Review - Distributed Management in the Tiger Video Fileserver

Video distribution over the network is complex due to quality of service
issues. This is typically time sensitive data where systems have real-time
requirements for serving content. The Tiger fileserver spreads an object
across hosts on the local network in an attempt to achieve some level of
load balancing. A centralized Tiger server manages schedules in order to
maintain file consistency. This implementation also uses mirroring
(replication) for availability.

The term hotspot is used by Bolosky et al. to describe when a cub is asked
to do more work then it can handle. Schedules are used to avoid these cases.
A scheduler describes the needed work to supply data to viewers and also
checks if starting a new viewer will create a hotspot. An interesting
concept presented in this paper is 'coherent hallucination.' Each cub
maintains a partial knowledge of the global schedule but lazily behaves as
if it knows the entire schedule. However, for fault tolerance, Tiger must
ensure that partial schedule information is not lost (replication).

This is an interesting classical paper. It seems that many of today's
content distribution questions were being looked at in the mid-90s. We can
see that they were struggling with scalability issues, replications issues,
bandwidth issues, etc. However, it is unclear weather or not this system
would be useful today. Bandwidth and disk capacity, as well as processing
power, have improved significantly and perhaps makes this technique
obsolete.
Received on Thu Dec 01 2005 - 01:11:13 EST