Ken Cantrell / NetApp Mark Rogov / EMC J Metz / Cisco October 21, 2015
© 2015 Storage Networking Industry Association. All Rights Reserved. SNIA Legal Notice The material contained in this tutorial is copyrighted by the SNIA unless otherwise noted. Member companies and individual members may use this material in presentations and literature under the following conditions: Any slide or slides used must be reproduced in their entirety without modification The SNIA must be acknowledged as the source of any material used in the body of any document containing material from these presentations. This presentation is a project of the SNIA Education Committee. Neither the author nor the presenter is an attorney and nothing in this presentation is intended to be, or should be construed as legal advice or an opinion of counsel. If you need legal advice or a legal opinion please contact your attorney. The information presented herein represents the author's personal opinion and current understanding of the relevant issues involved. The author, the presenter, and the SNIA do not assume any responsibility or liability for damages arising out of any reliance on or use of this information. NO WARRANTIES, EXPRESS OR IMPLIED. USE AT YOUR OWN RISK. 2
© 2015 Storage Networking Industry Association. All Rights Reserved. About The Speakers 3
© 2015 Storage Networking Industry Association. All Rights Reserved. SPEC SFS® 2014 See for detailshttp://spec.org/sfs2014/ SWBUILD VDA VDI DATABASE spec © P ROTOCOL S UPPORT A NY T RANSPORT P ROTOCOL TCP/IP, FC, I SCSI, ETC. A NY N ETWORK P ROTOCOL NFS, SMB, ETC. U SES POSIX FILE OPERATIONS M ETRICS B USINESS METRICS COUNTS SOFTWARE BUILDS, VIRTUAL DESKTOPS, VIDEO STREAMS, DATABASES R ESPONSE TIME
© 2015 Storage Networking Industry Association. All Rights Reserved. Storage Performance Benchmarking 5 SOLUTION UNDER TEST METRICS AND TERMINOLOGY FILE COMPONENTS BLOCK COMPONENTS WORKLOAD DEFINITIONS
© 2015 Storage Networking Industry Association. All Rights Reserved. Storage Performance Benchmarking 6 METRICS AND TERMINOLOGY FILE COMPONENTS BLOCK COMPONENTS WORKLOAD DEFINITIONS TODAY FUTURE WEBCASTSJULY 30, 2015 SOLUTION UNDER TEST ENDSUTINTROCACHEMOLEHEROMiB/SMiB/S
© 2015 Storage Networking Industry Association. All Rights Reserved. 7 END SUT CACHE MOLE HEROMiB/SMiB/S
© 2015 Storage Networking Industry Association. All Rights Reserved. Metrics and Terminology Review Part 1: (Both PDF and PPT available) 8 GRAPH FUN CONTEXT MAKES METRICS MATTER CONTEXT MAKES METRICS MATTER OPS COUNT EVERY PROTOCOL OPERATION PER SECOND MB/S PAYLOAD SUM OF EVERY OPERATION PER SECOND TERMINOLOGY IOPS COUNT EVERY IO OPERATION PER SECOND RESPONSE TIME TIME TARGET TAKES TO REPLY TO AN IO ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. Last Time: Which is Better? 9 Operations per sec, OPS Response time, ms ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. Last Time: Which is Better? 10 Operations per sec, OPS Response time, ms CONTEXT MAKES METRICS MATTER CONTEXT MAKES METRICS MATTER ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. This Time: Why Are They Different? 11 Operations per sec, OPS Response time, ms ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. 12 END INTRO CACHE MOLE HEROMiB/SMiB/S
© 2015 Storage Networking Industry Association. All Rights Reserved. What is a Solution Under Test (SUT)? 13 HYPERVISOR VM HYPERVISOR VM Increasing Complexity ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. SUT Layers 14 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. Front- and Back-End Basic Difference 15 2KiB IOs FRONT ENDBACK END 8KiB IOs CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. RUN ON Is a Workload / Application a Part of the SUT? 16 WORKLOADS HYPERVISOR VM HYPERVISOR VM SOLUTIONS UNDER TEST ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. 17 END SUT INTRO MOLE HEROMiB/SMiB/S
© 2015 Storage Networking Industry Association. All Rights Reserved. Which Elements in the SUT Affect Performance? 18 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. What is Common Between All Of The Components? 19 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT PROCESSING ELEMENTS (CPU, ASIC) MEMORY / CACHE ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. Answer To Ken’s Interview Question 20 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT CACHE IO ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. 3 Principles To Improve Performance 21 D O L ESS W ORK D O W ORK F ASTER I NCREASE P ARALLELISM ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. 22 END SUT INTRO CACHE HEROMiB/SMiB/S
© 2015 Storage Networking Industry Association. All Rights Reserved. Latency Or “Whack A Mole Game” 23 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT 7200 RPM 1000x Response Time ( ms) ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. Disk Bottleneck 24 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT DISK BOUND aka DISK BOTTLENECK Response Time ( ms) 1000x 7200 RPM ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. Why Are SSDs So Compelling? 25 Response Time ( ns) Random Write Latency 11ms 0.1ms ENDMiB/SMiB/SHEROMOLECACHESUTINTRO 25ns
