1 / 30

# Wire Efficient Topology

Wire Efficient Topology. In the 3D world. For n nodes, bisection area is O(n 2/3 ) For large n, bisection bandwidth is limited to O(n 2/3 ) Bill Dally, IEEE TPDS, [Dal90a] For fixed bisection bandwidth, low-dimensional k-ary n-cubes are better (otherwise higher is better)

## Wire Efficient Topology

E N D

### Presentation Transcript

1. Wire Efficient Topology

2. In the 3D world • For n nodes, bisection area is O(n2/3 ) • For large n, bisection bandwidth is limited to O(n2/3 ) • Bill Dally, IEEE TPDS, [Dal90a] • For fixed bisection bandwidth, low-dimensional k-ary n-cubes are better (otherwise higher is better) • i.e., a few short fat wires are better than many long thin wires

3. Equal cost in k-ary d-cubes • Equal number of nodes? • Equal number of pins/wires? • Equal bisection bandwidth? • Equal area? Equal wire length? What do we know? • switch degree: d diameter = dk/2 • total links = Nd • pins per node = 2wd • bisection = 2kd-1= 2N/k links in each directions • 2Nw/k wires cross the middle

4. Latency(d) for N with Equal Link Width • total links(N) = Nd • switch degree, number of channels, and channel length : for free

5. Latency with Equal Pin Count (total # of wires per node constant) • Baseline d=2, has w = 32 (128 wires per node) • fix2dw pins => w(d) = 64/d: routing delayD=2 cyc • With more dimension, more channels, but thinner • distance down with d, but channel time increases

6. Wire Efficient Topology • Selecting a network that gives us the lowest latency, T(0,L), for a given wiring density, BW, number of nodes, N, and message length, L. • trade off channel bandwidth Wagainst diameter D • wormhole routing makes latency the sum of two components D and L/W • the zero-load latency of the network • assuming a constant BW(N=2Nw/k), average distance : (k-1)d/2, channel width k/2 • the minimum latency occurs at a smaller d (larger k) than the point where D = L/W

7. Latency with Equal Bisection Width Distance dominatesMessage length dominates Low-dimensional networksare even more attractive than this comparison indicates - It is not feasible to scale Bw linearly with N Bw = Nr: 0.5 £r£ 0.66 is more reasonable. - This comparison assumes that all channels are the same speed. Lower dimensional networks have shorter wires. - Low-dimensional networks better exploit locality. - Router complexity is proportional to the number of dimensions. T 256 nodes 160 140 120 100 80 60 40 20 16K nodes 1M nodes 0 5 10 15 20 d L=150, Bw =N

8. Larger Routing Delay (w/ equal pin) • Dally’s conclusions strongly influenced by assumption of small routing delay D=20 D=2

9. Latency under Contention • Optimal packet size? Channel utilization? 1000 nodes distance m4-- Message size: 4 phits Equal channel width N=1024(d2,k32), 1000(d3,k10)

10. Saturation • Fatter links shorten queuing delays

11. 350 300 D=2 250 200 Latency 150 100 50 n8, d3, k10 n8, d2, k32 0 0 0.05 0.1 0.15 0.2 0.25 Phits per cycle per processor Phits per cycle • higher degree network has larger available bandwidth • cost?

12. Discussion • Rich set of topological alternatives with deep relationships • Design point depends heavily on cost model • nodes, pins, area, ... • Wire length or wire delay metrics favor small dimension • Long (pipelined) links increase optimal dimension • Need a consistent framework and analysis to separate opinion from design • Optimal point changes with technology

13. Networks: Routing and Design

14. Outline • Routing • Switch Design • Flow Control • Case Studies

15. Routing • Recall: routing algorithm determines • which of the possible paths are used as routes • how the route is determined • R: N x N -> C, which at each switch maps the destination node nd to the next channel on the route • Issues: • Routing mechanism • arithmetic • source-based port select • table driven • general computation • Properties of the routes • Deadlock free

16. Routing Mechanism • need to select output port for each input packet • in a few cycles • Simple arithmetic in regular topologies • ex: Dx, Dy routing in a grid • west (-x) Dx < 0 • east (+x) Dx > 0 • south (-y) Dx = 0, Dy < 0 • north (+y) Dx = 0, Dy > 0 • processor Dx = 0, Dy = 0 • Reduce relative address of each dimension in order • Dimension-order routing in k-ary d-cubes • e-cube routing in n-cube

17. Routing Mechanism (cont) • Source-based • message header carries series of port selects • used and stripped en route • CRC? Packet Format? • CS-2, Myrinet, MIT Artic • Table-driven • message header carried index for next port at next switch o = R[i] • table also gives index for following hop o, i’ = R[i ] • ATM, HPPI P3 P2 P1 P0

18. Properties of Routing Algorithms • Deterministic • route determined by (source, dest), not intermediate state (i.e. traffic) • Adaptive • route influenced by traffic along the way • Minimal • only selects shortest paths • Non-minimal • Deadlock free • no traffic pattern can lead to a situation where no packets move forward

19. Deadlock Freedom • How can it arise? • necessary conditions: • shared resource • cyclic dependency • incrementally allocated • non-preemptible • think of a channel as a shared resource that is acquired incrementally • source buffer then dest. buffer • channels along a route • How do you avoid it? • constrain how channel resources are allocated • ex: dimension order • How do you prove that a routing algorithm is deadlock free?

20. Proof Technique • resources are logically associated with channels • messages introduce dependences between resources as they move forward • need to articulate the possible dependences that can arise between channels • show that there are no cycles in Channel Dependence Graph • find a numbering of channel resources such that every legal route follows a monotonic sequence => no traffic pattern can lead to deadlock • network need not be acyclic, on channel dependence graph

21. Example: k-ary 2D array • Thm: x,y routing is deadlock free • Numbering • +x channel (i,y) -> (i+1,y) gets i+1 • similarly for -x with 0 as most positive edge • +y channel (x,j) -> (x,j+1) gets N+j+1 • similary for -y channels • any routing sequence: x direction, turn, y direction is increasing

22. Channel Dependence Graph

23. More examples: • Why is the obvious routing on X deadlock free? • butterfly? • tree? • fat tree? • Any assumptions about routing mechanism? amount of buffering? • What about wormhole routing on a ring? 2 1 0 3 7 4 6 5

24. Deadlock free wormhole networks? • Basic dimension order routing techniques don’t work for k-ary d-cubes • only for k-ary d-arrays (bi-directional) • Idea: add channels! • provide multiple “virtual channels” to break the dependence cycle • good for BW too! • Do not need to add links, or xbar, only buffer resources • This adds nodes the the CDG, remove edges?

25. Breaking deadlock with virtual channels

26. Up*-Down* routing • Given any bidirectional network • Construct a spanning tree • Number of the nodes increasing from leaves to roots • UP increase node numbers • Any Source -> Dest by UP*-DOWN* route • up edges, single turn, down edges • Performance? • Some numberings and routes much better than others • interacts with topology in strange ways

27. Turn Restrictions in X,Y • XY routing forbids 4 of 8 turns and leaves no room for adaptive routing • Can you allow more turns and still be deadlock free +Y -X +X -Y

28. Minimal turn restrictions in 2D +y -x +x north-last -y negative first

29. Example legal west-first routes • Can route around failures or congestion • Can combine turn restrictions with virtual channels

30. Adaptive Routing • Essential for fault tolerance • at least multipath • Can improve utilization of the network • Simple deterministic algorithms easily run into bad permutations • fully/partially adaptive, minimal/non-minimal • can introduce complexity or anomalies • little adaptation goes a long way!

More Related