Session 7: Towards A New Internet Routing Architecture Arguments for 
Separating Edges from Transit Core

The routing table size is growing at a rapid rate and will not stop 
growing.  The routing cost is becoming more expensive as the table 
size increases.  The main concern is whether or not the routers will 
be able to keep up with the growth.  Several practical solutions have 
been proposed to make routers more scalable.  Instead of comparing 
individual proposals, the authors categorized them into two groups: 
separation and elimination, and performed a comparison between them.  

The separation methods separate the edge networks and the Internet 
transit core.  The elimination methods require that edge networks take 
address assignments from their providers, which requires modification 
on the end-hosts to support multihoming.  The comparison shows that 
separation is in general a more suitable approach than elimination 
is.  The main reason is that separation methods treat the edge 
networks and transit core separately.  When the network needs a 
solution, separation methods allow quick deployment of the solution 
within each edge network, whereas the elimination methods might not be 
fast enough.

Discussion:  RRG is supposed to come up with recommendations for 
resolving the large routing table issue.  In case there is no good 
solution in place when the problem arises, separation methods might be 
helpful in keeping the routers running.  The speaker agreed that this 
might not be the ultimate solution.  The market and reality will bring 
up to the final solution.  Several other concerns are brought up 
during the discussion:

1.     Multipath is not supported by separation methods.

2.     Separation methods might introduce too much overhead than 
elimination methods do.

3.     It might not be a good idea to have the end-host or individual 
ISPs to make their own traffic engineering decisions.



Session 7: ViAggre: Making Routers Last Longer!

Routing table has been increasing significantly; and the trend will 
continue mainly due to IPv4 expansion and IPv6 deployment.   A router 
maintains a RIB and a FIB (commonly referred to as the routing 
table).  The growth will result in more processing time, more power 
and heat dissipation problems, and eventually more money to maintain a 
router.  ISPs are willing to undergo some pain to reduce FIB size in 
order to make their networks more cost-effective.  In effort to 
address this problem, ViAggre is proposed to shrink router FIBs.  In 
the ViAggre design, networks are configured into virtual networks, one 
for each virtual prefix.  The prefixes are made aggregatable, thus 
allowing for routing hierarchy that shrinks the routing table.  A 
router is required to maintain routes for prefixes in the virtual 
prefixes it is aggregating.  Since ViAggre is a configuration-only 
approach, it can be applied to any existing or future routers.  The 
authors showed that this design reduces stretch imposed on traffic as 
the FIB size increases.  ViAggre may cause an increase in traffic load 
across the ISP's routers and links.  However, the ISP can maintain 
routes to popular prefixes, that are carrying > 97% of the ISP's 
traffic, on all its routers to reduce both the load and the amount of 
traffic that gets stretched due to ViAggre.

Discussion:  It looks like ViAggre design is targeting at Tier-1 ISPs, 
and Tier-2 and Tier-3 ISPs might be benefited as well.  However, small 
ISPs might not benefit from it much.  Another concern is that whether 
ViAggre is still effective in the flat system imposed by P2P traffic.



Session 7: Good Things Come to Those Who (Can) Wait or How to Handle 
Delay Tolerant Traffic and Make Peace on the Internet

The main idea of this paper is to time shift the delay tolerant 
traffic away from peak hours to alleviate the wasteful usage of 
resources during the peak hours.  The ISPs peak hour load can be 
broken down into the interactive part and the delay tolerant (DT) 
part.  Due to the flat-rate pricing of residential broadband, users 
have no incentive to delay their non-interactive traffic to off-peak 
hours.  Any complicated pricing policy, or P2P traffic throttle will 
drive residential ISP subscribers away.  For this reason, the authors 
proposed a flat-rate-compatible incentive scheme.  On the end-hosts, 
the users who choose to consume small bandwidth during peak hours and 
will be rewarded with high bandwidth off the peak hours.  The ISP 
employs a storage enabled network consists of Internet Post Offices 
(IPO) and Transit Storage Nodes (TSN) to store delay tolerant data.  
The content will be forwarded to the destination during non-busy 
hours.  This method is referred to as the store-and-forward (SnF) 
transfer.  This study shows that an SnF transfer can utilize the TSN 
across different time zones since their peak hours are not 
overlapping, resulting in better performance than end-to-end routing.

Discussion:  Several concerns are brought up by the audience:

1.     The TSN and IPO can be the targets since they store many 
valuable data, especially personal information, in transmission.  
People have different views on this issue.  However, they all agree 
that this could be a serious problem.

2.     The storage requirement and cost on TSN and IPO could be a 
problem.  The speaker argued that the storage cost is decreasing at a 
rate exceeding the growth of network speed; hence, the storage won't 
be the main cost of the transmission.  However, depending on the 
application, more and more traffic will be stored on TSN or IPO.  
Would the backlog data eventually become obsolete?

3.     It is hard to determine the threshold bandwidth during and off 
the peak hours.

4.     If the delay tolerant data constitute majority of the peak-hour 
traffic, when time shifting their transmission, the peak-hour might 
shift as well.