Day 1: Tutorials: Network Design and Implementation
Man, getting up at 04:00 (Pacific time) after not being able to sleep before
a few hours earlier... It's only the first day and I'm already exhausted...
When's Red Hat moving to the Silicon Valley anyway?
The tutorial was given by Alex O. Yuriev from Netaxs, LLC (he designs networks
for a living).
I had originally meant to attend the Qmail tutorial by Russell Nelson but after
talking with Alex at the Ki Networks
welcome reception, I decided to attend his talk because I remembered having
missed a one hour talk from him last year and the slides he was flying through
at the end while he was trying to wrap up looked really interesting and he
definitely looked like a very knowledgeable speaker.
I guess I'll be better off reading Qmail's documentation
The talk was very informative, unfortunately Alex didn't get to finish it in
the morning session and while he was going to finish it in at the beginning of
his Managing Security Threats session, I just couldn't miss the
Kernel Programming session that afternoon.
Here are a few interesting facts he mentionned:
- 28G/day for a full news feed nowadays.
- Some telcos will take a Sonet ring and use the backup ring for additional
capacity so that when another telco comes in and cuts part of the ring, the
whole network goes down since the redundancy is gone.
- Gigabit ethernet isn't really proven yet and doesn't seem to yet handle
much more than a few hundred megabits. In a few months, things should get a
- The uplink and downlink ports are really identical on a switch, it's really
a labelling issue.
- Some catalist switches can get into a failure state where
they'll going in a spanning tree loop and you have to turn off all the
switches to clear the error (reported by Alan Cox)
- Flat networks are really bad because they can be used to create Smurf
attacks. But then, blocking ICMP on your router is not the way to go either
because it breaks path MTU discovery.
- Alex doesn't see a switch to IPV6 ever happening because it'd be way to
much work to do the switchover. It's true that Classless IP lets us save
IPs and reclaiming wasted ranges will also help, but I'm not sure how
far we're going to go. I guess we'll see.
- ATM clouds claim to have a one hop connection from anywhere to anywhere but
in real live, it's really a bunch of hops on layer 2, even though it looks
like one hop on layer 3
- Real routers should drop a static route linked to an interface if the
interface loses the physical link. Linux unfortunately doesn't withdraw
a static route temporarly if an ethernet link disappears (even though it's
definitely technically possible). Real routers do.
This is a very valid point. Let's hope this gets into linux in the future.
- Alex obviously recommended against using RIP v1, since it doesn't understand
classless routing. OSPF is kind of nice for a moderate size network if your
network has a backbone, but otherwise it's not the best choice either. ISIS
is similar to OSPF, but is supposed to work better. However only ciscos and
gated seem to support this protocol.
While I did like his talk, I would have prefered that he went over the basic
stuff much quicker (if at all for some things), and spent more time on the meaty
stuff, like BGP. Regardless, if you get a chance to attend one of his talks,
I recommend you check out the Picture
Library if you want to see the slides.
Link to Home Page
99/05/19 (13:31): Version 1.0