Download Offense - Northwestern Networks Group

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Network science wikipedia , lookup

Cracking of wireless networks wikipedia , lookup

Transcript
Spam???
Presented by
Shiva Srivastava
Ionut Trestian

There are research papers

There are Twitter papers
You can decrease RIB and FIB only by router
configuration changes
65 characters !!
You can decrease RIB and FIB only by router
configuration changes but packets take longer paths
96 characters !!




No need to change routers
That means that we use the same routers at
lower capacity
But if packets travel longer don’t we consume
more power with the extra forwarding?
Which is fine, but …




Biggest fragmentation one can have given by:
255x255x255 = about 16 million prefixes.
Actually lower than that (close to a few millions)
for various reasons
Paper says that FIB carries now about 1 million
such prefixes
To me the problem seems completely solved
one upgrade away

Unpopular prefixes used to get the short end
of the stick in current deployments (big
delays, jitter etc)

They seem to get the short end of the stick in
ViAggre too because even ViAggre seems to
treat popular prefixes better

Authors assume that no architecture changes will occur
but if one deploys ViAggre wouldn’t the ISPs tend to use
older/slower routers inside their network leading to
worse performance?

Maybe the performance benefits we are seeing is
because you deploy ViAggre over newer/faster routers

It seems to me that now you start designing your
network according to ViAggre so it induces architecture
changes !!

Everybody lets just use IBM 5150, you know
we can just upgrade the memory, it will be
great!
1)
2)
3)
Only a short term solution, that is going to
prevent from upgrading..?!
Effectively reducing routers by assigning
them as aggregate routers.
Design flaw: They try to send all the traffic
through one router: Congestion

They cannot attain the full potential of their
own ViAggre system as their design goals
restrict them

They only try to reduce the size of the routing
tables, which is not even that effective.

Why should we use a half solution to a
problem that does not even work to its full
potential.




Unnecessary complication by advertising!
And again waste of bandwidth.
Can cause looping of packets, use tunnels to
prevent them but yet again unnecessary
complication!
Routers have to maintain separate LSP
mappings
Figure 1 just shows how its making things
more complex unnecessarily.

This system makes the network more prone
to crashing!

If any aggregation point fails all the network
is transferred to second closest point…
AGAIN CONGESTION !



ISPs should configure so they can use
ViAggre!----- Why would you put energy in
making a short term fix.
The ViAggre system wont work on popular
prefixes! As they have to maintain whole table
For the rest of the network they need the ISP’s
to be smart in designating routers to
aggregate virtual prefixes! Then what good are
they for?

This method increases path length and
causes unnecessary traffic!

Greedy algorithm is too greedy and not
efficient and can cause the network to crash!




They had no information about the link
weights, they just assumed it to be distance
weighted!
Create too much complexity in the network!
They are basically stalling the problem and
not fixing it.
They would have to upgrade anyways
because of newer technologies and higher
data rate.

They are not sure about popular prefix ISP
traffic which renders ViAggre useless

Figure six shows how they are actually
affecting the network 100%!

No traffic matrices hence cannot analyze load
increase across routers. – Isn’t that’s what
they are studying?

Cannot use it, the ones who could have
actually used it.

Maybe they can reduce FIB size on them but
it isn't worth the trouble at this point.




Test system is way too small, no realization of
real world.
Only ‘Configuration approach’ – Only few
popular prefix add too many configuration
lines.
Their test tool is too specific to their system!
Design too maybe good – Too short term!





Their Goal: Reduce Monetary cost.
Their Achievement: Increase Management
Overhead cost and effectively doing nothing.
Add more slow fat routers---- We thought
they were trying to use the old ones!
Basically remodel the whole network for
stupid ViAggree
Finally they try to tell ISP’s that they should
do something!