Rui's Blog
  • Rui's Blog/Paper Reading Notes - Introduction
  • Personal Blog
    • Personal Blog - Index
      • How to Create Picture-in-Picture Effect / Video Overlay for a Presentation Video
      • How to Do Your Part to Protect the Environment in Wisconsin
      • How to Get a Driver's License in Wisconsin
      • How to Travel from the U.S. to China onboard AA127 in June 2021
      • How to Transfer Credits Back to UW-Madison
      • Resources on Learning Academic Writing (for Computer Science)
    • Towards applying to CS Ph.D. programs
  • Machine Learning Systems
    • Machine Learning Systems - Index
      • MLSys Papers - Short Notes
      • [2011 NSDI] Dominant Resource Fairness: Fair Allocation of Multiple Resource Types
      • [2014 OSDI] Scaling Distributed Machine Learning with the Parameter Server
      • [2018 OSDI] Gandiva: Introspective Cluster Scheduling for Deep Learning
      • [2018 SIGCOMM] Chameleon: Scalable Adaptation of Video Analytics via Temporal and Cross-camera ...
      • [2018 NIPS] Dynamic Space-Time Scheduling for GPU Inference
      • [2019 ATC] Analysis of Large-Scale Multi-Tenant GPU Clusters for DNN Training Workloads
      • [2019 NSDI] Tiresias: A GPU Cluster Manager for Distributed Deep Learning
      • [2019 SOSP] ByteScheduler: A Generic Communication Scheduler for Distributed DNN Training ...
      • [2019 SOSP] PipeDream: Generalized Pipeline Parallelism for DNN Training
      • [2019 SOSP] Parity Models: Erasure-Coded Resilience for Prediction Serving Systems
      • [2019 NIPS] GPipe: Efficient Training of Giant Neural Networks using Pipeline Parallelism
      • [2019 SC] ZeRO: memory optimizations toward training trillion parameter models
      • [2020 OSDI] Gavel: Heterogeneity-Aware Cluster Scheduling Policies for Deep Learning Workloads
      • [2020 OSDI] AntMan: Dynamic Scaling on GPU Clusters for Deep Learning
      • [2020 OSDI] BytePS: A High Performance and Generic Framework for Distributed DNN Training
      • [2020 SIGCOMM] Reducto: On-Camera Filtering for Resource-Efficient Real-Time Video Analytics
        • [2020 MLSys] Salus: Fine-Grained GPU Sharing Primitives for Deep Learning Applications
      • [2020 EuroSys] AlloX: Compute Allocation in Hybrid Clusters
      • [2020 VLDB] PyTorch Distributed: Experiences on Accelerating Data Parallel Training
      • [2020 NetAI] Is Network the Bottleneck of Distributed Training?
      • [2020 NSDI] Themis: Fair and Efficient GPU Cluster Scheduling
      • [2021 MLSys] Accordion: Adaptive Gradient Communication via Critical Learning Regime Identification
      • [2021 VLDB] Analyzing and Mitigating Data Stalls in DNN Training
      • [2021 FAST] CheckFreq: Frequent, Fine-Grained DNN Checkpointing
      • [2021 EuroMLSys] Interference-Aware Scheduling for Inference Serving
      • [2021 OSDI] Pollux: Co-adaptive Cluster Scheduling for Goodput-Optimized Deep Learning
      • [2021 MLSys] Wavelet: Efficient DNN Training with Tick-Tock Scheduling
      • [2021 NSDI] SwitchML: Scaling Distributed Machine Learning with In-Network Aggregation
    • Big Data Systems - Index
      • Big Data Systems Papers - Short Notes
      • [2003 SOSP] The Google File System
      • [2004 OSDI] MapReduce: Simplified Data Processing on Large Clusters
      • [2010 SIGMOD] Pregel: A System for Large-Scale Graph Processing
      • [2011 NSDI] Mesos: A Platform for Fine-Grained Resource Sharing in the Data Center
      • [2012 NSDI] Resilient Distributed Datasets: A Fault-Tolerant Abstraction for In-Memory Cluster ...
      • [2012 OSDI] PowerGraph: Distributed Graph-Parallel Computation on Natural Graphs
      • [2019 FAST] DistCache: Provable Load Balancing for Large-Scale Storage Systems with Distributed...
      • [2021 HotOS] From Cloud Computing to Sky Computing
      • [2021 EuroSys] NextDoor: Accelerating graph sampling for graph machine learning using GPUs
  • Earlier Readings & Notes
    • High Performance Computing Course Notes
      • Lecture 1: Course Overview
      • Lecture 2: From Code to Instructions. The FDX Cycle. Instruction Level Parallelism.
      • Lecture 3: Superscalar architectures. Measuring Computer Performance. Memory Aspects.
      • Lecture 4: The memory hierarchy. Caches.
