[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Urgent!
> X-Authentication-Warning: mash.CS.Berkeley.EDU: majrdomo set sender to
[email protected] using -f
> From: "Huiwen Li" <[email protected]>
> To: "'[email protected]'" <[email protected]>
> Subject: Urgent!
> Date: Wed, 2 Feb 2000 11:37:59 -0800
> MIME-Version: 1.0
>
> Hi,
>
> I have a question about monitoring ifq_ at a LAN. I modified the code in
> vlan.tcl so that it can monitor the ifq_, when I use 802_3 MAC at 100 source
> nodes.
>
> The topology is as follows. Each source node will form a LAN with node 0,
> while node0 is using DRR to schedule the packets to node1. The traffic
> source are 50 UDP one-way flows from snode(i) to dnode(i) (0<=i<50)and 50
> TCP one-way flows from snode(i) to dnode(i) (50<=i<100).
>
> snode(0) dnode(0)
> \ /
> snode(1) --- node 0 ------node1---- node2 ----- dnode(1)
> / \
> snode(99) dnode(99)
>
>
> I could see the throughput decrease for both UDP and TCP dramatically, but I
> could see no drops in ifq_'s! This is extremely strange for UDP flows,
> because there will be no ACK's for UDP flows, which means no contention at
> all for UDP source. I am wondering if there are other places that LANs could
Yes! At the Mac Layer, LL layer... See the code for LanIface instproc init and
trace to see what is usually traced. What is the ifq's max length? Maybe you
would want to monitor it's length to see how full it gets.
> possibly drop packets except ifq_? I also change the parameters for 802_3 to
> fit for 100Mbps Ethernet.
>
> Please give me some suggestions. Thanks in advance!
>
> Huiwen
Tarik