Difference between revisions of "CSIT/VPP-ubuntu-compare-1404to1604"

From fd.io
Jump to: navigation, search
Line 4: Line 4:
 
** 1. Ubuntu 14.04 - vpp build 101, csit-vpp-perf-master-long #1076
 
** 1. Ubuntu 14.04 - vpp build 101, csit-vpp-perf-master-long #1076
 
** 2. Ubuntu 16.04 - vpp build 136, csit-vpp-perf-master-long #1080(onwards)
 
** 2. Ubuntu 16.04 - vpp build 136, csit-vpp-perf-master-long #1080(onwards)
 +
** 3. Source data log files for above calcs available here: [https://wiki.fd.io/view/CSIT/long1076vs1080-ubuntu1404vs1604-analysis analysis], [https://wiki.fd.io/view/CSIT/long1076vs1080-ubuntu1404vs1604-thput-incr thput-increase], [https://wiki.fd.io/view/CSIT/long1076vs1080-ubuntu1404vs1604-thput-decr thput-decrease].
  
 
* '''B. Increased NDR throughput - all values for 1worker, unless noted:'''
 
* '''B. Increased NDR throughput - all values for 1worker, unless noted:'''
Line 14: Line 15:
 
* '''C. Decreased NDR throughput - all values for 1worker:'''
 
* '''C. Decreased NDR throughput - all values for 1worker:'''
 
** 1. IPv6 x520 -5%
 
** 1. IPv6 x520 -5%
 
Source data log files for above calcs available here: csit wiki links
 
  
 
* '''D. This is also reflected in NDR trending graphs in LF Jenkins:'''
 
* '''D. This is also reflected in NDR trending graphs in LF Jenkins:'''

Revision as of 14:27, 1 November 2016

Summary of performance trending from FD.io testbeds after upgrading from Ubuntu14.04 Trusty to 16.04 Xenial:-

  • A. Comparison based on:
    • 1. Ubuntu 14.04 - vpp build 101, csit-vpp-perf-master-long #1076
    • 2. Ubuntu 16.04 - vpp build 136, csit-vpp-perf-master-long #1080(onwards)
    • 3. Source data log files for above calcs available here: analysis, thput-increase, thput-decrease.
  • B. Increased NDR throughput - all values for 1worker, unless noted:
    • 1. l2bd x520 6%
    • 2. IPv4 x520 3%
    • 3. L2XC x520 3..9% (untagged, dot1q, dot1ad)
    • 4. L2XC xl710 3%
    • 5. L2XC xl710 81% 2workers
  • C. Decreased NDR throughput - all values for 1worker:
    • 1. IPv6 x520 -5%

but due to LF purged job history we only have vpp-1609 ("fib1.0") vs. vpp-master (fib2.0) trend graphs.

  • F. And few requests:
    • 1. Welcome all comments!
    • 2. Welcome any volunteers to help us catch any anomalies in perf trending.
    • 3. Pls send any wishes for what you would like to see for performance tests and trending, to csit-dev@lists.fd.io.