[Xorp-users] xorp_measures

[Atanu Ghosh]

Hi,

I don't understand your experiment but if the graphs that are being
tested are fully connected then for every additional vertex {vertex - 1}
edges will be added (in the 5000 vertex case each vertex will have 4999
edges), will any algorithm that has to traverse all edges perform
linearly as the vertexes are increased?

The SPT code in XORP is in a library and independant of OSFP the
interface to the library was designed to allow the implementation to use
an incremental algorithm. After we measured the performance of the SPT
code we didn't feel that it was necessary to add the incremental support
yet, for you experiment am incremental implementation would probably
have run slower as every vertex is modified.

I don't understand your questions about the code. It seems unlikely that
the choice of data structures in OSPF will make much difference to the
SPT computation time which is in a separate library. For very LSA there
is a single reference counted copy, to simplify management.

  Atanu.

>>>>> "ORANTE" == ORANTE TUCCERI <orante2003 at yahoo.it> writes:

    ORANTE>    hello to all, I am executing of the measures on the
    ORANTE> software router xorp.  The measures have the scope to
    ORANTE> estimate the performances of the router in terms of time
    ORANTE> demanded for the execution of the calculation of the
    ORANTE> shortest path tree.  From the measures conduct on topologies
    ORANTE> of net (emulated through PC) fully connected constituted
    ORANTE> from a number of router R 10: 10: 100 (min /step /max) with
    ORANTE> interposed of the transit network between router and an
    ORANTE> other, is possible to obtain the following course (Time to
    ORANTE> calculate the SPT in milli-seconds vs.Vertex N)

    ORANTE>    N= R*(R-1)*0.5 Xorp -1.2 vs. Quagga 0.98 not optimized
    ORANTE> see the attached graph

    ORANTE>    We think that the evidenced not linear behavior from the
    ORANTE> diagram is legacy to the choice of the given structures
    ORANTE> associates, second given definition into RFC 2328, to the
    ORANTE> list of the candidates, the list of the next-hop and to the
    ORANTE> database of the LSA.  I ask confirmation for following
    ORANTE> associations XORP/RFC2328 -----xorp/libproto/spt.hh @726 _
    ORANTE> tentative/list of the candidates
    ORANTE> -----xorp/libproto/area_router.hh @586 vector _ db//
    ORANTE> Database of LSAs Other aspect that me is not clear is this:
    ORANTE> If the association _ db/Database of the LSARef is corrected,
    ORANTE> · since LsaRef is a alias of ref_ptr · since the class
    ORANTE> ref_ptr (present in libxorp/ref_ptr.hh) is created with the
    ORANTE> scope to maintain the count of the number of how many
    ORANTE> references exists to an object me it is not clearly because
    ORANTE> the memory of the database of the LSA is implemented through
    ORANTE> carrier of type objects '' ref_ptr ' '.  tanks for
    ORANTE> attention.  < orante2003 at yahoo.it >

    ORANTE>    Chiacchiera con i tuoi amici in tempo reale!
    ORANTE> http://it.yahoo.com/mail_it/foot/*http://it.messenger.yahoo.com
    ORANTE> _______________________________________________ Xorp-users
    ORANTE> mailing list Xorp-users at xorp.org
    ORANTE> http://mailman.ICSI.Berkeley.EDU/mailman/listinfo/xorp-users