Docstoc

lec29-comparison.ppt

Document Sample
lec29-comparison.ppt Powered By Docstoc
					Comparing P2P Systems


          Anthony D. Joseph
          John Kubiatowicz
          CS294-4
Why so many systems?

   Many different types of target users
   Many different types of environments
   Many design choices
   Many hazards
   Many data types
   Many ….
Networks

   Chord         Similar interfaces
   CAN            –   DHT, DOLR
   Tapestry      Different design goals
                   –   Locality, Topology
   Pastry
                   –   Fault-tolerance
   Kademlia
   Viceroy
   Bamboo
   …
Systems We’ve Read About

   Freenet                    Gia
   Publis                     OceanStore
   SFS                        PAST
   Bayou                      Squirrel
   FARSITE                    CFS
   Logistical Networking      Ivy
   Pangaea                    PeerDB
   Pastiche                   PIER
Systems 1

   Freenet                                 Publis
     –   Anon, cens. resistant storage        –   FT, anon, censorship resistant
     –   Objects ref’d by SHA-1 hash              storage
         over content (GUID-CHK)              –   Tamper evident, src anon,
     –   Objs named by GUID-Signed                updatable, deniable
         Subspace Key pointing to             –   Persistent, extensible
         CHKs                                 –   Splits enc key into k shares
     –   Steepest Hill Climbing query         –   Retrieve k shares for content
         routing with TTL                     –   Static mapping of share
     –   Space allocated by popularity            locations to servers
     –   Power-law node degrees               –   Indirection-based (file) update
     –   Tolerates up to 30% failure              mechanism vulnerable to
                                                  server compromise
Systems 2

   SFS                                Bayou
    –   Auth, secure, encrypted         –   Replicated P2P DB
        client-server storage and               Atomic operations
        access control                          Whole DB replication
    –   ACL-based auth of               –   Operation-based updates
        individuals, groups, and        –   Tentative local commits
        groups of groups                    enforced by primary global
    –   Caching for speed and               commit
        availability                            Apps ctl data view
                                        –   Gossip-based info
                                            propagation
                                        –   Merge procedures for per-
                                            write conflict resolution
Systems 3

   FARSITE                               Logistical Networking
    –   P2P storage                        –   Network storage layer
    –   Max size ~105                      –   IBP: unreliable, transient
    –   Large-scale read-only                  byte-arrays on depots
        sharing, small-scale               –   Aggregation into exNodes
        read/write-sharing                         Can implement arbitrary
            Complex lease mechanism                reliability mechanisms
    –   Assumes user auth infra                    Analog to Unix inodes
    –   Byzantine ring formed for
        each namespace
    –   Reliability and availability
        through whole file
        replication
Systems 4

   Pangaea                                 Pastiche
    –   Server-based replication             –   P2P data replication for
    –   Assumes trusted servers                  whole machine backup
    –   Two-levels of servers:               –   Built on Pastry
            Gold                            –   Enc storage of
              – Fully connected                  immutable chunked data
                clique                       –   Network distance or
              – Strong maintenance
                                                 coverage based buddy
            Bronze                              choices
              –   Limited connectivity
    –   Last writer wins conflict
        resolution
Systems 5

   Gia                                       OceanStore
    –     Modified Gnutella protocol           –   Wide-area CS/P2P
    –     Argues against DHTs for                  replicated, robust, secure,
          this search type                         auth data storage
              Transient P2P clients           –   Built on Tapestry, Bamboo
              Keyword-based searches          –   Byzantine update commit
              Searching for hay instead       –   Per-write conflict resolution
               of needles
                                               –   Erasure coding based
    –     Capacity-based topology                  replication (robustness)
          adaptation                               with block caching
    –     Flow-ctrl for queries                    (performance)
Systems 6

   PAST                                Squirrel
    –   P2P archival storage             –   Decentralized P2P web
        model                                caching
            No updates                  –   Homestore model: stores
            Whole-file storage              content at home and
    –   Tries to balance per-                client nodes
        node storage load                –   Directory model: use
        (assumes ≤ 100x diff)                recent clients
    –   Replica and file diversion
        to maintain k copies
Systems 7
   CFS                                   Ivy
    –   P2P file storage                    –    R/W P2P file storage
             Lease-based                   –    Log-based, built on DHASH
             Read-only for clients         –    Snapshot and view-based
             Publishers can update              approach
             No explicit delete            –    User control over
    –   Built on Chord                           consistency/serialization
    –   Storage load-balancing
    –   Provably efficient and
        robust
    –   Built on DHASH xface
             File split into blocks
             k replication
Systems 8

   PeerDB                            PIER
    –   On path to a P2P DB            –   P2P DB
    –   No global schema                       Built on CAN and
                                                others
    –   Incomplete replication
                                       –   Relaxed consistency
    –   Dynamic reconfiguration
                                       –   Scalable with
    –   Requires small subset of
                                           namespace model
        persistent servers
                                       –   Std schemas
                                       –   Several join schemas
Evaluation Metrics

   Commit model (e.g., primary,            Reliability / robustness (i.e.,
    group, all)                              data that is eventually
   Information propagation model            available)
    (e.g., flood, epidemic, multicast)      Availabililty (i.e., data that is
   Topology                                 always available)
   Search model (e.g., targeted,           Quality of service
    flood, epidemic, multicast)             Anonymity/privacy
    - Expressiveness                       Censorship-resistance
    - Information                          Publisher/Server deniability
    placement/autonomy                      File integrity
   Scaleability                            File authenticity
   Target user?
Metrics from class
   Maintainability / Manageability            Trust model (physical vs virtual)
   Topology                                     –   Authentication
     –   Roles: client, supernode, server        –   Authorization
   Defense against selfish/ malicious           –   Admission control
    behaviors                                    –   Integrity
     –   Denial of svc resilience              Node heterogeneity
   Scope of knowledge                           –   Function, capabilities,
                                                     ownership, dynamic
   Needle vs Hay                                    election/configuration
     –   False negatives
                                               Indirection between obj lookup
   Static resilience vs MTTR                   and routing
   Performance under churn                    Application semantics used in
   Emergent behaviors                          routing
   Non-data services                          Data type / structured data
     –   GRID computing

				
DOCUMENT INFO
Categories:
Tags:
Stats:
views:0
posted:4/7/2013
language:Unknown
pages:14