Re-evaluating Measurement Algorithms in Software Omid Alipourfard, Masoud Moshref, Minlan Yu {alipourf, moshrefj,

Slides:



Advertisements
Similar presentations
IP Router Architectures. Outline Basic IP Router Functionalities IP Router Architectures.
Advertisements

Logically Centralized Control Class 2. Types of Networks ISP Networks – Entity only owns the switches – Throughput: 100GB-10TB – Heterogeneous devices:
A Search Memory Substrate for High Throughput and Low Power Packet Processing Sangyeun Cho, Michel Hanna and Rami Melhem Dept. of Computer Science University.
VCRIB: Virtual Cloud Rule Information Base Masoud Moshref, Minlan Yu, Abhishek Sharma, Ramesh Govindan HotCloud 2012.
Configuring a Load-Balanced Switch in Hardware Srikanth Arekapudi, Shang-Tse (Da) Chuang, Isaac Keslassy, Nick McKeown Stanford University.
OpenSketch Slides courtesy of Minlan Yu 1. Management = Measurement + Control Traffic engineering – Identify large traffic aggregates, traffic changes.
Estimating TCP Latency Approximately with Passive Measurements Sriharsha Gangam, Jaideep Chandrashekar, Ítalo Cunha, Jim Kurose.
PERSISTENT DROPPING: An Efficient Control of Traffic Aggregates Hani JamjoomKang G. Shin Electrical Engineering & Computer Science UNIVERSITY OF MICHIGAN,
Web Caching Schemes1 A Survey of Web Caching Schemes for the Internet Jia Wang.
1 Reversible Sketches for Efficient and Accurate Change Detection over Network Data Streams Robert Schweller Ashish Gupta Elliot Parsons Yan Chen Computer.
Efficient IP-Address Lookup with a Shared Forwarding Table for Multiple Virtual Routers Author: Jing Fu, Jennifer Rexford Publisher: ACM CoNEXT 2008 Presenter:
Reverse Hashing for High-speed Network Monitoring: Algorithms, Evaluation, and Applications Robert Schweller 1, Zhichun Li 1, Yan Chen 1, Yan Gao 1, Ashish.
Reverse Hashing for Sketch Based Change Detection in High Speed Networks Ashish Gupta Elliot Parsons with Robert Schweller, Theory Group Advisor: Yan Chen.
Translation Buffers (TLB’s)
Dream Slides Courtesy of Minlan Yu (USC) 1. Challenges in Flow-based Measurement 2 Controller Configure resources1Fetch statistics2(Re)Configure resources1.
DREAM: Dynamic Resource Allocation for Software-defined Measurement
A Scalable, Commodity Data Center Network Architecture Mohammad Al-Fares, Alexander Loukissas, Amin Vahdat Presented by Gregory Peaker and Tyler Maclean.
By- Jaideep Moses, Ravi Iyer , Ramesh Illikkal and
Software-defined Measurement
OpenFlow Switch Limitations. Background: Current Applications Traffic Engineering application (performance) – Fine grained rules and short time scales.
Software-Defined Networks Jennifer Rexford Princeton University.
Detail: Reducing the Flow Completion Time Tail in Datacenter Networks SIGCOMM PIGGY.
Practical Issues in Internet Measurement adapted from Mark Crovella and Balachander Krishnamurthy.
Log-structured Memory for DRAM-based Storage Stephen Rumble, John Ousterhout Center for Future Architectures Research Storage3.2: Architectures.
Resource/Accuracy Tradeoffs in Software-Defined Measurement Masoud Moshref, Minlan Yu, Ramesh Govindan HotSDN’13.
