minimal Port / Throughput Maximum Port / Throughput SYS-E100/302 SYS-510D Better Faster Greener™ © 2021 Supermicro 12 10/3/2022 SYS-110D SYS-110A SYS-120U SYS-210SE SYS-E403 SYS-220U
  • 13. 10/3/2022 Better Faster Greener™ © 2022 Supermicro 13 SYS-110D-4C/8C/16C/20C-FR(A/D)N8TP Intel® Xeon D Compact 1U Embedded System https://www.supermicro.com/en/products/system/IoT/1U/SYS-110D-20C-FRDN8TP Processor Intel® Xeon D-2700 Processor, 4/8/14/16/20 Cores, 65W~125W Memory Capacity DDR4 3200MHz 1DPC, 4 slots RDIMM/LRDIMM, up to 256GB/512GB Expansion Slots 1x PCIe 4.0 x 16 1x M.2 M-Key SATA3.0/PCI-E 3.0 x4, 2242/2280 I/O Ports 1x IPMI dedicated LAN 4x GbE, 2x 10GbE, 2x 25G SFP28 2x USB 3.0 type A 1x VGA Drive Bays 2x Internal 2.5" drive bays Power Supply Redundant 800 W, 100-240 Vac Redundant 600 W, -48 Vdc System Cooling 5x 40x56mm Hot-swap Counter-rotation PWM Fans (one is optional) Motherboard: X12SDV-4/8/16/20C-SPT8F (7.5”x 8.5”) Chassis: 515M-R804/515M-R601 Dimension: W17.2” x D15.7” x H1.7” (437mm x 399mm x 43mm) Intel® Xeon D-2700 4/8/14/16/20 Cores Processor Up to DDR4 3200Mhz 4 channels, 4 slots RDIMM/LRDIMM up to 512GB 1x IPMI dedicated LAN; 4x GbE; 2x 10GbE; 2x 25G SFP28 2x USB 3.0 type A; 1x M.2 M-Key SATA/PCI-E 3.0 x4, 2280 2x SATA/U.2 2.5” drives Key features Applications 1x VGA PCIe 4.0 x16 M-Key M.2 Networking Application, Firewall, uCPE, Telecom Application Xeon D-2700 4/8/16/20C 4x GbE LAN, 2x 10GbE LAN 2x 25G SFP28 1 x IPMI 2x 2.5” drive bays 1 2 3 4 5 6 7
  • 14. 10/3/2022 Better Faster Greener™ © 2022 Supermicro 14 SYS-510D-4C/8C/10C-FN6P Intel® Xeon D Compact 1U Embedded System https://www.supermicro.com/en/products/system/iot/1u/sys-510d-10c-fn6p Processor Intel® Xeon D-1700 Processor, 4/8/10 Cores, 25W/50W/88W Memory Capacity 4 DIMM slots DDR4 2933MHz up to 256GB RDIMM/UDIMM ECC/Non-ECC Expansion Slots 1 x PCIe 4.0 x 16 1x M.2 M-Key SATA3.0/PCI-E 3.0 x4, 2280 1x M.2 E-Key PCI-E 3.0 x1, 2230 1x M.2 B-Key SATA/PCI-E 3.0 x2/USB 3.0, 2280/3042 I/O Ports 1x IPMI dedicated LAN 4x GbE, 2x 25G SFP28 2x USB 3.0 type A 1x VGA Drive Bays 1x Internal 3.5" or 4x Internal 2.5" drive bays Power Supply 200W AC-DC power supply System Cooling 3x 40x28mm 4-PIN PWM fans Motherboard: X12SDV-4C/8C/10C-SP6F (9.0” x 7.25” Flex ATX) Chassis: CSE-505-203B Dimension: W17.2” x D9.8” x H1.7” (437mm x 249mm x 43mm) Intel® Xeon D-1700 4/8/10 Cores Processor Up to 256GB RDIMM or 128GB UDIMM, DDR4 2933MHz in 4 Slots 1x IPMI dedicated LAN, 4x GbE,2x 25G SFP28, 2x USB 3.0 type A 1x M.2 M-Key SATA/PCI-E 3.0 x4, 2280, 1x M.2 E-Key PCI-E 3.0 x1, 2230 1x M.2 B-Key SATA3.0/PCI-E 3.0 x2/USB 3.0, 2280/3042 4x SATA drives Key features Applications 1x VGA PCIe 4.0 x16 M, B, E-Key M.2 Retail Application, Firewall, Gateway, uCPE Xeon D-1700 4/8/10C 4x GbE LAN, 2x 25G SFP28 1 x IPMI 1x 3.5” or 4x 2.5” drive bays 1 2 3 4 5 6 7
