All things related to multicast

[Michael Blodgett <>]

We use frisbeed from the University of Utah,

http://www.cs.utah.edu/flux/papers/frisbee-usenix03/

which has a preset bandwidth cap set at the server, since we
have a number of 100baseT connected clients, it is set at 70mbit.
So comparing one machine unicast to multicast we would see a difference.

But for your multi hour lab imaging runs, do you know how a Zen client 
can request say a single missing block?  Do you have some machines that
actually complete in the 75 minute range but to get the last one 
completed is multiple hours?

Mike Blodgett
Wisconsin Advanced Internet Laboratory
University of Wisconsin - Madison

Ying Zhang wrote:
> Thanks for the reply. We'll double check that. When we set up a test 
> environment with three clients, they all run at 1-gig full-duplex. It 
> took 30 minutes to image all three with unicast but 75 minutes with 
> multicast.   
>  
> Regards.
> Ying
>
>     ----- Original Message -----
>     *From:* Yongrui Tang 
> <mailto:>
>     *To:* Ying Zhang 
> <mailto:>
>     *Sent:* Tuesday, May 06, 2008 10:42 AM
>     *Subject:* RE: Lab imaging using multicast
>
>     Most likely it is due to duplex mismatch.
>
>      
>
>     Yongrui Tang
>
>      
>
>     ------------------------------------------------------------------------
>
>     *From:* Ying Zhang 
> [mailto:]
>     *Sent:* Tuesday, May 06, 2008 8:18 AM
>     *To:* wg-multicast
>     *Subject:* Lab imaging using multicast
>
>      
>
>     Hello all,
>
>      
>
>     In our university, we using Novell ZENworks preboot services to
>     image thousands of lab computers. We found if use unicast to do
>     imaging, it takes about 30 minutes for each client, while using
>     multicast takes hours to image a lab with 20+ machines. And the time
>     increases proportionally with the number of machines. What is
>     strange about this is it takes a lot longer to do multicast, no
>     matter how many clients are being imaged.
>
>      
>
>     The server and clients are on the same VLAN. So no multicast routing
>     is involved. IGMP is enabled on the switch. It uses group address
>     231.164.242.170. Just wondering if any other universities have
>     similar experience and what could be done on the network side to
>     improve the performance. Any help is appreciated.
>
>      
>
>     Ying
>
>     UNB