Aggregate Traffic Performance with Active Queue Management and Drop from Tail

1
Aggregate Traffic Performance with Active Queue
Management and Drop from Tail

• Christophe Diot, Gianluca Iannaccone, Martin May
• Sprint ATL, Università di Pisa, Activia
• www.sprintlabs.com

2
Active Queue Management
queue
average
instantaneous
drop size
function
sharp
RED
Drop from Tail
smooth
Gentle
Gentle RED
RED
Instantaneous
3
Original RED
1
max-p
min-tresh
max-tresh
buffer size
4
Revised RED
1
max-p
min-tresh
max-tresh
buffer size
5
Gentle RED
1
max-p
min-tresh
max-tresh
buffer size
2 max-tresh
6
Experiments

• Testbed
• with CISCO routers (7500)
• with Dummynet
• We use recommended RED and GRED parameters
• Heterogeneous delays (120 to 180 ms)

7
Traffic characteristics

• 16 to 256 TCP connections sharing the bottleneck.
  
• Experimental traffic generated by Chariot
• long-lived TCP connections.
• more realistic traffic mix
• 90 short lived TCP connections (up to 20
  packets)
• 10 long lived TCP connections
• 1Mbps UDP in both cases

8
Testbed (CISCO routers)
7500
7500
10 Megs
9
Testbed (Dummynet)
7500
7500
10 Megs
Dummy net
100 Megs
10
What is Dummynet?
application
dummynet
network
11
Metrics observed

• Aggregate goodput through a router
• TCP and UDP loss rate
• Consecutive losses
• Queuing behavior

12
Aggregate goodput (long-lived TCP)
13
256 short and long lived TCP connections
14
Consecutive packet losses (long lived)
15
if we use optimal RED parameters
16
Consecutive packet losses (realistic traffic mix)
17
Queuing behavior (256 long lived connections)
18
Queuing behavior (256 connections, realistic mix)
19
In summary ...

• No significant difference on goodput, TCP losses
  and UDP losses.
• On consecutive losses, clear advantage to GRED
  and GRED-I.
• gentle modification solves many RED problems.
• Oscillations no clear winner. Traffic seems to
  be the determining factor.

20
From the ISP standpoint ...

• Not clear there is an advantage in deploying RED,
  GRED, or GRED-I.
• Maybe GRED-I is an option if one can find a
  universal exponential dropping function.
• ECN will work with any scheme.
• Not clear the solution is in the AQM space.

21
GRED-I with exponential dropping function
1
buffer size
22
About Fair Queuing ...

• Not only feasible easy at the edges!
• www.agere.com (an example)
• vendors support from 64k to 200k flows
• Really fair
• everybody gets what he/she paid for
• local signaling (end host to CPE)

23
Number of flows on an OC-3 link