  • 15. DISCLAIMER Super Micro Computer, Inc. may make changes to specifications and product descriptions at any time, without notice. The information presented in this document is for informational purposes only and may contain technical inaccuracies, omissions and typographical errors. Any performance tests and ratings are measured using systems that reflect the approximate performance of Super Micro Computer, Inc. products as measured by those tests. Any differences in software or hardware configuration may affect actual performance, and Super Micro Computer, Inc. does not control the design or implementation of third party benchmarks or websites referenced in this document. The information contained herein is subject to change and may be rendered inaccurate for many reasons, including but not limited to any changes in product and/or roadmap, component and hardware revision changes, new model and/or product releases, software changes, firmware changes, or the like. Super Micro Computer, Inc. assumes no obligation to update or otherwise correct or revise this information. SUPER MICRO COMPUTER, INC. MAKES NO REPRESENTATIONS OR WARRANTIES WITH RESPECT TO THE CONTENTS HEREOF AND ASSUMES NO RESPONSIBILITY FOR ANY INACCURACIES, ERRORS OR OMISSIONS THAT MAY APPEAR IN THIS INFORMATION. SUPER MICRO COMPUTER, INC. SPECIFICALLY DISCLAIMS ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR ANY PARTICULAR PURPOSE. IN NO EVENT WILL SUPER MICRO COMPUTER, INC. BE LIABLE TO ANY PERSON FOR ANY DIRECT, INDIRECT, SPECIAL OR OTHER CONSEQUENTIAL DAMAGES ARISING FROM THE USE OF ANY INFORMATION CONTAINED HEREIN, EVEN IF SUPER MICRO COMPUTER, Inc. IS EXPRESSLY ADVISED OF THE POSSIBILITY OF SUCH DAMAGES. 10/3/2022 Better Faster Greener™ © 2021 Supermicro 15
  • 17. Data Platforms Group 17 Intel® Xeon® D-1700 and D-2700 Nash Kleppan, Cloud Solutions Engineer, Intel
  • 18. Data Platforms Group 18 Intel® Xeon® D-1700 • For CDN applications, these processors offer: • Integrated Ethernet up to 100 Gbps • Increased memory bandwidth (2 or 3 channels DDR4 2933 MT/s) • Support for 16 lanes PCIe Gen4 and 24 lanes HSIO • Increased core count and cache over prior gen. Up to 10 cores. • Intel® Software Guard Extensions (Intel® SGX) for more protected execution. • Intel® QAT Gen 2
  • 19. Data Platforms Group 19 Intel® Xeon® D-2700 • For CDN applications, these processors offer: • Integrated Ethernet up to 100 Gbps • Increased memory bandwidth (4 channels DDR4 3200 MT/s 1DPC) • Support for 32 lanes PCIe Gen4 and 24 lanes HSIO • Increased core count and cache over prior gen. Up to 20 cores. • Intel® Software Guard Extensions (Intel® SGX) for more protected execution. • Intel® QAT Gen 3
  • 20. Data Platforms Group 20 Test Configuration: Xeon® D-2700 Nash Kleppan, Cloud Solutions Engineer, Intel
