Random Notes
  • Introduction
  • Reading list
  • Theory
    • Index
      • Impossibility of Distributed Consensus with One Faulty Process
      • Time, Clocks, and the Ordering of Events in a Distributed System
      • Using Reasoning About Knowledge to analyze Distributed Systems
      • CAP Twelve Years Later: How the “Rules” Have Changed
      • A Note on Distributed Computing
  • Operating System
    • Index
  • Storage
    • Index
      • Tachyon: Reliable, Memory Speed Storage for Cluster Computing Frameworks
      • Exploiting Commutativity For Practical Fast Replication
      • Don’t Settle for Eventual: Scalable Causal Consistency for Wide-Area Storage with COPS
      • Building Consistent Transactions with Inconsistent Replication
      • Managing Update Conflicts in Bayou, a Weakly Connected Replicated Storage System
      • Spanner: Google's Globally-Distributed Database
      • Bigtable: A Distributed Storage System for Structured Data
      • The Google File System
      • Dynamo: Amazon’s Highly Available Key-value Store
      • Chord: A Scalable Peer-to-peer Lookup Service for Internet Applications
      • Replicated Data Consistency Explained Through Baseball
      • Session Guarantees for Weakly Consistent Replicated Data
      • Flat Datacenter Storage
      • Small Cache, Big Effect: Provable Load Balancing forRandomly Partitioned Cluster Services
      • DistCache: provable load balancing for large-scale storage systems with distributed caching
      • Short Summaries
  • Coordination
    • Index
      • Logical Physical Clocks and Consistent Snapshots in Globally Distributed Databases
      • Paxos made simple
      • ZooKeeper: Wait-free coordination for Internet-scale systems
      • Just Say NO to Paxos Overhead: Replacing Consensus with Network Ordering
      • Keeping CALM: When Distributed Consistency is Easy
      • In Search of an Understandable Consensus Algorithm
      • A comprehensive study of Convergent and Commutative Replicated Data Types
  • Fault Tolerance
    • Index
      • The Mystery Machine: End-to-end Performance Analysis of Large-scale Internet Services
      • Gray Failure: The Achilles’ Heel of Cloud-Scale Systems
      • Capturing and Enhancing In Situ System Observability for Failure Detection
      • Check before You Change: Preventing Correlated Failures in Service Updates
      • Efficient Scalable Thread-Safety-Violation Detection
      • REPT: Reverse Debugging of Failures in Deployed Software
      • Redundancy Does Not Imply Fault Tolerance
      • Fixed It For You:Protocol Repair Using Lineage Graphs
      • The Good, the Bad, and the Differences: Better Network Diagnostics with Differential Provenance
      • Lineage-driven Fault Injection
      • Short Summaries
  • Cloud Computing
    • Index
      • Improving MapReduce Performance in Heterogeneous Environments
      • CLARINET: WAN-Aware Optimization for Analytics Queries
      • MapReduce: Simplified Data Processing on Large Clusters
      • Dryad: Distributed Data-Parallel Programs from Sequential Building Blocks
      • Resource Management
      • Apache Hadoop YARN: Yet Another Resource Negotiator
      • Mesos: A Platform for Fine-Grained Resource Sharing in the Data Center
      • Dominant Resource Fairness: Fair Allocation of Multiple Resource Types
      • Large-scale cluster management at Google with Borg
      • MapReduce Online
      • Delay Scheduling: A Simple Technique for Achieving Locality and Fairness in Cluster Scheduling
      • Reining in the Outliers in Map-Reduce Clusters using Mantri
      • Effective Straggler Mitigation: Attack of the Clones
      • Resilient Distributed Datasets: A Fault-Tolerant Abstraction for In-Memory Cluster Computing
      • Discretized Streams: Fault-Tolerant Streaming Computation at Scale
      • Sparrow: Distributed, Low Latency Scheduling
      • Making Sense of Performance in Data Analytics Framework
      • Monotasks: Architecting for Performance Clarity in Data Analytics Frameworks
      • Drizzle: Fast and Adaptable Stream Processing at Scale
      • Naiad: A Timely Dataflow System
      • The Dataflow Model:A Practical Approach to Balancing Correctness, Latency, and Cost in Massive-Scale
      • Interruptible Tasks:Treating Memory Pressure AsInterrupts for Highly Scalable Data-Parallel Program
      • PACMan: Coordinated Memory Caching for Parallel Jobs
      • Multi-Resource Packing for Cluster Schedulers
      • Other interesting papers
  • Systems for ML
    • Index
      • A Berkeley View of Systems Challenges for AI
      • Tiresias: A GPU Cluster Managerfor Distributed Deep Learning
      • Gandiva: Introspective Cluster Scheduling for Deep Learning
      • Workshop papers
      • Hidden Technical Debt in Machine Learning Systems
      • Inference Systems
      • Parameter Servers and AllReduce
      • Federated Learning at Scale - Part I
      • Federated Learning at Scale - Part II
      • Learning From Non-IID data
      • Ray: A Distributed Framework for Emerging AI Applications
      • PipeDream: Generalized Pipeline Parallelism for DNN Training
      • DeepXplore: Automated Whitebox Testingof Deep Learning Systems
      • Distributed Machine Learning Misc.
  • ML for Systems
    • Index
      • Short Summaries
  • Machine Learning
    • Index
      • Deep Learning with Differential Privacy
      • Accelerating Deep Learning via Importance Sampling
      • A Few Useful Things to Know About Machine Learning
  • Video Analytics
    • Index
      • Scaling Video Analytics on Constrained Edge Nodes
      • Focus: Querying Large Video Datasets with Low Latency and Low Cost
      • NoScope: Optimizing Neural Network Queriesover Video at Scale
      • Live Video Analytics at Scale with Approximation and Delay-Tolerance
      • Chameleon: Scalable Adaptation of Video Analytics
      • End-to-end Learning of Action Detection from Frame Glimpses in Videos
      • Short Summaries
  • Networking
    • Index
      • Salsify: Low-Latency Network Video through Tighter Integration between a Video Codec and a Transport
      • Learning in situ: a randomized experiment in video streaming
      • Short Summaries
  • Serverless
    • Index
      • Serverless Computing: One Step Forward, Two Steps Back
      • Encoding, Fast and Slow: Low-Latency Video Processing Using Thousands of Tiny Threads
      • SAND: Towards High-Performance Serverless Computing
      • Pocket: Elastic Ephemeral Storage for Serverless Analytics
      • Fault-tolerant and Transactional Stateful Serverless Workflows
  • Resource Disaggregation
    • Index
  • Edge Computing
    • Index
  • Security/Privacy
    • Index
      • Differential Privacy
      • Honeycrisp: Large-Scale Differentially Private Aggregation Without a Trusted Core
      • Short Summaries
  • Misc.
    • Index
      • Rate Limiting
      • Load Balancing
      • Consistency Models in Distributed System
      • Managing Complexity
      • System Design
      • Deep Dive into the Spark Scheduler
      • The Actor Model
      • Python Global Interpreter Lock
      • About Research and PhD
Powered by GitBook
On this page
  • Requirements for Ephemeral Storage
  • The design of Pocket
  • Rightsizing Resource Allocations

