• DocumentCode
    3441847
  • Title

    Exploiting the Thread-Level Parallelism for BGP on Multi-core

  • Author

    Lei, Gao ; Mingche, Lai ; Zhenghu, Gong

  • Author_Institution
    Sch. of Comput., Nat. Univ. of Defense Technol., Changsha
  • fYear
    2008
  • fDate
    5-8 May 2008
  • Firstpage
    510
  • Lastpage
    516
  • Abstract
    With the developing technology of multi-core processors, exploiting the thread-level parallelism (TLP) of BGP will highly improve the performance required by core routers on Internet, to relax the problems of BGP route outburst updates and persistent route oscillations. The TLP approach for BGP based on min-cut partition algorithm which introduces thread-level speculation (TLS) and decomposes the BGP program into granularity balanced threads is proposed in this paper. Meanwhile, the speculation strategies and relative implementations, which ensure the commits of speculative threads in sequence and maintain the memory consistency, are presented for TLS of BGP to improve the parallelism. Experiments on sun fire T1000 server show that our algorithm can achieve pretty speedup, which can reach 1.51, 1.82, 2.59 and 2.92 when partitioning 32, 64, 256 and 512 threads and allowing 4, 8, 12 and 16 threads in parallel. And the response time of processing 1 K, 5 K, 10 K and 30 K packets on single session has decreased 16.9% and 25.2% averagely when partitioning 64 and 256 threads.
  • Keywords
    Internet; internetworking; parallel programming; routing protocols; BGP; border gateway protocol; multicore; thread-level parallelism; thread-level speculation; Computer networks; Delay; Fires; Multicore processing; Parallel processing; Partitioning algorithms; Routing protocols; Sun; TCPIP; Yarn; BGP; multi-core; parallelism;
  • fLanguage
    English
  • Publisher
    ieee
  • Conference_Titel
    Communication Networks and Services Research Conference, 2008. CNSR 2008. 6th Annual
  • Conference_Location
    Halifax, NS
  • Print_ISBN
    978-0-7695-3135-9
  • Type

    conf

  • DOI
    10.1109/CNSR.2008.22
  • Filename
    4519901