  • 21. Data Platforms Group 21 Xeon® D-2700 Test Configuration Edge Cache Node CDN Application CPU Content Origin Client (wrk) 100GbE L2 Switch • 1RU server • Sufficient drive fanout • NICs bonded • 50Gbps line rate • Supermicro SuperServer SYS- 110D-8C-FRAN8TP • 1x Intel® Xeon® D-2733NT • 4x 32GB DDR4 3200 (128GB) • Integrated 2x25GbE • 2x Intel® SSD D7-P5510 3.84TB NVMe 25GbE
  • 22. Data Platforms Group 22 Xeon® D-2700 Test Configuration - Varnish Edge Cache Node Varnish CPU Content Origin Client (wrk) 100GbE L2 Switch • Supermicro SYS-110D-8C- FRAN8TP • 1x Intel® Xeon® D-2733NT • 4x 32GB DDR4 3200 (128GB) • Integrated 2x25GbE • 2x Intel® SSD D7-P5510 3.84TB NVMe • Ubuntu 22.04 • Varnish-Plus version 6.0.10r1 • OpenSSL 3.0.2 • Power consumption at both PSU inlets measured with Raritan PX3 PDU. 25GbE
  • 23. Data Platforms Group 23 Benchmarking Results: Intel Xeon® D-2700 Nash Kleppan, Cloud Solutions Engineer, Intel
  • 24. Data Platforms Group 24 Active-Idle Total System Power Measurements Power is measured at PDU outlets over 5 minutes. This measurement includes everything from the CPU to the LED status lights. See backup (slide 28) for workloads and configurations. Results may vary​. Edge Cache Node • Dedicated single socket server • Sufficient drive fanout • NICs bonded • 50Gbps line rate • Supermicro SYS-110D-8C-FRAN8TP • Ubuntu 22.04 • Using Intel_pstate CPU Scaling Driver • 90 watt average active-idle power in powersave mode • 113 watt average active-idle power in performance mode
  • 25. Data Platforms Group 25 Xeon® D-2733NT Video on Demand Test Results Up to 368Mb/s per watt. Power is measured at PDU outlets over duration of test. This measurement includes everything from the CPU to the LED status lights. Video-on-Demand, 100% TLS, 100% Cache Hit Rate, uniform random distribution, 315GB dataset on NVMe See backup (slide 28) for workloads and configurations. Results may vary​. Edge Cache Node Varnish • Dedicated single socket server • Sufficient drive fanout • NICs bonded • 50Gbps line rate
  • 26. Data Platforms Group 26 Xeon® D-2733NT Live-Linear Test Results Up to 386Mb/s per watt. Power is measured at PDU outlets over duration of test. This measurement includes everything from the CPU to the LED status lights. Live-Linear, 100% TLS, 93.3% Cache Hit Rate, uniform random distribution See backup (slide 28) for workloads and configurations. Results may vary​. Edge Cache Node Varnish • Dedicated single socket server • Sufficient drive fanout • NICs bonded • 50Gbps line rate
  • 27. Data Platforms Group 27 Sizing Considerations • With a good understanding of the workload and software efficiency, you can build the CDN system around expected NIC throughput. • Once NIC throughput is determined, the system can be selected. • The CPU(s) can then be specified based upon the performance requirements and solution capabilities. • Additional components (memory, storage, etc) can also then be specified based on performance requirements and system capabilities.
