Characterizing Load Imbalance in Real-World Networked Caches Qi - PowerPoint PPT Presentation

Characterizing Load Imbalance in Real-World Networked Caches Qi Huang Cornell U, Facebook Helga Gudmundsdottir Emory U, Reykjavik U Ymir Vigfusson Emory U, Reykjavik U Daniel A. Freedman Technion Ken Birman Cornell U Robbert van Renesse Cornell U

Networked Caches in Real-World Web Stack Web Servers Get X X Caches Get X X Backend Store

Partitioning Data Hash(A) � id: A type: USER name: Helga

Partitioning Data Hash(1) � Shard 1 A

Causes of Load Imbalance • Hashing schemes balance quantity of content per server. • However, content popularity varies! • Need to provision servers to handle peak load. • Lack of real-world data to verify suspicions.

Questions • What is the state of load imbalance? • What contributes to load imbalance? • How effective are current techniques? • How might they be improved? Sampled production traffic in TAO , serving Facebook’s social graph.

What is the State of Load Imbalance? Significant load imbalance observed in TAO.

Possible Causes: Content Popularity Skewed content popularity observed at both shard and object level.

Possible Causes: Hot Objects Very hot objects alone are not a major cause of load imbalance.

Possible Causes: Hot Shards �� Popular shards receive significantly higher load, compared to hot objects.

How Effective are Current Techniques? • Hashing: Balance number of shards across servers. • Replication: Divide load of popular shards among servers. Hashing Replication libketama TAO Perfect None TAO Perfect • Metric: Maximum load versus average .

Where We Are Today �� Hashing Replication libketama TAO Perfect None 1.53 TAO Perfect

Where We Are Today �� Hashing Replication libketama TAO Perfect None 1.53 TAO Perfect 1.00

Where We Are Today �� Hashing Replication libketama TAO Perfect None 1.53 TAO 1.25 Perfect 1.00

Hashing Schemes w/o Replication �� Hashing Replication libketama TAO Perfect None 1.53 TAO 1.25 Perfect 1.00

Hashing Schemes w/o Replication �� Hashing Replication libketama TAO Perfect None 1.53 1.46 TAO 1.25 Perfect 1.00

Hashing Schemes w/o Replication �� Hashing Replication libketama TAO Perfect None 1.53 1.46 1.34 TAO 1.25 Perfect 1.00

libketama Hashing �� Hashing Replication libketama TAO Perfect None 1.53 1.46 1.34 TAO 1.25 Perfect 1.00

libketama Hashing �� Hashing Replication libketama TAO Perfect None 1.53 1.46 1.34 TAO 1.53 1.25 Perfect 1.00

libketama Hashing �� Hashing Replication libketama TAO Perfect None 1.53 1.46 1.34 TAO 1.53 1.25 Perfect 1.41 1.00

TAO: Room For Improvement �� Hashing Replication libketama TAO Perfect None 1.53 1.46 1.34 TAO 1.53 1.25 Perfect 1.41 1.00

TAO: Room For Improvement �� Hashing Replication libketama TAO Perfect None 1.53 1.46 1.34 TAO 1.53 1.25 Perfect 1.41 1.18 1.00

TAO: Room For Improvement �� Hashing Replication libketama TAO Perfect None 1.53 1.46 1.34 TAO 1.53 1.25 1.17 Perfect 1.41 1.18 1.00

Streaming Replication �� Hashing Replication libketama TAO Perfect None 1.53 1.46 1.34 TAO 1.53 1.25 1.17 Perfect 1.41 1.18 1.00

Streaming Replication Streaming with �� Perfect Hashing �� achieves max/avg 1.12 �� Hashing Replication libketama TAO Perfect None 1.53 1.46 1.34 TAO 1.53 1.25 1.17 Perfect 1.41 1.18 1.00

Summary and Future Work • Characterized how hashing and replication affect load imbalance. • Can streaming algorithms replicate content before its popularity surges? • Can we predict popularity spikes and prevent hotspots? Questions?

Characterizing Load Imbalance in Real-World Networked Caches Qi - PowerPoint PPT Presentation

Characterizing Load Imbalance in Real-World Networked Caches Qi Huang Cornell U, Facebook Helga Gudmundsdottir Emory U, Reykjavik U Ymir Vigfusson Emory U, Reykjavik U Daniel A. Freedman Technion Ken Birman Cornell U Robbert van Renesse

Detecting Application Load Imbalance on Cray Systems Heidi Poxon Technical Lead, Performance

Characterizing Human Mobility in Networked Virtual Environments Siqi Shen, Niels Brouwers,

CHARACTERIZING LOAD TRANSFER EFFICIENCY IN CARBON NANOTUBES NANOCOMPOSITES USING MULTISCALE

Fractals exercise Investigating task farms and load imbalance Reusing this material This work is

