Re: [9fans] sources/contrib

[anyrhine@gmail.com]

[-- Attachment #1: Type: text/plain, Size: 638 bytes --]

the latency problem does not show up before you
have actual propagation delay. if you're next
door with a 34k modem, all your latency is bare
transmission time (time spent stuffing bits along
the wire), and you are prefectly happy because
your link gets full utilization.

now, if you are physically 200ms away and have a 
the fastest link on the planet, you might be less
than happy using 1 second [Twalk Topen, Tread, Tread,
Tclunk] to read a file of any size, because your
fast and fancy link isn't used, so you're paying for
nothing.

the problem really does exist, but there are bigger
ones that are easier to solve.

[-- Attachment #2: Type: message/rfc822, Size: 3515 bytes --]

From: erik quanstrom <quanstro@coraid.com>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] sources/contrib
Date: Tue, 20 Nov 2007 16:46:22 -0500
Message-ID: <333c3d5c4dfff180853969bcfdf2b867@coraid.com>

> Read 9fans archives, replica/pull has wiped out more than a handful of systems,
> if you run venti you can recover, but it is not fun.

i think a few well-choosen hurestics could solve most of these cases.

> And as erik points out, replica also has trouble with many corner cases.
> 
> And finally, it is incredibly slow, but that is probably mostly due to
> 9p's latency sensitivity.

i just don't see the latency problem.  even when i was using a 34k modem,
updates were pretty quick, 

i have seen the case where applychanges was very slow.  but using cphist
instead solved the problem.  i can't explain why applychanges can be very
slow for me.

- erik