Efficient, ProximityAware Load Balancing for DHTBased P2P Systems

1
Efficient, Proximity-Aware Load Balancing for
DHT-Based P2P Systems

• Yingwu Zhu, Yiming Hu

Presented by Ki
Appeared on IEEE Trans. on Parallel and
Distributed Systems, vol. 16, no. 4, April 2005
2
Outline

• Introduction
• System Design
• Proximity-Aware Load Balancing
• Experimental Evaluations
• Conclusions

3
Introduction

• DHT-based P2P systems like Chord, Pastry,
  Tapestry, CAN,
• Provide a distributed hash table abstraction for
  object storage and retrieval
• Assume nodes are homogeneous
• Two main drawbacks
• Imbalance load
• Node heterogeneity

4
Introduction

• Existing solutions for DHT load balancing
• Some ignore node heterogeneity
• Some ignore proximity information
• This work proposes a proximity-aware load
  balancing scheme which considered the above two
  aspects

5
System Design - Basic

• Virtual Server (VS)
• Act like an autonomous peer
• Responsible for a contiguous portion of the DHTs
  identifier space
• Responsible for certain amount of load
• Physical Peer
• Can host multiple virtual servers
• Heavily loaded physical peer
• Moves some of its virtual servers to other
  lightly loaded physical peers to achieve load
  balancing
• Movement of virtual server can be viewed as a
  leave operation followed by a join operation

6
System Design Four Phrases

• The load balancing scheme proceed in 4 phrases
• Load balancing information (LBI) aggregation
• Collect load and capacity information of whole
  system
• Node classification
• Classify nodes into HEAVY, LIGHT, NEUTRAL
• Virtual server assignment (VSA)
• Determine the VS assignment to make HEAVY nodes
  becomes LIGHT
• Proximity-aware
• Virtual server transferring (VST)

7
System Design k-ary Tree on DHT

• A k-ary tree (KT) is built on top of the DHT
• Occupying the same identifier space
• KT root node is responsible for the entire
  identifier space
• Each child node is responsible for a portion of
  their parents identifier space

8
System Design k-ary Tree on DHT

• A KT node is responsible for identifier space
  region
• key host is obtained by procedure
  plant_KT_node()
• Keeps track of its parent and children
• KT node, X, is planted into a virtual server
  which responsible for X.key
• Example (KT Node X, VS S)
• X.region (3,5 ? X.key 4
• Ss region (3, 6
• X is planted in S

9
System Design k-ary Tree on DHT

• For KT node X, its region
• is further divided into k parts, then taken by
  its k children
• Until
• Xs region is completely covered by its hosting
  VS
• Each KT node periodically check if its region is
  completely covered by its VS
• Yes ? delete the existing children
• No ? keep k children

10
Load Balancing Information Aggregation

• Load Balancing Information (LBI) of node i
• Li ? total load of VS in node i
• Ci ? capacity of node i
• Li,min ? minimum load of VS in node i

X.host randomly responds to one of its VS only
11
Load Balancing Information Aggregation

• KT root node obtains the system-wide LBI
• L ? Total load
• C ? Total capacity
• Lmin ? minimum load of all VS in system
• KT root node distribute the system-wide LBI
• Along the tree, back to the leaf nodes, VS and
  finally the DHT node

12
Node Classification

• System-wide utilization L / C
• Utilization of node i Li / Ci
• Define Ti (L / C e) Ci
• e is a parameter for trade off between amount of
  load movement and quality of load balance
• Classification
• HEAVY node if Li gt Ti
• LIGHT node if (Ti Li) gt Lmin
• NEUTRAL node if 0 lt (Ti Li) lt Lmin

13
Virtual Server Assignment

• Each HEAVY DHT node i
• Randomly choose a subset of its VS that minimizes
  s.t.
• Minimized the amount of load movement
• VSA info. ltLi,1, vi,1, IPigt ltLi,m, vi,m, IPigt
• Each LIGHT DHT node j
• VSA info. lt?Lj Tj Lj, IPjgt
• This VSA information propagates upward along the
  KT

14
Virtual Server AssignmentProximity Ignorant

• Each KT node i
• Collects the VSA information until the a
  pairing_threshold
• Uses a best-fit heuristic to reassign VS
• Reassign VS in HEAVY nodes to LIGHT nodes
• And minimize load movement
• DHT nodes of reassigned VS get notified while the
  rest of VSA information propagate to is parent

15
Virtual Server AssignmentProximity Aware

• Use landmark clustering
• Measure distance to a number of landmark nodes
• Obtain a landmark vector ltd1, d2, , dmgt which
  represent point in a m-dimensional space
• Nodes with close landmark vectors are in general
  physically close
• Transform the landmark m-dimensional space to the
  DHT identifier space and obtains a DHT key, LMi
• By Hilbert curve, i.e. Nm ? N
• Proximity preserving

16
Virtual Server AssignmentProximity Aware

• Each node i independently determines its landmark
  vector and its corresponding DHT key, LMi
• Node publish its VSA information in the DHT
  network with the DHT key LMi
• Node j that responsible for the region contains
  LMi receive the VSA information
• Node j propagate the received VSA information
  into the KT

17
Virtual Server Transferring

• Upon receiving the reassigned VSA info.
• HEAVY node transfer the reassigned VS to the
  LIGHT node
• The transfer of VS would cause KT to reconstruct
• Lazy migration
• Reconstruction of KT only after the completion of
  all transfers

18
Experimental Evaluations

• Load Balancing Information aggregation and
  virtual server assignment latency

19
Experimental Evaluations

• Underlying network
• Generated by GT-ITM with about 5000 nodes
• Underlying DHT overlay
• Chord with 4096 nodes
• 5 virtual servers in each node, exponential
  identifier space
• k-ary tree
• k 2
• Pairing threshold
• 50
• Landmark node count
• 15

20
Experimental Evaluations

• Nodes carry loads proportional to their
  capacities by reassigning virtual servers

21
Experimental Evaluations

• Cumulative distribution of moved load

• Proximity-aware
• 36 within 2 hops, 57 within 10 hops
• Proximity-ignorant
• 17 within 10 hops

• Proximity-aware
• Reduce load balancing cost
• Fast and efficient load balancing

22
Experimental Evaluations

• Effect of node churn (join leave)
• Overhead (Md Ms) / Ms
• Md number of VSA messages with node churn
• Ms number of VSA messages without node churn

23
Conclusions

• This work focuses on an efficient,
  proximity-aware load balancing scheme
• Align load distribution and node capacity
• Use proximity information to guide load
  reassignment and transferring