Fractals exercise Investigating task farms and load imbalance Reusing this material This work is

Simulating Real-world Load Patterns when playback just wont cut it Wayne Roseberry, M

GRAPH COLORING ON THE GPU AND SOME TECHNIQUES TO IMPROVE LOAD IMBALANCE SHUAI CHE, GREGORY

Load Imbalance Mitigation Optimizations for GPU-Accelerated Similarity Joins Benoit Gallet,

XML in the real world XML in the real world XML agentzh ( )

FreeSWITCH SIP and WebRTC Load Balancing and High Availability in Real World Giovanni

Where do you keep your photos? Personal Information Management in a Networked World

Approximate Computing Nikolai Lenney jlenney Charles Li cli4 18-742 S20 Load Value

KARMA KARMA P2P lending platform Investor presentation PREMISES Investment Imbalance

Characterizing Pixel Tracking through the Lens of Disposable Email Services Hang Hu, Peng Peng,

Decision on imbalance conformance enhancements proposal Greg Cook Director, Market and

Decision on Allocation of Real-time Imbalance Energy Offset Costs Greg Cook Manager, Market

Transient Fault Detection and Reducing Transient Error Rate Jose Lugo-Martinez CSE 240C:

Project TIER: Teaching Transparency in Empirical Research Richard Ball Associate Professor of

Strong Consistency & CAP Theorem CS 240: Computing Systems and Concurrency Lecture 15 Marco

A thin arbiter for glusterfs replication Ravishankar N. (@itisravi) Sr.Software Engineer,

Built-in Physical and Logical Replication in Postgresql Frat Gle - Company Hepsiexpress

Asynchronous Replication and Bayou Asynchronous Replication and Bayou Asynchronous Replication

Capacity Planning and Headroom Analysis for Taming Database Replication Latency - Experiences

Scaling State Machine Replication Fernando Pedone University of Lugano (USI) Switzerland State

Characterizing Load Imbalance in Real-World Networked Caches Qi - PowerPoint PPT Presentation

Characterizing Load Imbalance in Real-World Networked Caches Qi Huang Cornell U, Facebook Helga Gudmundsdottir Emory U, Reykjavik U Ymir Vigfusson Emory U, Reykjavik U Daniel A. Freedman Technion Ken Birman Cornell U Robbert van Renesse

Detecting Application Load Imbalance on Cray Systems Heidi Poxon Technical Lead, Performance

Characterizing Human Mobility in Networked Virtual Environments Siqi Shen, Niels Brouwers,

CHARACTERIZING LOAD TRANSFER EFFICIENCY IN CARBON NANOTUBES NANOCOMPOSITES USING MULTISCALE

Fractals exercise Investigating task farms and load imbalance Reusing this material This work is

Fractals exercise Investigating task farms and load imbalance Reusing this material This work is

Simulating Real-world Load Patterns when playback just wont cut it Wayne Roseberry, M

GRAPH COLORING ON THE GPU AND SOME TECHNIQUES TO IMPROVE LOAD IMBALANCE SHUAI CHE, GREGORY

Load Imbalance Mitigation Optimizations for GPU-Accelerated Similarity Joins Benoit Gallet,

XML in the real world XML in the real world XML agentzh ( )

FreeSWITCH SIP and WebRTC Load Balancing and High Availability in Real World Giovanni

Where do you keep your photos? Personal Information Management in a Networked World

Approximate Computing Nikolai Lenney jlenney Charles Li cli4 18-742 S20 Load Value

KARMA KARMA P2P lending platform Investor presentation PREMISES Investment Imbalance

Characterizing Pixel Tracking through the Lens of Disposable Email Services Hang Hu, Peng Peng,

Decision on imbalance conformance enhancements proposal Greg Cook Director, Market and

Decision on Allocation of Real-time Imbalance Energy Offset Costs Greg Cook Manager, Market

Transient Fault Detection and Reducing Transient Error Rate Jose Lugo-Martinez CSE 240C:

Project TIER: Teaching Transparency in Empirical Research Richard Ball Associate Professor of

Strong Consistency &amp; CAP Theorem CS 240: Computing Systems and Concurrency Lecture 15 Marco

A thin arbiter for glusterfs replication Ravishankar N. (@itisravi) Sr.Software Engineer,

Built-in Physical and Logical Replication in Postgresql Frat Gle - Company Hepsiexpress

Asynchronous Replication and Bayou Asynchronous Replication and Bayou Asynchronous Replication

Capacity Planning and Headroom Analysis for Taming Database Replication Latency - Experiences

Scaling State Machine Replication Fernando Pedone University of Lugano (USI) Switzerland State

Strong Consistency & CAP Theorem CS 240: Computing Systems and Concurrency Lecture 15 Marco