Was this helpful?

  1. Serverless
  2. Index

Pocket: Elastic Ephemeral Storage for Serverless Analytics

https://www.usenix.org/conference/osdi18/presentation/klimovic

PreviousSAND: Towards High-Performance Serverless ComputingNextFault-tolerant and Transactional Stateful Serverless Workflows

Last updated 4 years ago

Was this helpful?

Exchanging intermediate(ephemeral) data between execution stages(e.g., map and reduce) is a key challenge as direct communication between serverless tasks is difficult. Pocket is an elastic, distributed data store that provides serverless applications with the desired performance at a low cost.

Requirements for Ephemeral Storage

  • High performance for a wide range of object sizes: Ephemeral data access granularity varies greatly in size, ranging from hundreds of bytes to hundreds of megabytes.

  • Automatic and fine-grain scaling: An ephemeral data store can observe a storm of I/O requests within a fraction of a second. Thus, it must automatically rightsize resources to satisfy application I/O requirements while minimizing cost.

  • Storage technology awareness: The variety of storage media available in the cloud allow for different performance-cost trade-offs. An ephemeral data store must place application data on the right storage tier(s) for performance and cost efficiency.

  • Fault-(in)tolerance: Because ephemeral data has a short lifetime of 10-100s of seconds, an ephemeral storage does not have to provide high fault-tolerance as expected of traditional storage systems

However, existing storage systems do not satisfy the above combination of requirements, as shown in table 1.

The design of Pocket

Pocket consists of a logically centralized controller, one or more metadata servers, and multiple data plane storage servers. The three places can be scaled independently based on variations in load.

Controller

The controller allocates storage resources for jobs and dynamically scales Pocket metadata and storage nodes up and down as the number of jobs and their requirements vary over time. The controller also makes data placement decisions for jobs (i.e., which nodes and storage media to use for a job’s data).

Metadata Servers

Metadata servers enforce coarse-grain data placement policies generated by the controller by steering client requests to appropriate storage servers. Pocket’s metadata plane manages data at the granularity of blocks, whose size is configurable (64k by default). Objects larger than the block size are divided into blocks and distributed across storage servers, enabling Pocket to support arbitrary object sizes.

Rightsizing Resource Allocations

By default, Pocket uses a default allocation that conservatively over-provisions resources to achieve high performance. However, applications can provide hints about the job characteristics, such as its latency sensitivity, the maximum number of concurrent lambdas, and peak aggregate bandwidth. The controller uses these hints to determine a job's resource allocations (throughput, capacity, and the choice of storage media) to meet its requirements while minimizing the cost. For example, knowing a job's maximum number of concurrent lambdas allows Pocket to compute a less conservative estimate of the job's throughput requirement.

Pocket translates a job’s resource allocation into a resource assignment on specific storage servers by generating a weight map for the job. The weight map is an associative array mapping each storage server (identified by its IP address and port) to a weight from 0 to 1, which represents the fraction of a job’s dataset to place on that storage server