[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Supporting large very large number of nodes
> I've run multicast simulations with 1680 nodes under the normal
> simulator with recent snapshots, and _some_ of them have worked,
> albeit inconsistently; running-out-of-memory problems, I suspect,
> since there are rather a lot of link objects too.
>
> For large simulations where you're not interested in queuing delay,
> you should apparently use SessionSim, which discards queueing
> information etc to save memory. (Contrary to nsDoc, first part of
> chapter 23, up to and including the latest release of 29 September,
> SessionSim does not discard propagation delay - so I'm looking at it
> with renewed interest.)
Lloyd, thanks for clearing it out. SessionSim does take into account
propagation delay and this bug in nsDoc has been fixed and will appear in
the next nsDoc release.
As for the simulation scale, the following table is recently updated and I
thought should be helpful to Arnis and other ns-users.
# Created by Polly Huang, USC/ISI
# http://www-scf.usc.edu/~bhuang
# 7/16/98
ns Scaling - The Story
The bottleneck has been determined to be the routing table, which is
O(n^2) for flat routing.
In addition, n (number of routing entries) is always 2^k. That is why
we observe roughly the same routing table size for 513-node and
1024-node simulations (k=10).
The following table lists _approximately_ the memory consumption of
session sim with various memory conservation solutions.
# nodes in topo Mem consumption (in MB)
detailed session session+hier session+algo
512-1023 189 16 9 (1000 node) 6 (1000 node)
1024-2047 656 46 22 (2000 node) 10 (2000 node)
2048-4095 2432 180 77 (4000 node) 18 (4000 node)
4096-8191 720 190 (6000 node) 33 (8000 node)
8192-16384 2886 1049 (10075 node) 82 (20000 node)
217 (50000 node)
cheers,
-Polly
>
> You'll need to use one of:
> # traditional, deprecated
> Node expandaddr
>
> # new fashionable equivalent object syntax
> $ns set-address-format expanded
>
> # gives 1024 nodes with patch to support one level of hierarchy;
> # (bit reserved for multicast) I haven't played with multiple levels.
> $ns set-address-format hierarchical 1 11
>
> after initialising the simulator to get a large address space to
> create the nodes in.
>
> hth,
>
> L.
>
> > On Thu, 15 Oct 1998, Arnis Ziedins wrote:
> >
> > > Does somebody have experience of running ns script with a very large number
> > > of nodes?
> > >
> > > How many nodes running TCP can be simulated with ns?
> > > For my experiment I would like to have more than 1000 nodes.
> > > Is it feasible with ns?
>
> <[email protected]>PGP<http://www.ee.surrey.ac.uk/Personal/L.Wood/>
>
>