A Formal Analysis of Conservative Update Based Approximate Counting Gil Einziger and Roy Freidman Technion, Haifa.
Jennifer Rexford Princeton University MW 11:00am-12:20pm Measurement COS 597E: Software Defined Networking.
Boot Sequence, Internal Component & Cisco 3 Layer Model 1.
Efficient Cache Structures of IP Routers to Provide Policy-Based Services Graduate School of Engineering Osaka City University
Enabling a “RISC” Approach for Software-Defined Monitoring using Universal Streaming Vyas Sekar Zaoxing Liu, Greg Vorsanger, Vladimir Braverman.
SCREAM: Sketch Resource Allocation for Software-defined Measurement Masoud Moshref, Minlan Yu, Ramesh Govindan, Amin Vahdat (CoNEXT’15)
1 Building big router from lots of little routers Nick McKeown Assistant Professor of Electrical Engineering and Computer Science, Stanford University.
Open-source routing at 10Gb/s Olof Hagsand (KTH) Robert Olsson (Uppsala U) Bengt Görden (KTH) SNCNW May 2009 Project grants: Internetstiftelsen (IIS) Equipment:
SketchVisor: Robust Network Measurement for Software Packet Processing
NFP: Enabling Network Function Parallelism in NFV
Problem: Internet diagnostics and forensics
Architecture and Algorithms for an IEEE 802
FlowRadar: A Better NetFlow For Data Centers
Hydra: Leveraging Functional Slicing for Efficient Distributed SDN Controllers Yiyang Chang, Ashkan Rezaei, Balajee Vamanan, Jahangir Hasan, Sanjay Rao.
A Resource-minimalist Flow Size Histogram Estimator
Network Performance and Quality of Service
Data Streaming in Computer Networking
NOX: Towards an Operating System for Networks
Tapping Into The Unutilized Router Processing Power
Augmented Sketch: Faster and More Accurate Stream Processing
PA an Coordinated Memory Caching for Parallel Jobs
Load Balancing Memcached Traffic Using SDN
Srinivas Narayana MIT CSAIL October 7, 2016
NFP: Enabling Network Function Parallelism in NFV
Query-Friendly Compression of Graph Streams
Be Fast, Cheap and in Control
Optimal Elephant Flow Detection Presented by: Gil Einziger,
Qun Huang, Patrick P. C. Lee, Yungang Bao
NFP: Enabling Network Function Parallelism in NFV
SCREAM: Sketch Resource Allocation for Software-defined Measurement
Elastic Sketch: Adaptive and Fast Network-wide Measurements
Cloud computing mechanisms
Fast Congestion Control in RDMA-Based Datacenter Networks
Elastic Sketch: Adaptive and Fast Network-wide Measurements
Translation Buffers (TLB’s)
Ran Ben Basat, Xiaoqi Chen, Gil Einziger, Ori Rottenstreich
Translation Buffers (TLBs)
Lu Tang , Qun Huang, Patrick P. C. Lee
Toward Self-Driving Networks
Review What are the advantages/disadvantages of pages versus segments?
Toward Self-Driving Networks
Re-evaluating Measurement Algorithms in Software
Elmo Muhammad Shahbaz Lalith Suresh, Jennifer Rexford, Nick Feamster,
NitroSketch: Robust and General Sketch-based Monitoring in Software Switches Alan (Zaoxing) Liu Joint work with Ran Ben-Basat, Gil Einziger, Yaron Kassner,
2019/11/12 Efficient Measurement on Programmable Switches Using Probabilistic Recirculation Presenter:Hung-Yen Wang Authors:Ran Ben Basat, Xiaoqi Chen,
Presentation transcript:

