An Active Service Framework and its Application to Real-time Multimedia Transcoding
Elan Amir, Steven McCanne, and Randy Katz (Berkeley), 1998
Summary. In an effort to remain compatible with the present Internet
architecture, the authors propose an alternative to active networks for
the subset of applications that provide an application level service called
Active Services. They "inherit" all the programmability of active networks
but do not require drastic changes in the Internet architecture.
More Detail
The authors suggest that active networks will fix many of their target
application challenges, but that they are not necessary. IOW, it is not
necessary to [uproot and supplant] "over 20 years of Internet and design
experience". There are a number of applications which can be effectively
addressed by an the Active Service model they propose which features
Programmability, but restricted to the application level . They
detail the design of an active service framework which can be
specialized to suit each application's needs.
Model Overview. Parts:
- Clients.
- Servents (Service Agents) - an instance of the requested active service.
- Host Managers - daemon that receives client requests.
Key Concepts.
- Push for resource discovery. Use a well-known multicast address.
(Have other mechanisms for unicast client.)
- Soft-state.
- Announce-Listen protocol.
- timer-based aging of state
- periodic refresh
- error detection, recovery and (eventual) consistency
- Host Manager <-> servent launching
- Host Manager launching and fault tolerance
- Multicast damping.
- Used to avoid "servent flood"
- Gives you load mgt for free by adjusting the timer interval
according to host manager's load.
- Gives you simple admission control (timer interval of infinity
on when host at full capacity).
- Soft-state gateways.
- There exists two potential problems with Announce-Listen:
- Control traffic won't scale as the number of clients
increases.
- Multicast impoverished clients can't use it.
- SSGs address these problems.
- Filter the traffic.
- Serves as a rendezvous point for clients without multicast.
- SCUBA. Not covered in detail in this paper, but
neat software.
Potential Issues.
- Security. The authors admit that this is a significant
issue that they have chosen not to include in the scope of their work. What
if you have AS clusters in different administrative domains? Can you
use out-of-band security to issue certs to code servers?
- The SSG does not address the possible scalability problem of
Announce-Listen table size. Works for their application domain, but others?
- Choosing correct timer intervals for multicast damping in non-cluster
environment hard.