Those of you with long memories may remember discussing groups of groups in July last year, when I asked Transarc to consider the problem. The answers I got from most people seemed to try to tell me that while GOGs might be nice from the management point of view, they would slow the pt server unacceptably. This has been repeated recently I don't agree. GOGs reduce system administration and should make the protection database SMALLER and (if anything) FASTER. It's a basic characteristic of efficient search algorithms that they arrange items in trees so that search times depend on the log of the number of elements. Please don't tell me that arranging my AFS users in a sensible tree structure will make it significantly slower to use. OK, so it's possible to have a build pathologically bad group structure which would be slow to search, but I'm not going to do that, am I? There's no point. Similarly, it would be possible for Transarc to implement the above inefficiently, but I hope they wouldn't do that either. Furthermore, the vast majority of our users are students whose activity is almost solely within a particular group, department or school. In terms of applying ACLs, they rarely exist as individuals. It makes no sense to add individual users repeatedly to different groups for (say) access to a particular package when they will only ever be managed in bulk as part of their department or project group. Eliminating this repetition makes a smaller database. Yes, GOGs will probably require more work as each entry needs more consistency checking as it is added to the db. BUT since the total number of entrys in large groups will fall drastically, so will the number of db additions. For large groups, this will drop from thousands to dozens, so I'm quite happy with this tradeoff. Anyone care to prove me wrong? Please, Transarc, we really need this, and we can't hack AFS source like UMich. Peter Lister [EMAIL PROTECTED] Computer Centre, Cranfield Institute of Technology, Voice: +44 234 754200 ext 2828 Cranfield, Bedfordshire MK43 0AL England Fax: +44 234 750875
Groups of groups (and efficiency thereof)
Peter Lister, Cranfield Computer Centre Tue, 4 May 93 07:45:55 -0400
- Re: Groups of groups (and efficien... Peter Lister, Cranfield Computer Centre
- Re: Groups of groups (and eff... Pierette_Maniago_VanRyzin
- Re: Groups of groups (and... Jim Sullivan
- Re: Groups of groups ... Pierette_Maniago_VanRyzin
- Re: Groups of groups (and... peter honeyman
- Re: Groups of groups (and... mdw
- Re: Groups of groups (and eff... Peter Lister, Cranfield Computer Centre
- Re: Groups of groups (and... John Hascall
- Re: Groups of groups (and eff... Paul Howell
- Re: Groups of groups (and eff... Mahesh "BigMan" Subramanya
- Re: Groups of groups (and... Bill Fithen
