Unevenly Distributed Adrian Colyer @adriancolyer blog.acolyer.org - PowerPoint PPT Presentation

Unevenly Distributed Adrian Colyer @adriancolyer

blog.acolyer.org 350 Foundations Frontiers

5 Reasons to <3 Papers 03 Applied rainstorm 04 Lessons The Great 02 Conversation Raise Expectations 05 01 Uneven Thinking Brain Distribution tools storm 3

Scalability - but at what COST? Frank McSherry 4

5

But you have BIG Data! “Working sets are Zipf- distributed. We can therefore store in memory all but the very largest datasets.” Zipf Distribution 6

Musketeer One for all? 7

Approx Hadoop 32x! 8

The Scalable Commutativity Rule Improve your API Design 9

Raising Your Expectations 10

TLS CVEs 54 Jan ‘14 - Jan ‘15 ! Error prone languages ! Lack of Separation ! Ambiguous and Untestable Spec Surely we can do better? 11

Do Less Testing! Microsoft Windows 8.1 Relative Improvement Cost Improvement Test Executions 40.58% Test Time 40.31% $1,567,608 Test Result Inspection 33.04% $61,533 Escaped Defects 0.20% ($11,971) Total Cost Balance $1,617,170 12

13

Lessons from the Field 14

A Masterclass in Config Mgt at Facebook 15

Machine Learning Systems lessons from Google Feature 01 Management 02 Visualisation Relative Metrics 03 Systematic Bias 04 Correction Alerts on action 05 Thresholds 16

The Great Conversation And the Syntopicon 17

Cross-Fertilization Broad Exposure to Problems and their Solutions Security Robotics Distributed Machine Learning Systems Databases Programming Languages And Many More Operating Systems, Algorithms, Networking,Optimisation, SW Engineering,... 18

TPC-C - 1992 19

TPC-C Published Record Holder Date Mar 26th 2013 Database Manager Oracle 11g r2 Enterprise Edition w. Partitioning Performance (tpmC) 8,552,523 (8.5M) Performance (tps) 142,542 (143K) System Cost $4,663,073 #Processors 8 #Cores 128 #Threads 1024 20

Coordination Avoidance and I-Confluence Analysis TPC-C 21

Multi-Partition Transactions at Scale 22

Unevenly Distributed Turning your world Upside Down

Human computers at Dryden by NACA (NASA) - Dryden Flight Research Center Photo Collection http://www.dfrc.nasa. gov/Gallery/Photo/Places/HTML/E49-54.html. Licensed under Public Domain via Commons - https://commons.wikimedia.org/wiki/File: Human_computers_-_Dryden.jpg#/media/File: Human_computers_-_Dryden.jpg

Computing on a Human Scale Registers File on 10ns 10s & L1-L3 desk Office filing 1:10s 70ns Main cabinet memory 116d Trip to the 10ms HDD warehouse 25

All Change Please Next Generation Hardware Compute Networking HTM 100GbE Persistent Memory NI RDMA FPGA GPUs Memory Storage NVDIMMs NVMe Persistent Memory Next-gen NVM 26

Computing on a Human Scale 4x capacity File on 10s 2-10m fireproof local desk filing cabinets Phone Office filing 1:10s 23-40 m another office cabinet (RDMA) Next-gen Trip to the 116d 3h20m warehouse warehouse 27

The New ~Numbers Everyone Should Know Latency Bandwidth Capacity/IOPS Register 0.25ns L1 cache 1ns L2 cache 3ns 8MB L3 cache 11ns 45MB DRAM 62ns 120GBs 6TB - 4 socket NVRAM’ DIMM 620ns 60GBs 24TB - 4 socket 1-sided RDMA in Data Center 1.4us 100GbE ~700K IOPS RPC in Data Center 2.4us 100GbE ~400K IOPS NVRAM’ NVMe 12us 6GBs 16TB/disk,~2M/600K NVRAM’ NVMf 90us 5GBs 16TB/disk, ~700/600K 28

Low Latency - RAMCloud 5 μ s Reads 13.5 μ s Writes 20 μ s Transactions 27 μ s 5-object Txns 35K tps TPC-C (10 nodes) 29

No Compromises - FaRM 4.5M tps TPC-C (90 nodes) 1.9ms 99%ile 6.3M qps KV (per node) 41 μ s at peak throughput 30

No Compromises “This paper demonstrates that new software in modern data centers can eliminate the need to compromise. It describes the transaction, replication, and recovery protocols in FaRM, a main memory distributed computing platform. FaRM provides distributed ACID transactions with strict serializability, high availability, high throughput and low latency. These protocols were designed from first principles to leverage two hardware trends appearing in data centers: fast commodity networks with RDMA and an inexpensive approach to providing non-volatile DRAM .” 31

The Doctor will see you now DrTM 5.5M tps on TPC-C 6-node cluster. 32

Some things Change, Some stay the Same 33

A Brave New World Fast RDMA networks + Ample Persistent Memory + Hardware Transactions + Enhanced HW Cache Management + Super-fast Storage + On-board FPGAs + GPUs + … = ??? 34

5 Reasons to <3 Papers 03 Applied rainstorm 04 Lessons The Great 02 Conversation Raise Expectations 05 01 Uneven Thinking Brain Distribution tools storm 35

01 A new paper every weekday Published at http://blog.acolyer.org. 02 Delivered Straight to your inbox If you prefer email-based subscription to read at your leisure. 03 Announced on Twitter I’m @adriancolyer. 04 Go to a Papers We Love Meetup A repository of academic computer science papers and a community who loves reading them. 05 Share what you learn Anyone can take part in the great conversation.

THANK YOU ! @adriancolyer