Changes between Version 24 and Version 25 of Performance


Ignore:
Timestamp:
2007-12-19T04:13:07Z (16 years ago)
Author:
warner
Comment:

add mostly-download load test data

Legend:

Unmodified
Added
Removed
Modified
  • Performance

    v24 v25  
    178178factor than per-byte overhead.
    179179
     180=== test three: 80% upload, 20% download, 1MB mean file size ===
     181
     182Same environment as test 2, but 80% of the operations are uploads.
     183
     184{{{
     185clients: 150kBps down, 680kBps up, .14 fps down, .67 fps up
     186tahoecs1: 62% CPU, 11Mbps out, 2.9Mbps in, load avg .85
     187tahoecs2: 57% CPU, 10Mbps out, 4Mbps in, load avg .76
     188tahoebs4: 16% CPU, 700kBps out, 5.4Mbps in, load avg 0.4ish
     189tahoebs5: 21%, 870kBps out, 5.1Mbps in, load avg about 0.35
     190}}}
     191
     192Overall throughput is about half of the download case. Either uploading files
     193or modifying the dirnodes looks to be more expensive than downloading. The
     194CPU usage on the web servers was lower, suggesting that the expense might be
     195in round trips rather than actual computation.
     196
    180197=== initial conclusions ===
    181198