Presented by aditi bose hyma chilukuri
Download
1 / 15

FAWN: A Fast Array of Wimpy Nodes - PowerPoint PPT Presentation


  • 199 Views
  • Uploaded on

Presented by: Aditi Bose & Hyma Chilukuri. FAWN: A Fast Array of Wimpy Nodes. Motivation.

loader
I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
capcha
Download Presentation

PowerPoint Slideshow about 'FAWN: A Fast Array of Wimpy Nodes' - dakota


An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript
Presented by aditi bose hyma chilukuri

Presented by:

Aditi Bose & Hyma Chilukuri

FAWN: A Fast Array of Wimpy Nodes


Motivation
Motivation

Large-scale data-intensive applications like high performance key-value storage systems are being used by Facebook, LinkedIn, Amazon with more regularity. Being I/O, Requiring RA over large DB, performing parallel, concurrent and mostly independent operations, requiring large clusters and storing small sized objects are several common features these workloads share.System performance: queries/sec    Energy efficiency: queries/joule CPU performance and I/O bandwidth Gap : For data intensive computing workloads, storage, network and memory bandwidth bottlenecks lead to low CPU utilizationSolution: wimpy processors to reduce I/O induced idle cyclesCPU Power consumption: operating processors at higher freq requires more energy.                     techniques to mask CPU bottleneck cause energy inefficiency                     branch prediction, speculative execution – more processor  die areaSolution:  slower CPUs execute more instructions per joule               1 billion vs. 100 million instructions per Joule


FAWN

Efficient – 1W at heavy load Vs 10W at load        Fast random reads – up to 175 times faster        Slow random writes – updating a single page means erasing an entire block before writing the modified block in its placeCluster of embedded CPUs using flash storage        Efficient – 1W at heavy load Vs 10W at load        Fast random reads – up to 175 times faster        Slow random writes – updating a single page

means erasing an entire block before writing the modified

block in its place FAWN-KeyValue        nodes organized into a ring using consistent Hashing        physical node is a collection of virtual nodeFAWN-DS        Log structured key-value stores        contains values for key range associated with VID 


Fawn ds
FAWN - DS

Uses as in-memory Hash Index to map 160-bit key to a value stored in the data logstores only a fragment of the actual key.         Hash Index bucket = i low order index bits        key fragment = next 15 low order bitsEach bucket -6 bytes - stores frag, valid bit and 4-byte pointer


Fawn ds1

Virtual Node Maintenance:

    Split

    Merge

    Compact

FAWN - DS

Basic Functions:

        Store

        Lookup

        Delete

                               Concurrent operations


Fawn kv
FAWN - KV

FAWN-KV organizes the back-end VIDs into a storage ring-structure using consistent hashingManagement node        assigns each front-end to circular key space Front-end node        manages fraction of key-space        manages the VID membership list        forwards out-of-range request Back-end nodes – VIDs        owns a key range        contacts front-end when joining


Fawn kv1
FAWN - KV

Chain replication


Fawn kv2
FAWN - KV

Join

    split key range

    pre-copy

    chain insertion

    log flush

Leave

    merge key range

    Join into each chain


Individual node performance
Individual Node Performance

  • Lookup speed

  • Bulk store speed: 23.2 MB/s, or 96% of raw speed


Individual node performance1
Individual Node Performance

  • Put speed

  • Compared to BerkeleyDB: 0.07 MB/s – shows necessity of log-based filesystems


Individual node performance2
Individual Node Performance

  • Read- and write-intensive workloads


System benchmarks
System Benchmarks

  • System throughput and power consumption


Impact of ring membership changes
Impact of Ring Membership Changes

  • Query throughput during node join and maintenance operations


Alternative architectures
Alternative Architectures

Large Dataset, Low Query → FAWN+Disk                 number of nodes dominated by storage capacity per node                 has the lowest total cost per GBSmall Dataset, High Query → FAWN+DRAM                number of nodes dominated by per node query capacity                has the lowest cost for queries/secMiddle Range → FAWN+SSD               best balance of storage capacity, query rate and total cost


Conclusion
Conclusion

  • Fast and energy efficient processing of random read-intensive workloads

  • Over an order of magnitude more queries per Joule than traditional disk-based systems


ad