[clug] Feedback...

[Sam Couter]

steve jenkin <sjenkin at canb.auug.org.au> wrote:
> PS: There is a whole class of network problem which I've never heard a
> good name for ("cascade fault"?) like Martin's - some 'error recovery'
> mechanism kicks in and compounds the fault.
> Often causing accelerating damage and turning a recoverable/borderline
> situation into a guaranteed (massive) failure.

Reminds me of a time in the dim past when I used to tunnel PPP
connections over SSH. Bad idea. Works fine while the link is
underutilised, but as soon as packets start dropping there are two TCP
stacks trying to do error correction.

http://sites.inka.de/~W1011/devel/tcp-tcp.html

The end result of this is that you need an error-correcting, reliable
delivery protocol at exactly *one* place in the stack. If you're doing
stuff with TCP, such as fetching .debs via HTTP, just trust TCP to get
the job done right. If TCP is busted, you have bigger problems, and it's
not something you can easily work around anyway.
-- 
Sam Couter         |  mailto:sam at couter.id.au
OpenPGP fingerprint:  A46B 9BB5 3148 7BEA 1F05  5BD5 8530 03AE DE89 C75C
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://lists.samba.org/pipermail/linux/attachments/20100121/254e886e/attachment.pgp>