16.7

It effectively divides the namespace, and thus the load
on any one or set of servers.  It also allows local
control of local domains (presumably the folks
running the cs.washington.edu server know the most 
about the machines in *.cs.washington.edu).



16.9

Having it lower in the stack means you can use lower-level
information to possibly improve performance (such as doing
link-level retransmission), but it means that services
higher up that don't need this functionality pay for it
anyway.



16.15

The advantage is that everything works the same as it
did before, and users don't have to consider the fact that
the network exists and worry about maintaining it.  The
disadvantage is that if something goes wrong, users aren't
aware of why it's going wrong and can't make intelligent
decisions.



17.3

a) NFS is location transparent in the sense that the NFS
mount can be anywhere in the file system, and the user has
no notion of what server is backing that NFS mount.  AFS 
is also location transparent, since users have no notion of
servers at all, only of a shared namespace of files

b) NFS is not location independent.  Since each NFS mount is
at a different location in the file system, if a file moves
to a different server, it will be in a different part of the
file system and thus its name has changed.  AFS is location
independent because clients do not know which server the file
is on until they ask for it, and they ask for it by a consistent
name within the shared namespace



17.7

Caching locally means that a hit in the cache avoids the
network delay, and thus is faster.  However, since this
cache will only be for a limited user population (those
users on the client machine), the hit rate will be much 
lower.  In contrast, caching on the server means a higher
hit rate, but with a larger cost for hits.  A second issue
is cache coherence: if we cache on the client side, we have
to make sure these cache entries are invalidated when the
info changes.  This is easy to do on the server side since
it's the server that knows when the entries change



17.8

a) Clients are clustered within a local LAN with a representative
server, which keeps performance acceptable despite a growing
network, and reduces network load.

b) There is extensive use of caching at the client side, so that
most operations can be performed locally with no server involvement

c) Clients maintain the local namespace for files that can't or don't
need to be managed in the shared namespace, further reducing the load
on the servers



15.1

Better programming methodology, such as using functions that are not
susceptible to buffer overflows and more stringent code review, 
reduces the number of buffer overflows that exist in a given program.
Hardware methods include hardware support to note and prevent out-of-
bounds memory access or protect key data structures, like the return
address on the stack, reducing the possibility that an exploit
of a buffer overflow bug will lead to a compromise.  As always,
the combination of these two methods provides the best defense.



15.7

If they had correctly written a set of firewall rules that only allowed
out expected traffic (from explicitly installed services, say), then even
if a worm infected a system, it could not spread to others via the system,
limiting its effectiveness.  This limits the flexibility of users within
the system.  If they run an unexpected service, even if benign it will 
prevented from running until the firewall rules are changed.