Network and Protocol Mechanisms: How well do they collaborate?
Post on 03-Jan-2016
19 Views
Preview:
DESCRIPTION
Transcript
Network and Protocol Mechanisms:
How well do they collaborate?
Ageliki Tsioliaridou
What we investigate
• Transport Protocols come at various versions - some aggressive some conservative
• Network mechanisms differ in sophistication regarding the scheduling, forwarding and dropping policy
We claim that
• Evaluation of a new mechanism cannot be investigated alone; that is, one has to study its impact on the different mechanisms.
• A protocol may lack the sophistication needed to exploit the potential of a new network mechanism, and vice versa
• the chicken or the egg
More specifically
• We select two widely used network mechanisms (DT and RED); we also introduce a new mechanism, namely Fr-RED; and we discuss the potential of another mechanism which we will develop soon
• We monitor the interaction of these mechanisms with the congestion control mechanisms of Tahoe, Reno, NewReno and Vegas
1st scenarioMany flows compete for low bandwidth. The contention level is high. Congestion event is persistent2nd scenarioA small number of flows occupy the transmission channel. The contention level is low. Congestion event is transient3rd scenarioSome flows co-exist in the communication channel and suddenly some other flows enter the link 4th scenario Some flows co-exist in the communication channel and suddenly some of them finish their task and leave the channel
Experiments
Topology: dumbbell
1st scenario
Topology:dumbbellbw_1=0.1Mbps, bw_2=1Mbps, bw_3=0.1Mbps
Remarks that have to be highlighted:1. If the protocol is Vegas or Tahoe, the combination with
drop give us better results in throughput2. If the protocol is Reno or Newreno, the combination
with red give us better results not only in goodput but also in throughput
3. When the router’s algorithm is drop, the performance of Vegas in goodput is higher than the other three
TCP Tahoe
goodput throughput
tahoe
020000400006000080000
100000120000140000160000180000
50 60 70 80 90 100 110 120
number_of_flows
(bps
)
DropTail
Red
tahoe
100000
110000
120000
130000
140000
150000
160000
170000
180000
50 60 70 80 90 100 110 120
number_of_flows(b
ps)
DropTail
Red
TCP Vegas
goodput throughput
vegas
020000400006000080000
100000120000140000160000180000
50 60 70 80 90 100 110 120
number_of_flows(b
ps)
DropTail
Red
vegas
020000400006000080000
100000120000140000160000180000
50 60 70 80 90 100 110 120
number_of_flows
(bps
)
DropTail
Red
TCP Reno
goodput throughput
reno
020000400006000080000
100000120000140000160000180000
50 60 70 80 90 100 110 120
number_of_flows
(bps
)
DropTail
Red
reno
0
50000
100000
150000
number_of_flows(b
ps)
DropTail
Red
TCP NewReno
goodput throughput
nr
020000400006000080000
100000120000140000160000180000
50 60 70 80 90 100 110 120
number_of_flows
(bps
)
DropTail
Red
nr
020000400006000080000
100000120000140000160000180000
50 60 70 80 90 100 110 120
number_of_flows
(bps
)
DropTail
Red
goodput throughput
DropTail
020000400006000080000
100000120000140000160000180000
50 60 70 80 90 100 110 120
number_of_flows
(bps
) tahoe
reno
nr
vegas
Droptail
020000400006000080000
100000120000140000160000180000
50 60 70 80 90 100 110 120
number_of_flows
(bps
) tahoe
reno
nr
vegas
DropTail
2nd scenario
Topology:dumbbell
bw_1=1Mbps, bw_2=50Mbps, bw_3=10Mbps
Remarks that have to be highlighted:
1. If the protocol is Tahoe the combination with drop give us better results
2. If the transport protocol is Vegas the value of goodput doesn’t get influence from the router’s algorithm, and it is higher than the value of other three protocols.
TCP Tahoe
goodput throughput
tahoe
0
1000000
2000000
3000000
4000000
5000000
6000000
7000000
10 20 30 40 50 60
number_of_flows(b
ps)
droptail
red
tahoe
0
1000000
2000000
3000000
4000000
5000000
6000000
7000000
10 20 30 40 50 60
number_of_flows
(bps
)
droptail
red
TCP Vegas
vegas
0
1000000
2000000
3000000
4000000
5000000
6000000
7000000
10 20 30 40 50 60
number_of_flows
(bps
)
droptail
red
vegas
1000000
2000000
3000000
4000000
5000000
6000000
7000000
10 20 30 40 50 60
number_of_flows(b
ps)
droptail
red
goodput throughput
goodput throughput
red
0
2000000
4000000
6000000
10 20 30 40 50 60number_of_flows
(bps
) tahoe
reno
nr
vegas
red
1000000
2000000
3000000
4000000
5000000
6000000
7000000
10 20 30 40 50 60
number_of_flows(b
ps)
tahoe
reno
nr
vegas
RED
3nd scenario
Topology:dumbbell
bw_1=1Mbps, bw_2=50Mbps, bw_3=10Mbps
Remarks that have to be highlighted:
• The combination of Vegas and Drop Tail algorithm gives us the worst value in fairness
• When of the transport protocol is Tahoe, the Drop Tail algorithm performs better in goodput
• If the transport protocol is Vegas it gives the best value in goodput
Fairness
droptail
0,600000,650000,700000,750000,800000,850000,900000,950001,00000
10-100
20-100
30-100
40-100
50-100
60-100
tahoe
reno
nr
vegas
TCP Tahoe
tahoe
3000000
35000004000000
45000005000000
5500000
60000006500000
7000000
10-100
20-100
30-100
40-100
50-100
60-100
number_of_flows
(bps
)
droptail
red
goodput
goodput throughput
goodput throughput
TCP Vegas
red
3000000
3500000
4000000
4500000
5000000
5500000
6000000
6500000
7000000
10-100 20-100 30-100 40-100 50-100 60-100
number_of_flows
(bps)
tahoe
reno
nr
vegas
red
3000000
3500000
4000000
4500000
5000000
5500000
6000000
6500000
7000000
10-100 20-100 30-100 40-100 50-100 60-100
number_of_flows
(bp
s)
tahoe
reno
nr
vegas
droptail
3000000
3500000
4000000
4500000
5000000
5500000
6000000
6500000
7000000
10-100 20-100 30-100 40-100 50-100 60-100
number_of_flows
(bps)
tahoe
reno
nr
vegas
droptail
3000000
3500000
4000000
4500000
5000000
5500000
6000000
6500000
7000000
10-100 20-100 30-100 40-100 50-100 60-100
number_of_flows
(bps)
tahoe
reno
nr
vegas
4th scenario
Topology:dumbbellbw_1=1Mbps, bw_2=50Mbps, bw_3=10Mbps
Remarks that have to be highlighted:• When the transport protocol is Reno, the Drop
algorithm results better in fairness• When of the transport protocol is Tahoe, the drop
algorithm perfumes better in goodput• If the transport protocol is Vegas the value of goodput
is higher than the value of other three protocols.
