Professional Documents
Culture Documents
i+
tive bound 1
i Ki of WFQ, especially when the hop count Ki is large. We
call our scheme C OORDINATED -E ARLIEST-D EADLINE -F IRST (CEDF) since
it uses an earliest-deadline-first approach in which simple coordination is
applied to the deadlines for consecutive hops of a session. The key to the
bound is that once a packet has passed through its first server, it can pass
through all its subsequent servers quickly.
We conduct simulations to compare the delays actually produced by the
two scheduling disciplines. In many cases, these actual delays are comparable to their analytical worst-case bounds, implying that CEDF outperforms
WFQ.
I. I NTRODUCTION
The provision of end-to-end delay guarantees in high-speed
networks remains one of the most important and widely studied Quality-of-Service (QoS) issues. Many real time audio and
video applications rely on the ability of the network to provide small delays. One key mechanism for achieving this aim
is scheduling at the outputs of the switches. In this paper, we
attempt to minimize end-to-end delay using a novel scheduling
scheme.
Before we introduce our scheme we first recall the delay
bounds for the much studied Weighted Fair Queueing (WFQ)
scheduling discipline, also known as Packet-by-Packet Generalized Processor-Sharing (PGPS). In their seminal papers [1], [2],
Parekh and Gallager showed that WFQ achieves the following
session-i delay bound for Rate Proportional Processor Sharing
(RPPS).1
Ki L
i + (Ki ; 1)Li + X
max
m
i
m=1 r
i + Ki ; 1 + K :
i
i
Hence, for a small burst size, e.g. i = 1, the multiple-hop delay
is essentially
i Ki
and the single-hop delay is essentially 1i . Moreover, it is possible to construct an example in which this bound is achieved
since a packet can wait for time 1i at every hop. This illustrates
our earlier observation.
In this paper, we demonstrate with both analysis and simulation that even for small burst sizes, a bound of 1i Ki is not
necessary, i.e. the K -hop delay does not have to be K times the
1-hop delay. Indeed, in the case of uniform packet sizes, uniform service rates and small burst sizes, [3] showed that each
session i can achieve a delay bound, 2
1
O + Ki
i
1
n
O + Ki log
:
i
min
(1)
O 1
i
+ Ki
150
150
rate=0.03
rate=0.05
rate=0.1
rate=0.15
rate=0.2
rate=0.3
rate=0.4
rate=0.5
rate=0.6
rate=0.7
140
130
120
110
130
120
110
100
90
90
80
80
Delay
Delay
100
rate=0.03
rate=0.05
rate=0.1
rate=0.15
rate=0.2
rate=0.3
rate=0.4
rate=0.5
rate=0.6
rate=0.7
140
70
70
60
60
50
50
40
40
30
30
20
20
10
10
0
0
10
15
20
25
30
Session Length
35
40
45
50
Ki L
i + 4Li =" + X
max log():
m
i
r
m=1
(2)
The parameter " is the server utilization factor defined later. The
logarithmic term, although small, is somewhat involved. We
give the full definition later. In Section IV we provide simulation
results to compare the actual performance of our protocol and
WFQ.
The basic ideas of our protocol are an earliest-deadline-first
approach coupled with randomization and coordination. We assign a deadline for every server through which a packet passes.
By introducing some randomness, the deadlines can be sufficiently spread out so that all the packets can meet all their
deadlines. By introducing simple coordination among the deadlines, we can ensure that once a packet has passed through
its first server, it can pass through all its subsequent servers
quickly. We refer to our protocol as C OORDINATED -E ARLIESTD EADLINE -F IRST (CEDF).
The traffic lights in Manhattan provide an intuitive analogue
to CEDF. Since the lights are coordinated, when one traffic light
turns green, many lights further down the street turn green also.
This means that once a car waits through one red light it can
then drive through many green lights quickly. In this way, delay
does not have to accumulate at every light.
From now on, we refer to a delay bound of the form 1i Ki
as a multiplicative bound and a bound of the form 1i + Ki as
an additive bound. In Figures 1 and 2, we plot these bounds for
different values of K i and i . The curves for the multiplicative
bound have different slopes for different i , whereas the curves
for the additive bound all have the same slope. We can see that
in general it is desirable to have an additive bound. We note that
the bound (2) of CEDF is close to an additive bound. (It does not
contain a term K i =i .) Apart from the bound in reference [3] we
know of no previous end-to-end delay bound that is close to an
additive bound.
In our simulations, we observe that the actual delays under WFQ and CEDF are often comparable to their analytical
10
15
20
25
30
Session Length
35
40
45
50
+ Ki .
Ai (t1 t2 ] i + i (t2 ; t1 )
8 t2 t1 0:
i2I (m)
i (1 ; ")rm
m
P i m rm
j 2B j
m
where Bm is the set of backlogged sessions at server m and the
m
i are a set of allocated weights.