© 2015 Storage Networking Industry Association. All Rights Reserved. Why Are SSDs So Compelling? Response Time ( ns) Random Write Latency 11ms 0.1ms ENDMiB/SMiB/SHEROMOLECACHESUTINTRO 25ns 100 X
© 2015 Storage Networking Industry Association. All Rights Reserved. Change SUT: Upgrade With SSDs 27 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT SSD WAS 12ms Response Time ( ms) 6x D O W ORK F ASTER ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. Change SUT: Upgrade With SSDs 28 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT Response Time ( ms) D O W ORK F ASTER ENDMiB/SMiB/SHEROMOLECACHESUTINTRO CONTROLLER BOUND WAS 12ms
© 2015 Storage Networking Industry Association. All Rights Reserved. Controller Bottleneck 29 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT VENDOR Response Time ( ms) ~1.8x CLIENT BOUND WAS 0.7ms D O W ORK F ASTER ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. Bottlenecks Always Exist 30 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT BOTTLENECK MOVES Response Time ( ms) D O W ORK F ASTER ENDMiB/SMiB/SHEROMOLECACHESUTINTRO WAS 12ms
© 2015 Storage Networking Industry Association. All Rights Reserved. 31 END SUT INTRO CACHE MOLE MiB/SMiB/S
© 2015 Storage Networking Industry Association. All Rights Reserved. Back to the Original Problem … 32 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT Response Time ( ms) D O L ESS W ORK ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. Avoid Slow Parts And Generate “Hero Numbers” 33 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT ADD MORE CACHE DON’T USE THIS Response Time ( ms) D O L ESS W ORK ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. Use More Cache 34 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT Response Time ( ms) WAS 12ms CONTROLLER BOUND D ISK L ATENCY IS S TILL 11 MS B UT M ASKED D O L ESS W ORK ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. Caching Isn’t Just For Slow Drives 35 3,000 MILES ~70 MS LATENCY CLOUD STORAGE CONTROLLER BACK - END CONNECT ENDMiB/SMiB/SHEROMOLECACHESUTINTRO D O L ESS W ORK
© 2015 Storage Networking Industry Association. All Rights Reserved. 36 END SUT INTRO CACHE MOLE HERO
© 2015 Storage Networking Industry Association. All Rights Reserved. Latency Isn’t Everything 37 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT STILL A BOTTLENECK HOW? Response Time ( ms) ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. MiB/s Limits 38 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT NOT ENOUGH DISKS MiB/s “M ORE ” IS BETTER ENDMiB/SMiB/SHEROMOLECACHESUTINTRO I NCREASE P ARALLELISM
© 2015 Storage Networking Industry Association. All Rights Reserved. MiB/s Limits 39 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT MiB/s “M ORE ” IS BETTER ENDMiB/SMiB/SHEROMOLECACHESUTINTRO I NCREASE P ARALLELISM BOTTLENECK MOVES
© 2015 Storage Networking Industry Association. All Rights Reserved. Network Bound Or Design Problem 40 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT ENDMiB/SMiB/SHEROMOLECACHESUTINTRO I NCREASE P ARALLELISM BOTTLENECK
© 2015 Storage Networking Industry Association. All Rights Reserved. Network Bound Or Design Problem 41 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT I NCREASE P ARALLELISM ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. 42 SUT INTRO CACHE MOLE HEROMiB/SMiB/S
© 2015 Storage Networking Industry Association. All Rights Reserved. Which SUT Component Matters? 43 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. Which Component Matters? 44 CLIENTS / HOSTS STORAGE CONTROLLER PHYSICAL STORAGE FRONT - END CONNECT BACK - END CONNECT THE SLOWEST SUT COMPONENT MATTERS MOST THE SLOWEST SUT COMPONENT MATTERS MOST ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. Solution Under Test Review 45 DISK BOUND CLIENT BOUND SLOW COMPONENT MATTERS MOST BOTTLENECKS ALWAYS EXIST 3 PERFORMANCE PRINCIPLES I NCREASE P ARALLELISM D O L ESS W ORK D O W ORK F ASTER ENDMiB/SMiB/SHEROMOLECACHESUTINTRO
© 2015 Storage Networking Industry Association. All Rights Reserved. Storage Performance Benchmarking 46 METRICS AND TERMINOLOGY FILE COMPONENTS BLOCK COMPONENTS WORKLOAD DEFINITIONS TODAY FUTURE WEBCASTSJULY 30, 2015 SOLUTION UNDER TEST ENDSUTINTROCACHEMOLEHEROMiB/SMiB/S
© 2015 Storage Networking Industry Association. All Rights Reserved. After This Webcast 47 A PDF and a PPT of the slides for this and all previous parts of this Webcast series will be posted to the SNIA Ethernet Storage Forum (ESF) website and available on-demand PPT and PDF: Storage Performance Benchmarking: Part 1 Recording: A full Q&A from this webcast, including answers to questions we couldn't get to today, will be posted to the SNIA-ESF blog Follow us Next Webcast – First Quarter 2016 “Storage Performance Benchmarking: Part 3”