home *** CD-ROM | disk | FTP | other *** search
-
- On Sat, 3 Jul 1993, John Gardiner Myers wrote:
-
- > If the advice to clients is reasonably interpreted as "ignore the
- > /SEEN flag for bboards and handle the per-user state yourself", then I
- > fear clients will store this information in the simplest and most
- > logical place (the local disk) and be done with it.
-
- Again, in many cases there won't be any per-user /SEEN flags stored on the
- archive server. Whether or not client-writers choose the easy way out and
- store state locally depends largely on how easy we make it to access
- remote state files. But I'll also mention a scenario below where having
- the state stored locally on the client is the *right* --indeed, only--
- thing to do.
-
- > Any distributed mechanism available to the client for handling this
- > information is at least equally available to the server.
-
- In principle, Yes. In practice, Not necessarily. I see a significant
- difference in operational cost when a (non-Kerberos) archive server
- suddenly needs to traffic in user-specific data.
-
- > As there are
- > usually fewer varieties of servers than clients, and as servers
- > usually don't run on low-resource machines, I believe it is much
- > simpler to get all servers to handle this correctly than to get all
- > clients to do so.
-
- We have to solve the same problem for addressbooks and existing newsrc
- files, so I don't believe there is much increase in complexity or resource
- requirements for the client. Moreover, as time goes on, shared server
- resources will become much more scarce than dedicated client resources.
- Besides, the which-Unix-flavor server wars will go on forever, but in a
- couple of years all clients will be running Win NT, won't they? :) :)
-
- Having the per-user state info manipulated on the server *requires* that
- the server have some cognizance of "user". Within a Kerberos realm, this
- may not be a big problem, but in many other situations it will make the
- difference between whether or not the per-user services are available at
- all.
-
- Consider a small K12 school that may be running their own mail server, but
- is relying on another site for access to news and public archive data. If
- per-user state data can *only* be manipulated on the server, a teacher
- who's entire data world lives on their PC's hard disk is powerless to take
- advantage of these per-user services. (I'm assuming a realistic view of
- how easy it is to get guest accounts set up at some other site as K12
- demand scales!) In contrast, if the per-user data *may* be manipulated by
- the client, then we can accommodate sites where central *support*
- resources are slim-to-none as well as sites with relatively rich computing
- infrastructures (both systems & staff). NNTP is clearly a successful
- precedent for this model.
-
- I also want to see IMAPd code surface in the bazillion anonymous ftp
- servers around the world and suddenly have their mail archives available
- via IMAP. These systems don't know about specific users in general, and
- 99.99% will never know about me in particular. But I still want to be
- able to keep track of what messages I've seen/pseudo-deleted/answerd/etc.
-
- IF the IMAPd is the only one allowed to manipulate per user state info, I
- now have several additional dependencies when compared to client-side state
- manipulation:
- -My MUA must know how to authenticate itself to an IMAPd in any realm or
- location, even for *public* data access,
- -The IMAPd must know how and where to get my per-user state,
- -The IMAPd must be able to authenticate me and/or it to the state server,
- -and There must be a "state" server ready and willing to accommodate that
- request.
-
- Within a particular site, having the server do the per-user state
- processing may be perfectly reasonable, though I'm not yet convinced it is
- preferable. I am convinced that it is the wrong answer for the broader
- context cited above.
-
- -teg
-
-
-
-