Ti = 2dlog2 "i e
M = max
T
i i
Si = Ti i (1 + "=2)
We define Gm for each server m, which determines how the
deadline for a packet is incremented when it advances from one
server to the next. Let,
nMrm "
Gm = Lrmax
log
e Lmin
m
where Lmax = maxi Li and Lmin = mini Li . The parameter
= O(";3 log 1;p1suc ), where psuc is the success probability
of the protocol. (We discuss this success probability in the Remarks section.) Note that is independent of L i , i , i , Ki and
rm .
D1 = + Gm1
Dj = Dj;1 + Gmj
Now that all deadlines are defined, each server gives priority
to the packet that has the earliest deadline.
Remarks
1. The only coordination required comes from the above iterative definition of the deadlines. This coordination can be
achieved simply by stamping each packet with its current deadline.3 Each server can then update the deadlines of its pending
3 This can be done using techniques similar to the protocols of [26].
Ki L
i + 4Li =" + X
max log nMrmk " :
e
mk
i
Lmin
k=1 r
i + 2(Ki ; 1)Li
i
i + 4Li =" :
i
We note that the bound for CEDF does not contain K i .
IV. S IMULATION R ESULTS
Our experiments simulate a simple situation with uniform
packet sizes and uniform server rates. Since CEDF involves
WFQ
Simple-CEDF
150
150
rate=0.03
rate=0.05
rate=0.1
rate=0.15
rate=0.2
rate=0.3
rate=0.4
rate=0.5
rate=0.6
rate=0.7
140
130
120
110
130
120
110
100
90
Mean Delay
Mean Delay
100
rate=0.03
rate=0.05
rate=0.1
rate=0.15
rate=0.2
rate=0.3
rate=0.4
rate=0.5
rate=0.6
rate=0.7
140
80
70
90
80
70
60
60
50
50
40
40
30
30
20
20
10
10
0
0
10
15
20
25
30
Session Length
35
40
45
50
10
15
20
45
50
Simple-CEDF
130
120
110
100
rate=0.03
rate=0.05
rate=0.1
rate=0.15
rate=0.2
rate=0.3
rate=0.4
rate=0.5
rate=0.6
rate=0.7
140
130
120
110
100
90
98th Percentile
98th Percentile
40
150
rate=0.03
rate=0.05
rate=0.1
rate=0.15
rate=0.2
rate=0.3
rate=0.4
rate=0.5
rate=0.6
rate=0.7
140
80
70
60
90
80
70
60
50
50
40
40
30
30
20
20
10
10
0
0
Fig. 5.
10
15
20
25
30
Session Length
35
40
45
50
p:
tinj :
Dk :
many parameters, we simulate a simplified version, S IMPLE CEDF, which nevertheless contains the essence of CEDF. Under S-CEDF, the deadline for the first server is chosen randomly
(without reference to periodic tokens). Every subsequent deadline is the deadline for the previous server incremented by one
packet service time. (See Figure 11.) As we shall see, the performance of S-CEDF corresponds to the analytical bounds of
Section III.
1
2
3
35
WFQ
150
25
30
Session Length
A session-i packet
Injection time of p
Deadline of p at its k th hop
10
15
20
25
30
Session Length
35
40
45
50
Fig. 6.
link speed
1Mb/sec
buffer size
Fig. 12. Session 0 is the long session with 5 hops. Sessions 1 through 5 are the
1-hop sessions.
WFQ
Simple-CEDF
150
150
rate=0.03
rate=0.05
rate=0.1
rate=0.15
rate=0.2
rate=0.3
rate=0.4
rate=0.5
rate=0.6
rate=0.7
140
130
120
110
130
120
110
100
90
Mean Delay
Mean Delay
100
rate=0.03
rate=0.05
rate=0.1
rate=0.15
rate=0.2
rate=0.3
rate=0.4
rate=0.5
rate=0.6
rate=0.7
140
80
70
90
80
70
60
60
50
50
40
40
30
30
20
20
10
10
0
0
10
15
20
25
30
Session Length
35
40
45
50
10
15
20
35
40
45
50
WFQ
Simple-CEDF
150
150
rate=0.03
rate=0.05
rate=0.1
rate=0.15
rate=0.2
rate=0.3
rate=0.4
rate=0.5
rate=0.6
rate=0.7
140
130
120
110
100
rate=0.03
rate=0.05
rate=0.1
rate=0.15
rate=0.2
rate=0.3
rate=0.4
rate=0.5
rate=0.6
rate=0.7
140
130
120
110
100
90
98th Percentile
98th Percentile
25
30
Session Length
80
70
60
90
80
70
60
50
50
40
40
30
30
20
20
10
10
0
0
10
15
20
25
30
Session Length
35
40
45
50
10
15
20
25
30
Session Length
`
s
WFQ
S-CEDF
0.03
0.77
1.0
1.06
0.1
0.7
1.0
1.16
35
40
45
50
0.2
0.6
1.0
1.26
0.3
0.5
1.0
1.3
0.4
0.4
1.4
1.42
0.5
0.3
1.5
1.53
0.6
0.2
1.8
1.79
0.7
0.1
2.2
2.15
WFQ
Simple-CEDF
100
100
rate=0.03
rate=0.05
rate=0.1
rate=0.15
rate=0.03
rate=0.05
rate=0.1
rate=0.15
90
80
80
70
70
60
60
Mean Delay
Mean Delay
90
50
50
40
40
30
30
20
20
10
10
0
0
10
15
20
25
30
Length of network
35
40
45
50
10
15
20
25
30
Length of network
40
45
50
WFQ
Simple-CEDF
100
100
rate=0.03
rate=0.05
rate=0.1
rate=0.15
90
rate=0.03
rate=0.05
rate=0.1
rate=0.15
90
80
80
70
70
60
60
98th Percentile
98th Percentile
35
50
40
50
40
30
30
20
20
10
10
0
0
10
15
20
25
30
Length of Network
35
40
45
50
10
15
20
25
30
Length of Network
35
40
45
50
line of 41 nodes and 80 links, where neighboring nodes are connected by double links. All sessions have 40 hops, starting from
the node on the left end and finishing at the node on the right
end. Each hop along the path of a session can follow either the
upper or the lower link. The choice is made randomly, subject
to the constraint that each link has an equal number of sessions
passing through it. All sessions have the same injection rate.
We vary the number of sessions in order to achieve the desired
session rate. Figures 17 and 19 illustrate the end-to-end delays
due to WFQ averaging over all the 40-hop sessions. These delays have little multiplicative behavior. This is because in this
network there is little contention among packets. S-CEDF produces similar end-to-end delays.
V. C ONCLUSION
We have described a work-conserving scheduling discipline
C OORDINATED -E ARLIEST-D EADLINE -F IRST with end-to-end
delay bound,
Ki L
i + 4Li =" + X
max log():
mk
i
r
k=1
CEDF uses randomization and simple coordination to ensure
that once a packet passes through its first server it can pass
through all its subsequent servers quickly. Under CEDF, a
i
session-i packet does not accumulate a delay of LiK
i over Ki
hops, and therefore its delay bound is smaller than that of the
WFQ
Simple-CEDF
150
150
rate=0.03
rate=0.05
rate=0.1
rate=0.15
rate=0.2
140
rate=0.03
rate=0.05
rate=0.1
rate=0.15
rate=0.2
140
130
120
120
110
110
100
100
90
90
Mean Delay
Mean Delay
130
80
70
80
70
60
60
50
50
40
40
30
30
20
20
10
10
0
0
10
15
20
25
30
Session Length
35
40
45
50
10
15
20
130
120
120
110
110
100
100
90
90
98th Percentile
98th Percentile
50
rate=0.03
rate=0.05
rate=0.1
rate=0.15
rate=0.2
140
80
70
60
80
70
60
50
50
40
40
30
30
20
20
10
10
0
0
10
15
20
25
30
Session Length
35
40
45
50
Weighted Fair Queueing discipline. We have also presented simulation results to show that the performance of CEDF and WFQ
can be comparable to the analytical bounds.
The major open problem is to reduce the delay bound still further. The ultimate goal is a simple protocol with a delay bound,
Ki L
X
i + Li +
max :
mk
i
r
k=1
10
15
[5]
[6]
[7]
[8]
[9]
ACKNOWLEDGMENTS
[4]
45
Simple-CEDF
130
[3]
40
150
rate=0.03
rate=0.05
rate=0.1
rate=0.15
rate=0.2
140
[2]
35
WFQ
150
[1]
25
30
Session Length
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
20
25
30
Session Length
35
40
45
50
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
be close to 1.
X Si
i + (tinj ; t0 )i :
Gm (1 ; "=2)rm Gm :
T
i
i2N
i2N
The total number of session-i tokens during (t 0 ] is at least
;t0 ;Ti . Therefore, the total number of session-i bits consumed
P
Ti
m Gm
A Chernoff-type argument shows that Pr
X
r
i
i2N
during t0 ] is at least,
is small. (We omit details here, since the calculation is standard.) In particular,
; t0 ; Ti (S ; L ):
i
i
Ti
P
;"3 (1;")rm Gm =(48Lmax)
m
Pr i2N Xi r Gm e
:
Hence,
E
Xi ]
; t0 ; Ti (S ; L ) + (t ; t )
i
i
i
inj 0 i
Ti
) ;t0T; Ti (i Ti + Li ; Li ) i + tinj ; t0
i i
i
4
L
) tinj + i + " i :
i
i