Channel-bonding - choice of switch?

[Josip Loncaric]

Jeff Largent wrote:
> 
> Josip Loncaric wrote:
> >
> > Sounds reasonable, and if we did use channel bonding, that's what we'd
> > do.  This "separate switches" solution results in two identical but
> > physicaly separate LANs, which means that you also need a bridge machine
> > between channel-bonded nodes and normal machines.  The bridge machine
> > can be a bottleneck for applications that need high bandwidth to normal
> > machines (data servers, video displays).
> 
> Don't you have this anyway?
> lan--gateway--beowulf cluster
>        ^
>        bridge machine

Yes.  And the gateway can be a performance bottleneck.  Of course, the
best solution strongly depends on the application.

The alternative (no channel bonding) has lower bandwidth between cluster
nodes, but many switches provide a Gigabit Ethernet uplink port, which
simplifies access to large central servers within the organization.

Sincerely,
Josip

-- 
Dr. Josip Loncaric, Research Fellow               mailto:josip at icase.edu
ICASE, Mail Stop 132C           PGP key at http://www.icase.edu./~josip/
NASA Langley Research Center             mailto:j.loncaric at larc.nasa.gov
Hampton, VA 23681-2199, USA    Tel. +1 757 864-2192  Fax +1 757 864-6134