wiki:GENIRacksHome/Performance

Version 10 (modified by lnevers@bbn.com, 5 years ago) (diff)

--

0.1 GENI Racks Performance

This page captures performance measurements for GENI Racks using OpenFlow and Stitching Scenarios. The following topologies will be used with raw-pc/bare metal dedicated hosts to collect performance data:

  • Site Confirmations test OpenFlow connections over Meso-scale:
    • New IG or EG Site to GPO IG, GPO EG, Utah PG
  • Site Confirmations test for OpenFlow connection over AL2S:
    • New IG or EG Site to GPO IG, GPO EG, Utah PG
  • Site Stitching Confirmation test via stitched VLAN:
    • New IG or EG Site to GPO IG
    • New IG or EG Site to GPO EG Bare Metal

For details about the performance test criteria see the Test Methodology section.

1.0 Performance Results

1.1 OpenFlow connections over Meso-scale Connection

End points One Client(tcp)5 Clients(tcp)10 Clients(tcp)1 client(udp)Ping rtt min/avg/max/mdev
Clemson IG to GPO IG
Clemson IG to GPO EG
Clemson IG to PG Utah
Wisconsin IG to GPO IG
Wisconsin IG to GPO EG
Wisconsin IG to PG Utah
GATech IG to GPO IG
GATech IG to GPO EG
GATech IG to PG Utah
Northwestern IG to GPO IG
Northwestern IG to GPO EG
Northwestern IG to PG Utah
Utah IG to GPO IG
Utah IG to GPO EG
Utah IG to PG Utah
Stanford IG to GPO IG
Stanford IG to GPO EG
Stanford IG to PG Utah
NYU IG to GPO IG
NYU IG to GPO EG
NYU IG to PG Utah
FIU EG to GPO IG
FIU EG to GPO EG
FIU EG to PG Utah

1.2 OpenFlow connection over AL2S Connection

End points One Client(tcp)5 Clients(tcp)10 Clients(tcp)1 client(udp)Ping rtt min/avg/max/mdev
Clemson IG to GPO IG
Clemson IG to GPO EG
Clemson IG to PG Utah
Wisconsin IG to GPO IG
Wisconsin IG to GPO EG
Wisconsin IG to PG Utah
GATech IG to GPO IG
GATech IG to GPO EG
GATech IG to PG Utah
Northwestern IG to GPO IG
Northwestern IG to GPO EG
Northwestern IG to PG Utah
Utah IG to GPO IG
Utah IG to GPO EG
Utah IG to PG Utah
Stanford IG to GPO IG
Stanford IG to GPO EG
Stanford IG to PG Utah
NYU IG to GPO IG
NYU IG to GPO EG
NYU IG to PG Utah
FIU EG to GPO IG
FIU EG to GPO EG
FIU EG to PG Utah

1.3 Stitched connection

End points One Client(tcp)5 Clients(tcp)10 Clients(tcp)1 client(udp)Ping rtt min/avg/max/mdev
Wisconsin IG to GPO IG
Wisconsin IG to GPO EG
MAX IG to GPO IG
MAX IG to GPO EG
NYSERnet IG to GPO IG
NYSERnet IG to GPO EG
Illinois IG to GPO IG
Illinois IG to GPO EG
Kentucky IG to GPO IG
Kentucky IG to GPO EG
Missouri IG to GPO IG
Missouri IG to GPO EG
Utah IG to GPO IG
Utah IG to GPO EG

2.0 Test Methodology

Test Characteristics:

  • All test hosts are dedicate hosts (bare metal/raw-pc)
  • All results are captured for default bandwidth allocation (100Mb/s shaped), unless otherwise stated.
  • All results collected between 2014-01-xxx and xxx???

Iperf TCP measurements captured in this page have the following assumptions:

  • All results are for a 60 second test run.
  • for OpenFlow scenarios, traffic is exchanged to establish data flows before measurement test runs are executed.
  • iperf '1 client' scenario command: 'iperf -c dest_host -t 60'
  • iperf '5 clients' scenario command: 'iperf -c dest_host -t 60 -P 5'
  • iperf '10 clients' scenario command: 'iperf -c dest_host -t 60 -P 10'

Iperf UDP measurements captured in this page have the following assumptions:

  • Each test run is 60 seconds.
  • UDP measurements between racks requested a bandwidth of 100 Mbits/sec, if results showed that 100Mb/s was possible, then increased to 1Gb/s.
  • UDP measurements within a rack requested a bandwidth of 1Gbits/sec for test run

3.0 Historical Performance Data

Various effort have captured performance for GENI Racks, following is a list of earlier pages that have capture some aspect of performance with OpenVZ, Xen and dedicated hosts. These pages should be considered obsolete:

Attachments (1)

Download all attachments as: .zip