🚀 go-pugleaf

In article <1992Sep10.152047.641@nntpd.lkg.dec.com>, koning@koning.enet.dec.com (Paul Koning) writes:
> 
> Unfortunately, there are a number of similar-sounding terms around to 
> describe very different functions.  You were caught by one of them.
> 
> What the Motorola chipset can do is "ring purging".  This function -- usually
> enabled in Digital products -- indeed removes fragments left by the Strip
> function.  Its main purpose, though, is to remove "no owner" frames, which
> are things that can cause real trouble.  You can find more information about
> the ring purger in the following paper:
> 
> 	Proceedings of 16th Conf on LCN, Oct 14-17, 1991, 
> 	"A Ring Purger for the FDDI Token Ring", by Henry Yang and KK
> 	Ramakrishnan, page 503-514.
> 
> Fragments are harmless things, since they consume no ring bandwidth....


Do you agree that no owner frames also consume no bandwidth?  That they
are only harmful if they have broadcast or multicast DA's or if you are
using MAC's in promiscuous mode?  (e.g. in bridges and ring analayzers;
DEC makes the former, SGI the latter)


Has anyone seen throughput problems when purging is happening?
As in reducing TCP from 6MByte/sec to 2 MB/s?  I have one such report,
but have not been able to reproduce it, nor to figure out a way in
which it might happen.  I have seen 2-6% drops from 6MB/s by turning on
a purger on a ring with 50 usec latency and sub-millisecond average
token latency (even at 6.5MB/s), which I guess sounds benign and
reasonable, given that any jostling in timing hurts throughput, at
least for me.  I just can't see how the purger might reduce performance
by 60%.


Vernon Schryver,  vjs@sgi.com