RE: Do we need a reliable multicast in kernel?

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tue, 21 Oct 2003, Zhu, Yi wrote:
> On Mon, 21 Oct 2003, Donald Becker wrote:
> > But none of these uses indicates that we
> > need reliable multicast implemented in the kernel.
> 
> Reliable multicast transport layer is useful in the intranet.
> If kernel can provide this, a lot of cluster softwares can
> benefit from it.

I really doubt that.  A few years ago, with mostly repeater-based
networks, multicast did marginally make sense.  But Ethernet switches
drop multicast traffic in different, unpredictable ways.  And cascading
two switches compounds the problem characterizing the loss pattern.
End-point: high-bandwidth Ethernet multicast is not a deployable technology.
You end up developing and tuning algorithms that work on your specific
machine, but don't work with other hardware or scale to a different
size/topology.

Multicast does have a much better chance of working if the application
carefully manages the group size and restricts the bandwidth
i.e. implements significant policy intertwined with the mechanism.
With low bandwidth and complicated policy, there is no performance or
structural reason to put multicast into the kernel vs. leaving it as a
library implementation.  And with a library implementation, it's easier
to turn off when it doesn't work for the majority of end users!

If you are making general-purpose cluster software, the communication
design must also work on Myrinet, SCI, Giganet, and Infiniband.  None of
those have hardware-based multicast.  They all emulate it in software or
switch firmware.  (Most Gigabit Ethernet switches emulate multicast in
firmware.)  I don't see any change that motivates new development of
physical-layer-oriented multicast protocols for cluster use.

-- 
Donald Becker				becker@scyld.com
Scyld Computing Corporation		http://www.scyld.com
914 Bay Ridge Road, Suite 220		Scyld Beowulf cluster system
Annapolis MD 21403			410-990-9993

-
: send the line "unsubscribe linux-net" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[Index of Archives]     [Netdev]     [Ethernet Bridging]     [Linux 802.1Q VLAN]     [Linux Wireless]     [Kernel Newbies]     [Security]     [Linux for Hams]     [Netfilter]     [Git]     [Bugtraq]     [Yosemite News and Information]     [MIPS Linux]     [ARM Linux]     [Linux RAID]     [Linux PCI]     [Linux Admin]     [Samba]

  Powered by Linux