  • 28. Data Platforms Group 28 Conclusions • Solutions are systems of systems. Every component matters. • Determine the requirements and priorities • Software must be efficient and high performance • System must be appropriately sized for the workload • Processor(s) must also be appropriately sized for the workload and system constraints • Very low idle power, as low as 90 watts in powersave mode. • This solution provides excellent performance per watt. Up to 368Mb/s per watt VoD and 386Mb/s per watt for live-linear. • Efficient performance helps achieve both OpEx and sustainability goals
  • 29. Data Platforms Group 29 Configuration Details (Intel® Xeon® D-2700) • As low as 90 watts active-idle power. Active-idle, intel_pstate in active mode, powersave governor. 1 node, Supermicro 110D-8C-FRAN8TP,1x Intel® Xeon® D-2733NT Processor, 8 core HT ON Turbo ON, Total Memory 128GB (4 slots/32GB/3200MT/s), BIOS 1.0 (ucode: 0x10001a0), 2x Intel® P5510 3.84TB, Integrated 2x25GbE ethernet, Ubuntu 22.04, kernel 55.15.0-40-generic, gcc 11.2.0- 19ubuntu1 compiler, openssl 3.0.2, Varnish-Plus version 6.0.10r1 revision 46f744d2ab5e7c30737dfcf8bbe823b01341efe8. Test by Intel as of 7/27/2022. • Up to 50Gbps network throughput and 368Mbps per watt. CDN Video On Demand with Intel NVMe: 1 node, Supermicro 110D-8C-FRAN8TP,1x Intel® Xeon® D-2733NT Processor, 8 core HT ON Turbo ON, Total Memory 128GB (4 slots/32GB/3200MT/s), BIOS 1.0 (ucode: 0x10001a0), 2x Intel® P5510 3.84TB, Integrated 2x25GbE ethernet, Ubuntu 22.04, kernel 55.15.0-40-generic, gcc 11.2.0- 19ubuntu1 compiler, openssl 3.0.2, Varnish-Plus version 6.0.10r1 revision 46f744d2ab5e7c30737dfcf8bbe823b01341efe8. 1 client, wrk master 4/17/2019 (keep alive on, 100,1000, and 5000 total connections). Test by Intel as of 7/27/2022. Throughput measured with 100% Transport Layer Security (TLS) traffic with 100% target cache hit ratio. • Up to 50Gbps network throughput and 386Mbps per watt. CDN Live-Linear: 1 node, Supermicro 110D-8C-FRAN8TP, 1x Intel® Xeon® D-2733NT Processor, 8 core HT ON Turbo ON, Total Memory 128GB (4 slots/32GB/3200MT/s), BIOS 1.0 (ucode: 0x10001a0), 2x Intel® P5510 3.84TB, Integrated 2x25GbE ethernet, Ubuntu 22.04, kernel 55.15.0-40-generic, gcc 11.2.0-19ubuntu1 compiler, openssl 3.0.2, Varnish-Plus version 6.0.10r1 revision 46f744d2ab5e7c30737dfcf8bbe823b01341efe8. 1 client, wrk master 4/17/2019 (keep alive on, 100,1000, and 5000 total connections). Test by Intel as of 7/27/2022. Throughput measured with 100% Transport Layer Security (TLS) traffic with 93.3% target cache hit ratio.
  • 30. Data Platforms Group 30 To Learn More: • https://www.varnish-software.com/solutions/cdn/ • https://networkbuilders.intel.com/ecosystem/varnish-software • https://www.supermicro.com/en/products/system/IoT/1U/SYS- 110D-8C-FRAN8TP • https://www.intel.com/content/www/us/en/products/docs/processors /xeon-d/network-segments-product-brief.html • https://www.intel.com/content/www/us/en/products/docs/processors /xeon-d/intelligent-iot-edge-product-brief.html
  • 31. 31

Editor's Notes

  • #8: Varnish is infrastructure software - it is the bridge between the hardware and apps on top, a part of the stack. You sandwich it in between the clients and the conventional web servers, where it will cache data / assets, or anything HTTP, really. This is the basic picture, but it is more, of course. Why do we add a cache? Web servers are not necessarily slow, but they are complex, and a cache will always outperform generating content again. Also, the cache may be close to the clients, reducing bandwidth, and have a strategy for when the backend server goes down. Quite often, it is a good idea to serve stale content instead of 404s or 5xx. (Next slide, please). Image credits, Creative Commons Pixabay (flip phone) openclipart.org (PC), non-attribution license Wikimedia Commons (laptop). Server rack: https://www.maxpixel.net/Stack-Store-Drive-Storage-Rack-Hard-Disk-30459
  • #9: More beyond hardware NUMA awareness and TLS in process … The threading model with worker threads works because we use few context switches compared to the amount of requests and data flowing through the system. We optimize to reduce the amount of context switches and locking. The asynchronous direct IO is what separates MSE 3 from MSE 2 - we take control over the disk, performing disk optimization past challenges in Linux with the volume of soft page faults. A bit more on VCL - The VCL language provides a lot of flexibility when it comes to making decisions in the cache, for example when it comes to caching behavior. VCL is also allows logic for routing, can do request manipulation of various kinds, business logic, tool connectors … the opportunity is endless to add value while keeping the core highly tuned and reliable.