Fast and scalable priority queue architecture for highspeed network switches

1
Fast and scalable priority queue architecture for
high-speed network switches
Authors Ranjita Bhagwan, Bill Lin Publisher
IEEE INFOCOM 2000 Present Min-Yuan Tsai
(???) Date January, 04, 2007
Department of Computer Science and Information
Engineering National Cheng Kung University,
Taiwan R.O.C.
2
Outline

• 1. Introduction
• 2. The P-HEAP Data Structure
• 3. Priority Queue Operations on the P-HEAP
• 4. Pipelining The P-HEAP Operations

3
Introduction

• To implement the priority-based scheduling
  policy, priority queues can be used to maintain a
  real-time sorting of the queue elements in a
  decreasing order of priorities.
• The task of highest-priority-first scheduling can
  be reduced to a simple removal of the top queue
  element.
• To maintain this real-time sorting at link
  speeds, a fast hardware priority queue
  implementation is essential.

4
Introduction (contd.)

• In this paper, a new pipelined priority queue
  architecture is described.
• The architecture is based on a novel data
  structure called a Pipelined-heap (or P-heap) for
  short. This data structure is similar to a
  conventional binary heap.
• In addition to being very fast, our priority
  queue architecture also scales very well with
  respect to both the number of priority levels and
  the queue size.

5
Introduction (contd.)

• The Queue Manager consists
• primarily of three parts
• Priority Assignment Unit
• Stamps the incoming packets with
• a certain priority (decided depending
• upon the scheduling algorithm)
• P-heap Manager
• Contains the P-heap priority queue
• Queue Controller
• Maintains a lookup table with entries
  corresponding to each
• priority value.
• Each entry consists of a pointer to a list of
  packets of the same priority. (refer this list as
  a priority list)

6
Outline

• 1. Introduction
• 2. The P-HEAP Data Structure
• 3. Priority Queue Operations on the P-HEAP
• 4. Pipelining The P-HEAP Operations

7
The P-HEAP Data Structure
8
The P-HEAP Data StructureB

• Definition of Binary array (B)
• Bi a node in B with index i. (Contain 3
  fields)
• Bi.active Boolean field. (True means the node
  Bi is filled with a valid priority value
  (active). False means empty (inactive))
• Bi.value If the node is active, holds the
  actual priority value.
• Bi.capacity Contains the number of inactive
  nodes in the sub-tree rooted at Bi .
• Lj the set of nodes of the jth level of B
• We define a trio for every node in B as follows
  
• The ith trio of B ?i Bi, Bleft(i),
  Bright(i)

9
The P-HEAP Data StructureB (contd.)

• The P-heap property has to be satisfied by every
  node Bi of B except the leaves.

10
The P-HEAP Data StructureT

• Definition of Token array (T)
• Ti associate with level Li in of B. (Contain 3
  fields)
• Ti.operation Holds an instruction (the
  operation need to be executed at level Li)
• only holds one of the four instructions, enq,
  deq, edq, nop
• Ti.value Holds the priority value that need to
  be inserted into B.
• Ti.position Holds the index of a node at level
  Li of B.

11
Outline

• 1. Introduction
• 2. The P-HEAP Data Structure
• 3. Priority Queue Operations on the P-HEAP
• 4. Pipelining The P-HEAP Operations

12
Enqueue OperationValid Paths

• A valid path Bi1 ? Bi2 ? ? Bil where
  Bij belong to Lj and ij parent (ij 1)

• ? For all Bij, Bij.capacity gt 0.

13
Enqueue procedure (Enqueue 9)
9
9lt16
9
9lt14
9gt7, SWAP
9
9
7
7
14
Enqueue procedure (contd.)

• After Enqueue

15
Dequeue procedure
16
Dequeue procedure (contd.)

• After Dequeue

17
Enqueue-Dequeue procedure
18
Outline

• 1. Introduction
• 2. The P-HEAP Data Structure
• 3. Priority Queue Operations on the P-HEAP
• 4. Pipelining The P-HEAP Operations

19
Pipelining The P-HEAP Operations

• Define the ith pipeline window ai as
• The function f ( i ) executes one of the three
  local procedures on pipeline window ai depending
  on the operation field of Ti

20
The Pipelining Operations of a 4-level
P-HEAP Example

• We define a P-heap cycle is the maximum time
  required to execute any instance of the three
  local operations.
• Operations as follows
• Enqueue (9)
• Enqueue (4)
• Dequeue
• Enqueue-Dequeue (2)

21
The Pipelining Operations of a 4-level
P-HEAP Cycle 1

• Enqueue (9) a1

9lt16
22
The Pipelining Operations of a 4-level
P-HEAP Cycle 2

• Enqueue (9) a2

9lt14
23
The Pipelining Operations of a 4-level
P-HEAP Cycle 3

• Enqueue (9) a3
• SWAP
• Enqueue (4) a1

4lt16
9gt7, SWAP
24
The Pipelining Operations of a 4-level
P-HEAP Cycle 4

• Enqueue (9) a4
• Complete
• Enqueue (4) a2

4lt10
25
The Pipelining Operations of a 4-level
P-HEAP Cycle 5

• Enqueue (4) a3
• Complete
• Dequeue (4) a1

14gt10
26
The Pipelining Operations of a 4-level
P-HEAP Cycle 6

• Dequeue (4) a2

9gt8
27
The Pipelining Operations of a 4-level
P-HEAP Cycle 7
Replace by 2

• Dequeue (4) a3
• Complete
• Enqueue
• -Dequeue (2) a1

2lt10,SWAP
7gt5
28
The Pipelining Operations of a 4-level
P-HEAP Cycle 8

• Enqueue
• -Dequeue (2) a2
• Complete

2lt4, SWAP
29
The Pipelining Operations of a 4-level
P-HEAP Result