Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
Voldemort on Solid
State Drives
Vinoth Chandar, Lei Gao, Cuong Tran
Linkedin Corp, Mountain View, CA
What is Voldemort?
• Distributed key-value storage system
  • Based on Amazon Dynamo
• Open source
  • https://github.com/...
SSD Migration
• Improve average latency
  • 20 ms to 2 ms
• Cluster expansion & data restoration
  • Improved 10x
• 95th &...
Benchmarking for SSD
• Java based systems
• Need for End-End Correlation
  • Linux page stealing
  • Use AlwaysPreTouch
• ...
Takeaways
• GC impact significant
  • Big heaps = Big pauses
• Longer runs to account for SSD GC
  • SSD is log structured...
Upcoming SlideShare
Loading in …5
×

of

Voldemort on Solid State Drives Slide 1 Voldemort on Solid State Drives Slide 2 Voldemort on Solid State Drives Slide 3 Voldemort on Solid State Drives Slide 4 Voldemort on Solid State Drives Slide 5
Upcoming SlideShare
Composing and Executing Parallel Data Flow Graphs wth Shell Pipes
Next
Download to read offline and view in fullscreen.

4 Likes

Share

Download to read offline

Voldemort on Solid State Drives

Download to read offline

Voldemort on Solid State Drives

  1. 1. Voldemort on Solid State Drives Vinoth Chandar, Lei Gao, Cuong Tran Linkedin Corp, Mountain View, CA
  2. 2. What is Voldemort? • Distributed key-value storage system • Based on Amazon Dynamo • Open source • https://github.com/voldemort/voldemort • Widely used • Several TB of data @ Linkedin • Eharmony, Nokia, JIVE Software etc. • We are on SSD!
  3. 3. SSD Migration • Improve average latency • 20 ms to 2 ms • Cluster expansion & data restoration • Improved 10x • 95th & 99th latencies shot up • 30ms to 130ms and 240ms to 380ms • GC Issues !! Client operations BDB Cleaner Retention Job BDB Cache App1 App2 .. AppN Server JVM Heap Linux
  4. 4. Benchmarking for SSD • Java based systems • Need for End-End Correlation • Linux page stealing • Use AlwaysPreTouch • Run Workload causing Fragmentation • Promotion failure more likely • Defragmentation costly • Run Workload with High memory churn • cost of CMS Initial Mark • High CMS initiating fraction • Factor in higher IO Parallelism • Sequential IO benchmarks no longer use single thread
  5. 5. Takeaways • GC impact significant • Big heaps = Big pauses • Longer runs to account for SSD GC • SSD is log structured • Worst case 2 ms • Special attention to 95th & 99th • Fragmentation eats up gains
  • YoungjinYu1

    Feb. 5, 2013
  • flyinweb

    Jan. 5, 2013
  • parkghost

    May. 17, 2012
  • tanbamboo

    May. 16, 2012

Views

Total views

2,070

On Slideshare

0

From embeds

0

Number of embeds

116

Actions

Downloads

22

Shares

0

Comments

0

Likes

4

×