20 October 2011, HWB

On Wednesday, 18 October 2011, a new Tier-based QoS paradigm was implemented on HPWREN backbone nodes after the following tests were done on a separate test network.

Tier

DSCP

Hex

Class

E-queue

Threshold

E-Alloc 1

E-Alloc 2

Testport

3

CS1

0x20

default

2

1

8

5

Undef.

2

CS2

0x40

116

3

1

22

20

5003

1-B

CS4

0x80

132

4

1

65

65

5002

1-E

CS5

0xa0

140

1

1

5

10

5001


3560 testing with 95Mbps iperf UDP loads injected from a 1Gbps interface onto a 100Mbps substrate:

Egress buffer allocation 1

Egress buffer allocation 2

Port

Single

All

partial

partial

partial

partial

Single

All

partial

partial

partial

partial

5004

28

7.5

-

9.6

26

7.9

25

4.7

-

5.9

21

5.2

5003

45

21

23

-

38

22

40

19

20

-

37

21

5002

96

63

68

80

-

66

96

63

66

78

-

70

5001

29

4.7

5

6

21

-

25

9.4

9.9

12

26

-


3550 testing with 20Mbps iperf UDP loads injected from a 100Mbps interface onto a 10Mbps substrate (“Egress buffer allocation 1 - Single” was tested at 95Mbps onto a 100Mbps substrate):

Egress buffer allocation 1

Egress buffer allocation 2

Port

Single

All

partial

partial

partial

partial

Single

All

partial

partial

partial

partial

5004

95

0.75

-

1

2.2

0.8

9.6

0.47

-

1.2

1.4

0.5

5003

95

2.1

2.3

-

6

2.2

9.6

1.9

2

-

5.5

2.1

5002

95

6.2

6.8

8

-

6.5

9.6

6.3

6.6

7.8

-

6.9

5001

95

0.48

0.52

0.62

1.4

-

9.6

0.94

1

1.2

2.7

-


3550 CoS maps for HPWREN-internal connections:


E-queue

CoS value

2

0, 1

3

2, 3

4

4

1

5, 6, 7


Subsequent traffic was verified to contain the correct DSCP values per the access-list definitions.