Thursday, November 13, 2008

Post-lecture: Middleboxes done right?

Last lecture on Thursday again ended in a very interesting discussion.

Randy commented on my last post: "I always like the format you use for these comments! Any thoughts on whether the idea of outsourcing middlebox functions makes any sense?"

Here is my answer:
Yes:
  • framework provides the same functionality without breaking the two principles mentioned in the introduction
  • performance will be competitive only if the middlebox will be close or on the network path, i.e. where the middlebox is anyway
  • user benefits from being able to choose between the middlebox service and direct service
No:
  • extra conceptual overhead
  • how to gradually deploy?
  • because middlebox is not phyisically on the network path blocking IP packets, certain attacks on computers are still possible
  • user needs to configure their computer to use middlebox service (half will forget and leave the network vulnerable)
  • middleboxes might be deployed to actually prevent certain kinds of services, with middlebox being outsourced this can be circumvented
  • NAT on traditional boxes offer anonymity, invisibility from the outside

No comments: