ETSN10 Formula Sheet

Probability Continuous random variables

P (Ā) = 1 − P (A) Mean: Z
P (A ∪ B) = P (A) + P (B) − P (A ∩ B) µ= xp(x)dx
P (A ∩ B) Variance:
P (A|B) =
P (B) Z
A, B mutually exclusive: V ar(X) = (x − µ)2 p(x)dx
P (A ∩ B) = 0
Negative exponential:
A, B independent:
p(x) = αe−αx
P (A ∩ B) = P (A)P (B)
P (X ≤ x) = 1 − e−αx
CDF:
F (x) = P (X ≤ x) 1
µ=
α
PDF:  2
pk = p(k) = P (X = k) 1
V ar(X) =
α
Discrete random variables Gaussian:
1 (x−µ)2

Mean: p(x) = √ e− 2σ2

X σ 2π
µ= kpk
Variance: Multiple random variables
X
V ar(X) = (k − µ)2 pk = E[X 2 ] − (E[X])2 Joint CDF:
Standard deviation:
p F (x, y) = P (X ≤ x, Y ≤ y)
σ= (V ar(X))
X, Y independent:

E[aX + b] = aE[X] + bV ar(aX + b) = a2 V ar(X) F (x, y) = F (x)F (y)

Bernoulli random variable:

E[X + Y ] = E[X] + E[Y ]
p0 = 1 − p
p1 = p E[XY ] = E[X]E[Y ]
µ=p
V ar(X + Y ) = V ar(X) + V ar(Y )
σ 2 = p(1 − p)
Properties of random variables X, Y , constants
Geometric:
a, b:
pk = (1 − p)k−1 p E[aX + b] = aE[X] + b
1
µ= V ar(aX + b) = a2 V ar(X)
p
1−p E[X + Y ] = E[X] + E[Y ]
V ar(X) =
p2 Covariance:
Poisson random variable:
Cov(X, Y ) = E[(X−E[X])(Y −E[Y ])] = E[XY ]−E[X]E[Y ]
λk
pk = e−λ
k! Correlation coefficient:
µ=λ Cov(X, Y )
r(X, Y ) =
σ2 = λ σx σy

1
Stochastic Processes CSMA utilisation:
Autocorrelation:
A = kp(1 − p)k−1
R(t1 , t2 ) = E[x(t1 )x(t2 )]
1−A
Autocovariance: E[w] =
A
C(t1 , t2 ) = Cov(x(t1 ), x(t2 )) 1
U=
1 + 2a + a(1 − A)/A
Sampling and random numbers
Sample mean:
n
Queueing Systems
1X
z̄ = zi
n i=1 Kendall Notation parameters:
Sample variance: 1. Arrival distribution
n
1X
V̂ = (zi − z̄)2 2. Service distribution
n i=1
3. Number of servers
Unbiased sample variance (Bessel’s correction):

1 X
n 4. Total capacity (default: infinite)
s2 = (zi − z̄)2
n − 1 i=1
5. Population size (default: infinite)
Confidence intervals:
6. Service disciplien (default: FIFO)
σ
Confidence = P r(|z̄ − µ| ≤ α × √
n Little’s Law:
α = 1.96 gives confidence of 95%. E[R] = λE[TR ]
Inverse method:
Occupancy:
X = F −1 (Y ) λ
ρ=
µ

Medium Access Control

M/M/1
FDMA/TDMA rate of work:
Number of items in the system:
n
1 X
η= ρi ρ
n × ν i=1 E[R] =
1−ρ
Polling efficiency:
Total time in system:
Tt
E= 1
Tt + Tidle + Tpoll E[TR ] =
µ(1 − ρ)
ALOHA throughput:
Waiting time:
S = Ge−G
ρ
Slotted ALOHA throughput: E[TW ] =
µ(1 − ρ)
S = Ge−2G Delay bound:
1-persistent CSMA throughput:
P r(TR ≤ t) = 1 − e−µ(1−ρ)t
−G
Ge (1 + G)
S=
G + e−G
M/M/2
Non-persistent CSMA throughput:
1
G E[TR ] =
S= µ(1 + ρ)(1 − ρ)
1+G

2
M/M/1/n Network Architectures
Blocking probability: Packet reception rate:
1−ρ n
PB = ρ S
1 − ρn+1 P RR =
T
Carried traffic:
Cellular frequency re-use:
γ = λ(1 − PB )
√
Little’s Law: D = R 3K
E[R] = γE[TR ]
Number of items in the system:
n
Congestion Control
1−ρ X i
E[R] = iρ
1 − ρn+1 i=0 Token bucket:
R = ρT + β
Erlang (M/M/n/n)
Offered traffic: Random Early Discard (RED)
A = λh
(avg − T hmin
Service rate: Pb = Pmax
1 T hmax − T hmin
µ=
h Pb
Erlang loss function: Pa =
1 − count × Pb
An
En (A) = PB = Pnn! Aj
j=0 j!
TCP
Carried traffic: Max bandwidth:
Ac = A(1 − PB )
r
M SS × C 3
BWmax = √ C=
Lost traffic: RT T × p 2
A − Ac
Normalised throughput:
Jackson Networks (
1 W ≥ 2RD
Probability a packet leaves the network from node S= W
2RD W < 2RD
i:
Xn
1− rij Expected (average) round trip time:
j=1
K 1
Total arrival rate to node i: ERT T (K+1) = ERT T (K)+ RT T (K+1)
n
K +1 K +1
X
λi = γi + λj rji Smoothed round trip time:
j=1

SRT T (K+1) = α×SRT T (K)+(1−α)×RT T (K+1)

Queueing Disciplines
Kleinrock Conservation Law: Retransmission timeout with SRTT:
N
X RT O(K+1) = min{U B, max{LB, β×SRT T (K+1)}}
ρn qn = C
n=1
Van Jacobson’s algorithm:
Processor Sharing:
DRT T (K+1) = (1−α)×DRT T (K)+α×(SRT T −ERT T )
Fiα = Siα + Piα
Siα = max{Fi−1
α
, Aα
i } RT O = ERT T + 4 × DRT T
Generalised Processor Sharing:
Exponential backoff:
Pα
Fiα = Siα + i
wα RT Oi+1 = q × RT Oi

3
Other useful formulas
Geometric series:
n−1
X 1 − rn
ark = a
1−r
k=0

∞
X a
ark =
1−r
k=0

Expectation of geometric series:

∞
X ar
kark =
(1 − r)2
k=0