Sorting in the Bit Model, Lower Bounds

Tom Kelliher, CS 315

Feb. 3, 1999

Administrivia

Announcements

Assignment

Read Sections 1.1.4, 1.1.5.

From Last Time

Sorting in the Bit Model.

Outline

1. Sorting in the bit model.

2. Example problems.

3. Lower bound techniques.

Coming Up

Lower bound counterexample, properties of fixed-connection networks.

Sorting in the Bit Model

1. Why get more detailed? To see how much work (e.g., #number of transistors) is involved in, for instance, comparison.

2. A linear array bit-level model of a compare cell:

   

   Compare, notify.

   Assume we have k-bit numbers. Run time?

3. Can we do better? Yes. We can do comparisons in parallel and merge the partial results:

   

   Compare, notify.

   Run time?

   Notes:

   1. Why not use a height 1 tree?

   2. Fan-in and fan-out in circuits.

Sorting

1. Run-time for sorting on a linear array of linear array comparators. Binary tree comparators.

   What do they look like?

2. Pipelining the linear comparators. Pipelining the compare:

   

   Five in, four out.

   A lot of parallel work. Systolic computation.

3. Pipelining the sort. Just a linear array of the comparators:

   

   Running time?

Example Problems

1.2

Consider two algorithms for solving a a problem of size M, one that runs in M steps on an M-processor machine and one that runs in steps on an -processor machine. Which algorithm is more efficient?

Assume that the cost of running a P-processor machine for T steps is , where and are constants and P = M or . For what values of and is the first algorithm cheaper to run? For what values of and is the second algorithm cheaper to run?

1.3

Consider two algorithms for solving a problem of size M, one that runs in M steps on an M-processor machine and one that runs in steps on an -processor machine. Which algorithm will run faster on an N-processor machine? (Hint: Your answer depends on the relative sizes of M and N, and you will need to use the fact that an N-processor machine can simulate a P-processor machine with slowdown .)

Lower Bound Arguments

The following are tools we can sometimes use to characterize a problem-solving network and, hence, understand its failings.

Assume the bit model, so we have an sorting network in the following.

1. I/O bandwidth.

   What is the I/O bandwidth of the sorting network? This makes sorting .

2. Diameter of the network: Defined as the maximum distance between two vertices in the graph.

   What is the diameter of the sorting graph?

   Can we construct a sorting problem that will force the two most distant cells to communicate?

   What's the lower bound from this perspective?

3. Bisection width of the network: Defined as the number of edges which must be cut to split the graph in ``half.''

   What is the bisection width of the sorting graph?

   Can we construct a sorting problem that will force all the data to cross this boundary?

   What's the lower bound from this perspective?