Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

  •  Michael Birch (Unlicensed): gist from(deprecated) Measurement of RChain Network Performance needs to be revised.
    •  Once updated, need the output integrated into the performance harness.
  •  Stable performance - Dupe(5) contract - updates to the scheduler (CORE-1033) will address.
  •  What is the hardware configuration we need to recommend?  Do we need people focusing on this?
    •  We know we need at least 4 GB of memory
    •  How many CPU's?
    •  Approach - What does it take for a node to hit 1K COMM / second? We send a dupe(3) out 1X /second.  Have versions of dupe(3) that write on different channels.  - The above is in progress by Medha Parlikar (Unlicensed)
      •  Create a version of dupe that makes a new name. 
        Jira Legacy
        serverSystem JIRA
        serverId50130123-f232-3df4-bccb-c16e7d83cd3e
        keyRHOL-645
      •  How are validators deciding when to propose? Gatling is also making a propose. 
      •  We will need to determine what the interval of propose will be & block size (as a result).  We can tune the number of proposes per deploy.

...

  •  Mike to provide an updated wide loop contract.
  •  Will run dupe against stress-docker
    •  Have a network ready to do this
  •  Will run wide against stress-docker
    •  Have a network ready to do this, with the contract set up.
  •  Will run token transfer contract against test net
  •  Will speak to our roadmap for performance
    •  No silver bullet for performance.  It's a lot of little fixes.
    •  Created a performance harness, will become part of our test criteria. 
    •  Re-engineer appends in RSpace
    •  Optimize how we store blocks in Casper, presently there is a good amount of duplicated data
    •  Optimize how we pass blocks around then network
    •  Optimize our utilization of LMDB
    •  -   Issue is reading something out of the DB, checking for a match, and then putting it back.  
      •  to append, requires that a blob needs to be deserialized, checked for a match and then appended to (no match for a produce)
    •  Customize write locks for channels that are disjoint.
    •  Examine the performance of the matcher.
    •  Examine the performance of block storage.  
    •  Examine the performance of cryptographic functions.
    •  Optimize how we store blocks in Casper, presently there is a good amount of duplicated data
    •  Optimize replay (proposing node also replays the block, which is not necessary)
    •  Optimize how we pass blocks around then network
    •  Optimize read /write locks and threading in transaction handling code.
    •  Implement peek in Rholang + RSpace 
  •  Will speak to how long we have been performance testing
    •  When we are feature complete
    •  What we have built.
  •  Question about the token transfer contract code - per Michael Stay (Unlicensed) the code appears correct, not sure what the 0, 11 represent.
  •  Run the token transfer contract and see how the system performs.
  •  Run with Kent's PR to observe the Neglected invalid block + bonds cache issue

...