      • Lecture 5: Caches, wrap up. Virtual Memory.
      • Lecture 6: The Walls to Sequential Computing. Moore’s Law.
      • Lecture 7: Parallel Computing. Flynn's Taxonomy. Amdahl's Law.
      • Lecture 8: GPU Computing Intro. The CUDA Programming Model. CUDA Execution Configuration.
      • Lecture 9: GPU Memory Spaces
      • Lecture 10: GPU Scheduling Issues.
      • Lecture 11: Execution Divergence. Control Flow in CUDA. CUDA Shared Memory Issues.
      • Lecture 12: Global Memory Access Patterns and Implications.
      • Lecture 13: Atomic operations in CUDA. GPU ode optimization rules of thumb.
      • Lecture 14: CUDA Case Studies. (1) 1D Stencil Operation. (2) Vector Reduction in CUDA.
      • Lecture 15: CUDA Case Studies. (3) Parallel Prefix Scan on the GPU. Using Multiple Streams in CUDA.
      • Lecture 16: Streams, and overlapping data copy with execution.
      • Lecture 17: GPU Computing: Advanced Features.
      • Lecture 18: GPU Computing with thrust and cub.
      • Lecture 19: Hardware aspects relevant in multi-core, shared memory parallel computing.
      • Lecture 20: Multi-core Parallel Computing with OpenMP. Parallel Regions.
      • Lecture 21: OpenMP Work Sharing.
      • Lecture 22: OpenMP Work Sharing
      • Lecture 23: OpenMP NUMA Aspects. Caching and OpenMP.
      • Lecture 24: Critical Thinking. Code Optimization Aspects.
      • Lecture 25: Computing with Supercomputers.
      • Lecture 26: MPI Parallel Programming General Introduction. Point-to-Point Communication.
      • Lecture 27: MPI Parallel Programming Point-to-Point communication: Blocking vs. Non-blocking sends.
      • Lecture 28: MPI Parallel Programming: MPI Collectives. Overview of topics covered in the class.
    • Cloud Computing Course Notes
      • 1.1 Introduction to Clouds, MapReduce
      • 1.2 Gossip, Membership, and Grids
      • 1.3 P2P Systems
      • 1.4 Key-Value Stores, Time, and Ordering
      • 1.5 Classical Distributed Algorithms
      • 4.1 Spark, Hortonworks, HDFS, CAP
      • 4.2 Large Scale Data Storage
    • Operating Systems Papers - Index
      • CS 736 @ UW-Madison Fall 2020 Reading List
      • All File Systems Are Not Created Equal: On the Complexity of Crafting Crash-Consistent Applications
      • ARC: A Self-Tuning, Low Overhead Replacement Cache
      • A File is Not a File: Understanding the I/O Behavior of Apple Desktop Applications
      • Biscuit: The benefits and costs of writing a POSIX kernel in a high-level language
      • Data Domain: Avoiding the Disk Bottleneck in the Data Domain Deduplication File System
      • Disco: Running Commodity Operating Systems on Scalable Multiprocessors
      • FFS: A Fast File System for UNIX
      • From WiscKey to Bourbon: A Learned Index for Log-Structured Merge Trees
      • LegoOS: A Disseminated, Distributed OS for Hardware Resource Disaggregation
      • LFS: The Design and Implementation of a Log-Structured File System
      • Lottery Scheduling: Flexible Proportional-Share Resource Management
      • Memory Resource Management in VMware ESX Server
      • Monotasks: Architecting for Performance Clarity in Data Analytics Frameworks
      • NFS: Sun's Network File System
      • OptFS: Optimistic Crash Consistency
      • RAID: A Case for Redundant Arrays of Inexpensive Disks
      • RDP: Row-Diagonal Parity for Double Disk Failure Correction
      • Resource Containers: A New Facility for Resource Management in Server Systems
      • ReVirt: Enabling Intrusion Analysis through Virtual-Machine Logging and Replay
      • Scheduler Activations: Effective Kernel Support for the User-Level Management of Parallelism
      • SnapMirror: File-System-Based Asynchronous Mirroring for Disaster Recovery
      • The Linux Scheduler: a Decade of Wasted Cores
      • The Unwritten Contract of Solid State Drives
      • Venti: A New Approach to Archival Storage
    • Earlier Notes
      • How to read a paper
  • FIXME
    • Template for Paper Reading Notes
Powered by GitBook
On this page
  • One-line Summary
  • Paper Structure Outline
  • Background & Motivation
  • Design and Implementation
  • RAID-0: Striping, no redundancy
  • RAID-1: Mirroring
  • RAID-4: Parity
  • RAID-5: Rotating parity
  • Evaluation
  • Links