Re-evaluating Measurement Algorithms in Software Omid Alipourfard, Masoud Moshref, Minlan Yu {alipourf, moshrefj,

Software Switches are Popular Data centers: “Use the cloud to manage the cloud” Load balancer and Firewall on VMs ISPs: AT&T replaces hardware routers with NFV Allow customers to run network functions on commodity servers 2

Network Function Virtualization 3 Typical X86 server NATIDS DPIFirewall

Measurement is Critical for NFVs Make decisions based on measurement input Firewall, load balancing, intrusion detection systems (IDS) We also need measurement for managing NFVs Profiling NFV usage, resource scheduling.. 4 “If you can’t measure it, you can’t manage it”

What is the best algorithm for measurement in software? 5

New Design Concerns in Software 6 Software switches Working-set (Cache size) Maximizing Throughput Hardware switches Limited Memory Size Fit in the Memory Domain Constraint Objective It’s time to reevaluate measurement algorithms in software context!

Previous Works on Measurement Algorithms 7 Measurement task Heavy-hitters detection Super-spreaders detection Flow-size distribution Change detection Entropy estimation Quantiles

Measurement taskSketchHeap/tree Heavy-hitters detection Super-spreaders detection Flow-size distribution Change detection Entropy estimation Quantiles Measurement Algorithms 8 Popular measurement algorithms in the hardware switch and database domain

Measurement taskSketchHeap/tree Heavy-hitters detectionNSDI’ 13, JoA’ 05ICDT’ 05, ANCS’ 11 Super-spreaders detectionNSDI’ 13, PODS’ 05 Flow-size distributionSIGMETRICS’ 04 Change detectionTON’ 07CONEXT’ 13 Entropy estimationCOLT’ 11 QuantilesHot ICE’ 11SIGMOD’ 99, 01, 13 Measurement Tasks 9

Measurement taskSketchHeap/tree Heavy-hitters detectionNSDI’ 13, JoA’ 05ICDT’ 05, ANCS’ 11 Super-spreaders detectionNSDI’ 13, PODS’ 05 Flow-size distributionSIGMETRICS’ 04 Change detectionTON’ 07CONEXT’ 13 Entropy estimationCOLT’ 11 QuantilesHot ICE’ 11SIGMOD’ 99, 01, 13 Measurement Tasks 10

Algorithms for Heavy-hitter Detection Heavy hitters: Detect the most popular flows in the traffic. Different memory-computation tradeoffs: Simple hash table: For every packet, hash the header, update a counter Sketches: For every packet, hash the header several times, update relevant counters. Heap: Keep a heap of counters, remove smallest counter when there is no space for new packets. 11 Precision MemoryProcessing Precision Memory Processing Different trade offs for algorithms: which one is the best?

Compare different measurement algorithms Hash, sketch, heap Metrics Performance: network throughput, per-packet delay Precision: % of selected heavy-hitters that are correct. Evaluation setting CAIDA traces Click modular router modified with DPDK 12 Evaluating the algorithms

Simple Hash Table Works the Best Throughput Simple hash table has 34% higher throughput over sketches, and 108% over heap. Precision Simple hash table is only 4% less accurate for sizes greater than 200KB. The accuracy difference is less than 1% when the size is greater than 10MB. 13

Hash has the lowest average per packet delay 14 1 hash + 2 mem. accesses 3 hashes + 6 mem. accessesO(log(N)) memory accesses

Tail Latency Jump from L3 to Memory 15

Simple hash table works well for heavy-hitter detection. What about the other tasks? Observation All measurement tasks maintain a table of entries (e.g., counters). Gbps, 67 ns to process each packet on average (worst case).

17 L3 Memory Memory access is too slow 16 ns 100 ns

Plenty of memory for measurement #entries that can be used for line-rate packet processing ~5 Mil NetFlow Entries ~15 Mil Hash Entries with Distinct counters ~200 Mil Hash Entries 18

Enough memory to hold everything! 5 Mil NetFlow Entries 15 Mil Hash Entries with Distinct counters 200 Mil Hash Entries Average Flow size of 1KB  Gbps More than enough space to hold everything! 19

Effects of Skew and Multicore Different skews of traffic Simple hash table result is consistently the best Sharing data across multiple cores Because of cache contention, performance drops 20

Future Work 21 Where simple hash table fails? Solution to the failed cases Improving the model

Conclusion NFV is the new trend in data-centers and ISPs Measurement is a key component for NFV Simple hash table works best For many tasks, the working set fits in the cache Especially when the traffic is skewed We expect this to be true in the future with newer servers Larger cache, better efficient cache 22