Re: nfs kernel booting is slow

[mlelstv%serpens.de@localhost (Michael van Elst)]

scole_mail%gmx.com@localhost (scole_mail) writes:

>The rsize/wsize options didn't seem to make a difference either way.

Since kernel loading is the issue and not the kernel accessing the
root volume you have to look at the macppc bootloader. The bootloader
uses the NFS client in libsa which does UDP only and is already limited
to rsize=1024.

What you see in the bad trace is a NFS request that is quickly answered
by the NFS server, and then a pause of about two seconds. Then the same
NFS request is retried, answered as quickly but then followed by the next
request.

This means that the first response isn't seen by the client.

One possible reason for this is a duplex mismatch between client and switch.
The bootloader doesn't configure the interface but relies on OpenFirmware
to do this and handle autonegotiation correctly. Maybe there are properties
in OpenFirmware where you can configure the interface.

Another possibility is that the Gigabit switch doesn't buffer packets, then
it's possible that packets from 1Gbps to 10Mbps port are dropped when they
arrive too fast. For a test you could either try to throttle the NFS server
to 100Mbps or maybe the switch supports 802.3x style flow control and you
can enable this on the interface of the NFS server.

-- 
-- 
                                Michael van Elst
Internet: mlelstv%serpens.de@localhost
                                "A potential Snark may lurk in every tree."