Was this helpful?

  1. Earlier Readings & Notes
  2. Operating Systems Papers - Index

RAID: A Case for Redundant Arrays of Inexpensive Disks

One-line Summary

Redundant arrays of inexpensive disks are used to address the I/O crisis. A tradeoff analysis is presented.

Paper Structure Outline

  1. Background: Rising CPU and Memory Performance

  2. The Pending I/O Crisis

  3. A Solution: Arrays of Inexpensive Disks

  4. Caveats

  5. And Now The Bad News: Reality

  6. A Better Solution: RAID

  7. First Level RAID: Mirrored Disks

  8. Second Level RAID: Hamming Code for ECC

  9. Third Level RAID: Single Check Disk Per Group

  10. Fourth Level RAID: Independent Reads/Writes

  11. Fifth Level RAID: No Single Check Disk

  12. Discussion

  13. Conclusion

Background & Motivation

The performance in I/O is not catching up with that in CPU and memory. Gene Amdahl (an American computer architect who completed his Ph.D. at UW-Madison!) presented Amdahl's Law, which stated that the overall speedup relies on the worst-performing sections the most:

  • S = the effective speedup

  • f = fraction of work in fast mode

  • k = speedup while in fast mode

The implication of this rule is as follows: Suppose that 10% of some application is spent in I/O. If we have computers 100 times faster (through the evolution of uniprocessors or by multiprocessors), the overall speedup will be less than 10 times faster, wasting ~90% of the potential speedup. This is the I/O crisis people were facing.

Inexpensive disks have lower I/O (but not that low), and in some aspects, they are superior or equal to the larger disks. This inspired people to use an array of inexpensive disks. However, there is very little fault-tolerance: The Mean Time To Failure (MTTF) of an array of disks is:

Therefore, people made use of extra disks containing redundant information for recovery.

Design and Implementation

RAID-0: Striping, no redundancy

RAID-0 is actually not RAID, as there is no redundancy. However, by striping the data, RAID-0 achieves the upper-bound on performance and capacity.

RAID-1: Mirroring

RAID-1 is sometimes called RAID-10 (1+0, stripe of mirrors). RAID-1 is capacity-expensive (only 50% with mirroring level = 2) but very good against failures (can tolerate up to half of the disk failing if we are lucky).

RAID-4: Parity

The parity disk uses XOR to calculate the parity of the data disks and can tolerate the loss of any one block from our stripe. A drawback is the small-write problem: see the evaluation section below for more detail.

RAID-5: Rotating parity

An attempt to increase the performance of random writes. Now the parity block is rotated across drives instead of on one single disk.

Evaluation

There are two types of workloads:

  • Sequential: Requests to the array come in large contiguous chunks. E.g., a request that accesses 1 MB of data that starts at block x and ending at block (x+1 MB). Sequential workloads are very common.

  • Random: Each request is to a different random location on the disk. DBMS-related workloads often exhibit this pattern.

  • N: Number of disks

  • B: Number of blocks per disk

  • S: I/O for one disk under a sequential workload

  • R: I/O for one disk under a random workload

  • T: The time that a request to a single disk would take

Some interesting data points:

  • Sequential read for RAID-1: Unlike random read where we can distribute the reads to all the disks and thus obtain full possible bandwidth, in sequential reads, the bandwidth utilization is only 50%. Consider disk 0 in figure 38.3. When a sequential read is issued (0, 1, 2, 3, 4, ...), disk 0 gets a request for block 0 and then block 4, skipping block 2. While it is rotating over the skipped block, it is not delivering useful bandwidth to the client.

  • Sequential write for RAID-4: A simple optimization, full-stripe write, is used. We first calculate the value of the new parity and then write all of the blocks to all disks above in parallel.

  • Random write for RAID-4: Both the data disk and the parity disk needs to be written. There are two methods: the additive/subtractive parity method. In subtractive, for each write, RAID has to perform 4 I/Os (2 reads & 2 writes). When a series of writes comes in, the parity becomes a bottleneck in that all requests have to read the related parity blocks. This is known as the small-write problem.

  • Random read for RAID-5: Slightly better than that for RAID-4 as now we can utilize all disks.

  • Random write for RAID-5: Parallelism across requests. We can assume that given a large number of random requests, we can keep all the disks about evenly busy.

Links

PreviousOptFS: Optimistic Crash ConsistencyNextRDP: Row-Diagonal Parity for Double Disk Failure Correction

Last updated 4 years ago

Was this helpful?

S=1(1−f)+f/kS = {1 \over (1 - f) + f / k}S=(1−f)+f/k1​
MTTFofaDiskArray=MTTFofaSingleDiskNumberofDisksintheArrayMTTF of a Disk Array = {MTTF of a Single Disk \over Number of Disks in the Array}MTTFofaDiskArray=NumberofDisksintheArrayMTTFofaSingleDisk​

Paper PDF
RAID in OSTEP
RAID in CS 537 @ UW-Madison
2MB
L3+L4+L5-RAID+RDP+iBench.pptx
Prof. Andrea's slides on RAID and RDP
Amdahl's Law in the UW-Madison Alumni Park. Photo by Jeff Miller from OnWisconsin (https://onwisconsin.uwalumni.com/features/6-surreptitious-science-lessons-in-alumni-park/).
Chunk size = 1 block. This can be changed to more blocks.
The XOR parity function