Skip to main content

Clos network and STS/TST blocking

Short answer

Cls network is a three-stage switch: N inputs divided into N/n first-stage n×k arrays, k middle (N/n)×(N/n) arrays, N/n third-stage k×n arrays; strictly non-blocking if k = 2n-1. Blocking probability: STS: P_B = [1 - (1 - p/beta)^2]^k, beta = k/N; TST: P_B = [1 - (1 - p/alpha)^2]^l, alpha = l/c. TST blocking is far lower because l >> k.

Define what the Clos network is. Derive the expression for blocking probability of both STS and TST switches, and compare them.

Clos network

A Clos network is a three-stage (multi-stage) switching network, proposed by Charles Clos (1953), in which NN inputs are divided into N/nN/n first-stage arrays of size n×kn \times k, connected to kk middle arrays of size Nn×Nn\frac{N}{n} \times \frac{N}{n}, which connect to N/nN/n third-stage arrays of size k×nk \times n. Every array of one stage has exactly one link to every array of the next stage.

Nx=Nn(nk)⏟stage 1+k(Nn)2⏟stage 2+Nn(kn)⏟stage 3=2Nk+k(Nn)2\begin{aligned} N_x &= \underbrace{\tfrac{N}{n}(nk)}_{\text{stage 1}} + \underbrace{k\left(\tfrac{N}{n}\right)^2}_{\text{stage 2}} + \underbrace{\tfrac{N}{n}(kn)}_{\text{stage 3}} \\ &= 2Nk + k\left(\frac{N}{n}\right)^2 \end{aligned}

Clos showed that the network is strictly non-blocking when k=2n−1k = 2n - 1 (worst case: n−1n-1 other inputs of the calling array and n−1n-1 outputs of the called array use different middle arrays, so one more is needed). With the best choice n≈N/2n \approx \sqrt{N/2}, the minimum is Nx≈4N(2N−1)N_x \approx 4N(\sqrt{2N} - 1), much less than N2N^2.

Blocking probability of STS

 STS:     S (N x k) -> T (k TSI modules) -> S (k x N)
 Lee graph: k parallel paths, each = 2 links in series

In STS, a call from input highway A to output highway B in given slots can go through any of the kk time-switch modules. Each path uses a link from the first space stage to a module and a link from the module to the second space stage. With link occupancy p′=p/βp' = p/\beta (β=k/N\beta = k/N space expansion):

  • Probability a link is free =1−p′= 1 - p'; both links of a path free =(1−p′)2= (1-p')^2.
  • Path blocked =1−(1−p′)2= 1 - (1-p')^2; all kk paths blocked:
p′=pβ,β=kNPB(STS)=[1−(1−p′)2]k\begin{aligned} p' &= \frac{p}{\beta},\quad \beta = \frac{k}{N} \\ P_B(\text{STS}) &= \left[1 - (1-p')^2\right]^{k} \end{aligned}

Blocking probability of TST

 TST:     T (c -> l slots) -> S (N x N) -> T (l -> c)
 Lee graph: l parallel paths (internal time slots)

In TST, the path from A to B passes through the space stage in any of the ll internal time slots. Each path uses one internal slot on A's highway and the same slot on B's highway. With time expansion α=l/c\alpha = l/c, p′=p/αp' = p/\alpha, and by the same argument:

p′=pα,α=lcPB(TST)=[1−(1−p′)2]l\begin{aligned} p' &= \frac{p}{\alpha},\quad \alpha = \frac{l}{c} \\ P_B(\text{TST}) &= \left[1 - (1-p')^2\right]^{l} \end{aligned}

Comparison

Since the number of paths in TST equals the number of internal time slots ll (32 to several hundred), while in STS it equals the number of time-switch modules kk (usually small), TST blocking is far lower for the same hardware. Example with p′=0.2p' = 0.2:

STS, k=8: (1−0.82)8=0.368≈2.8×10−4TST, l=32: (1−0.82)32=0.3632≈6.3×10−15\begin{aligned} \text{STS},\ k = 8 &: \ (1-0.8^2)^{8} = 0.36^{8} \approx 2.8 \times 10^{-4} \\ \text{TST},\ l = 32 &: \ (1-0.8^2)^{32} = 0.36^{32} \approx 6.3 \times 10^{-15} \end{aligned}
PointSTSTST
Number of alternative pathskk (time modules)ll (internal slots)
BlockingHigherMuch lower
Expansion byMore space arrays/modulesMore internal slots (cheap memory)
Cost for large exchangeHigherLower (time stages cheap)
ControlSimpler for small sizesMore memory, path search over slots
UseSmall exchangesMost large digital exchanges

✓ Answer checked by Claude Opus 5.5 (high effort)

All Telecommunications chapter 4 questions →