TCP Reno
fairness
reno
0,0000
0,2000
0,4000
0,6000
0,8000
1,0000
100-50 100-40 100-30 100-20 100-10
number_of_flows
droptail
red
tahoe
200000030000004000000500000060000007000000
100-50
100-40
100-30
100-20
100-10
droptail
red
TCP Tahoe
goodput
goodput throughput
goodput throughput
droptail
20000002500000300000035000004000000450000050000005500000600000065000007000000
100-50 100-40 100-30 100-20 100-10
number_of_flows
tahoe
reno
nr
vegas
red
2000000
3000000
4000000
5000000
6000000
7000000
100-50 100-40 100-30 100-20 100-10
number_of_flows
tahoe
reno
nr
vegas
droptail
20000002500000300000035000004000000450000050000005500000600000065000007000000
100-50 100-40 100-30 100-20 100-10
number_of_flows
tahoe
reno
nr
vegas
red
2000000
3000000
4000000
5000000
6000000
7000000
100-50 100-40 100-30 100-20 100-10
number_of_flows
tahoe
reno
nr
vegas
Random Early Drop (RED)
A router that implements RED uses two threshold values to mark positions in the queue: Tmin and Tmax
AvgLen
TminTmax2* Tmax
A drop event is characterize either as FORCED drop nor as UNFORCED drop
DROP LOGIC
1.If avg > 2* maxthresh , this is a FORCED drop
2.If Tmin < avg < 2*maxthresh, this may be an UNFORCED drop. The drop probability changes from 0 to max_p as the avg varies from Tmin to Tmax and from max_p to 1 as the avg varies from Tmax to twice Tmax (max_p=1/linterm)
3.If (q+1) > hard q limit, this is a FORCED drop
RED drop principle:
FORCED dropThe victim is either the arriving packet (default) or the front
packet of the queue or any packet of packet (random)UNFORCED dropThe victim is the arriving packet
fr-RED drop principle :
FORCED dropThe victim is the arriving packetUNFORCED dropThe victim is the front packet of the queue
Our goal is:
To indicate senders faster that the congestion is going to happen The TCP congestion mechanism of senders will be triggered faster
Experiments
Topology: dumbbell
scenario
bw_1=bw_2=bw_3=1Mbps
TCP Tahoe
fairness
0,5000
0,6000
0,7000
0,8000
0,9000
1,0000
3 4 5 6 7 8 9 10
number_of_flows
tahoe
tahoefr
throughput
100000
110000
120000
130000
140000
3 4 5 6 7 8 9 10
number_of_flows
(bps
) tahoe
tahoefr
goodput
100000
110000
120000
130000
140000
3 4 5 6 7 8 9 10
number_of_flows
(bps
) tahoe
tahoefr
Buffer_size (bs):100Link_delay:7msTmax:3*bs/4=75Tmin:Tmax/3=25
TCP Tahoe
fairness
0,5000
0,6000
0,7000
0,8000
0,9000
1,0000
3 4 5 6 7 8 9 10
number_of_flows
tahoe
tahoefr
fairness
0,5000
0,6000
0,7000
0,8000
0,9000
1,0000
3 4 5 6 7 8 9 10
number_of_flows
tahoe
tahoefr
goodput
100000
105000
110000
115000
120000
125000
130000
135000
140000
3 4 5 6 7 8 9 10
tahoe
tahoefr
throughput
100000
105000
110000
115000
120000
125000
130000
135000
140000
3 4 5 6 7 8 9 10
number_of_flows
(bp
s)
tahoe
tahoefr
Buffer_size (bs):100 Buffer_size (bs):200
goodput
100000105000110000115000120000125000130000135000140000
3 4 5 6 7 8 9 10
number_of_flows
(bp
s)
tahoe
tahoefr
throughput
100000105000110000115000120000125000130000135000
3 4 5 6 7 8 9 10
number_of_flows
(bp
s)
tahoe
tahoefr
The concept of a new network mechanism:
When congestion is going to happen rearrange the order of the packets at the queue
The concept of a new congestion control mechanism:
The sender should adjust its rate, depending on the reordering of the incoming packets at the receiver.
Future work
• Evaluation of fr-red at high-speed networks• Implementation of the new concept of
congestion avoidance mechanism
top related