🚀 go-pugleaf

In article <1992Jan7.072754.10321@cbfsb.att.com> eajjs@cbnewsf.cb.att.com (john.
sottile) writes:
>We tuned NFS via the /etc/fstab options to use a retrans=300,
>rsize=4096,wsize=4096.
>Retransmission  = 300 tenths of a second (30 Seconds).  This is a bit

------------
Don't you mean 'timeo=300' ?
I believe default timeo=7 (7/10 sec) & default retrans=3 (3 retransmissions).

We have people mounting over 56k, T1, Ethernet, & faster. We have no control
or warning as to what who is going to do when, and to whom.  ;-)

I read something somewhere (I thought it was a Sun man page on 'mount') &
tried what they said in a lab with various conditions. I found that :

'rsize=1024 wsize=1024 timeo=30 (3 seconds) retrans=5'

worked fine - we suffered little performance loss in good (eg, clean ethernet)
situations while still allowing things to work in bad (marginal 56k links)
situations. I recommended this to everyone in my company that I could get to,
& some even listened to me. We've had very few calls since.

I am not an expert at this, & this was by no means a 'good scientific study'.
But it was all I had time for, & it sure seemed to work.

ps: I sure appreciate that recommendation of the O'Reilly book, we're going
    to order one.

pps: It looked to me that rsize & wsize < 1256 (or there abouts) would allow
    enough room for ip/udp/nfs headers & still stay within ONE ethernet packet
    (MTU). 1024 was a NICE figure that was < 1256. The default 8k can result
    in a fast system rapid-firing out 6 ethernet packets, which in a non-perfect
    environment, may NOT all get through. IF one or more packets don't get
    ALL the way to the other process, then ALL 6 packets get rapid-fired again,
    & the cycle MAY repeat (until RPC timeout error).
    At least it looked that way to me. Again, I'm just a network grunt, so what
    do I know ? Don't answer that !! ;-)