DONAR Decentralized Server Selection for Cloud Services

Document Sample
DONAR Decentralized Server Selection for Cloud Services Powered By Docstoc
					                      DONAR
Decentralized Server Selection for Cloud Services

B96B02016 生化科技四 張煥基
B97901184 電機三  姜慧如

                                                    2011.06.21
       Introduction
The trend toward geographically-diverse
server placement will only continue and
increasingly include smaller enterprises, with
the success of cloud-computing platforms
like Amazon AWS .
These services all need an effective way to
direct clients across the wide area to an
appropriate service location (or “replica”).
Replication Selection 兩大主
             流
                                             distributed
     central coordination
                                             heuristics(DONAR)


     (1) reliability                         同時解決
                                             (1) client performance
優點   (2) security                            (2) server load



     (1) single point of failure

     (2) attractive target for attackers     (1) nodes over-react based on their own local
                                             information
     (3) overhead
缺點                                           (2) the system does not balance replica load
                                             effectively
     (4) less responsive to sudden changes

     (5) scalability limitations
Replica-Selection System 必備特質
                     Customers should have a sufficiently
                     expressive interface to specify policies based
    [1] Expressive   on (some combination of) (1) performance,
                     (2)replica load, and (3) server and bandwidth
                     costs.

                     The system should offer reliable service to
     [2] Reliable    clients, as well as stable storage of customer
                     policy and replica configuration data.

                     Client requests should be directed to the
     [3] Accurate    service replicas as accurately as possible,
                     based on the customer’s replica-selection
                     policy.

                     The replica-selection system should respond
    [4] Responsive   quickly to changing client demands and
                     customer policies without introducing
                     instability.


     [5] Flexible    The nodes should support a variety of replica-
                     selection mechanisms

                     Only the customer, or another authorized
      [6] Secure     party,should be able to create or change its
                     selection policies.
本篇主角:DONAR

This paper presents DONAR, a distributed
system that can offload the burden of replica
selection, while providing these services with a
sufficiently expressive interface for specifying
mapping policies.
                           Roadmap
            Simple and expressive interface for customer policies
section 2

            Stable, efficient, and accurate distributed replica-selection algorithm
section 3

            Scalable, secure, reliable, and flexible prototype system
section 4
            Experiments in Section 5 evaluate both our distributed algorithm
            operating at scale and a small-scale deployment of our prototype
section 5   system


            compares DONAR to related work
section 6

            concludes
section 7
      2.1 Customer Goals
     •    Customers use DONAR to optimally pair clients with service
          replicas



minimize the network latency


balance load across all replicas


billing costs
        2.2 Application Programming
                  Interface
                              s = create ()
create a DONAR service

add a replica instance        i = add (s, repl, ttl) time-to-live period (ttl)


set split weight              set (s, i, wi, εi)

set bandwidth cap             set (s, i, Bi)
match a client-replica pair
                              match (s, clnt, i)

prefer a particular replica   preference (s, clnt, i)

remove a replica instance
                              remove (s, i)
                           Roadmap
            Simple and expressive interface for customer policies
section 2

            Stable, efficient, and accurate distributed replica-selection algorithm
section 3

            Scalable, secure, reliable, and flexible prototype system
section 4
            Experiments in Section 5 evaluate both our distributed algorithm
            operating at scale and a small-scale deployment of our prototype
section 5   system


            compares DONAR to related work
section 6

            concludes
section 7
3.1 Global Replica-Selection Problem


  若想提高網路效能,就得以accurate load
  distribution為代價。
  Our goal is to minimize this performance penalty
3.2 Distributed Mapping Service


每個 mapping node 各有其負責的 clients


The node maps the client to a replica    ,
and returns the result to that client.
3.2 Distributed Mapping Service




 所有          mapping node
 clients 的   n 所有的          所有從 client c 而來
 traffic,    traffic 中,從    ,經過 mapping
 node n 所    client c 而來的   node n 的 traffic,
 佔的比例        比例             流入replica i的比例,
                            i.e., ∑i Rnci = 1
  3.3 Decentralized Selection Algorithm

        optimization decomposition : 藉由 algorithmic
        iterations, 讓 local decisions converge to the global
        optimum.




global performance
                      local client performance
local replica selection
                       (以一個 mapping node 的視野看世界)




 某特定 mapping node n, 將其所負    loadn = load,∀n
 責之所有 clients 的 traffic 引到   ;超出預期流量的罰款。
 replicas 所需的 performance
 penalty.
The core components of the algorithm are the local
 updates by each mapping node, and the periodic
            updates of replica prices.
                                  overhead
centralized
solution

                                                                          每個 mapping node 會有 |N-1| 個
                                                                          mapping node 的鄰居。他們會告訴
                       Each node needs to share its mapping decisions      此mapping node   個與replica
                       of size    and each replica’s price λi needs to be      mapping 有關的消息。
distributed solution   known by each node. This implies      messages,
                                                                           個 replicas 會告訴每一個
                       each of size                                      mapping node 他們的 price
                       computational complexity is of size
DONAR’s system design
DONAR’s system
   design
Distribution optimization---tracking requests
               geographically
        A group of
     similarly located
        end-hosts.
Distributed Optimization--
Tracking Requests Geographically
      Distributed Optimization--
Exponentially weighted moving average
Distributed Optimization--
    Known cost assumption
DONAR’s System Design
Decomposed Local Problem
   For Some Node (n*)
DONAR Algorithm
DONAR Algorithm
DONAR Algorithm
DONAR Algorithm
DONAR Algorithm
Better!
DONAR’s System Design
Protocol-level mechanisms for
 Wide-area replica selection
Data Retrieving
    Steps
Data Retrieving Steps(cont.)
DONAR’s System Design
Secure Registration and
  Dynamic Updates
DONAR’s System Design
Distributed Data Storage
                          CRAQ
(Chain Replication with Apportioned Queries)

“while maintaining the strong consistency
properties of chain replication, provides lower
latency and higher throughput for read operations
by supporting apportioned (分攤) queries: that is,
dividing read operations over all nodes in a chain, as
opposed to requiring that they all be handled by a
single primary node.”
DONAR’s System Design
IP Anycast
Software Architecture
Results: DONAR Curbs Volatility
Results: DONAR Minimizes Distance
Conclusions
Thank You

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:10
posted:5/4/2012
language:English
pages:49