Messages in this thread | | | Date | Thu, 27 May 1999 14:00:21 +0000 | From | Glen Turner <> | Subject | Re: Quota file format proposal |
| |
Jan Kara wrote:
> * We don't have good estimation of how large should hash table be > and resizing it is really pain.
Most system administrators know how many users are on their machines. Set a reasonable default for small machines and allow this to be overridden on the command line by the utility that writes the quota file.
From a user interface point of view, don't ask for the size of the hash table. Ask for the estimated number of disk-using users.
Have an option that allows a space/time tradeoff (which would technically be the probability of a hash value collision). Than derive the hash table size from these two values. The hash divisor should be prime, as user IDs tend to be clustered. Finally, a real-life use for those prime-searching algorithms that clutter CompSci textbooks :-)
The size of the hash table would be recorded in the quota file header.
Changing the size of the hash table can be done by exporting and importing the quota file -- something already done by edquota.
----------------------------
On 32-bit UIDs and GIDs in general. Linux *really* needs these. In an enterprise trying to implement enterprise- wide security, UNIXen UIDs have to be unique and can never be reused. For universities with large student populations, a 16-bit UID space is consumed very quickly.
Linux's 16-bit UIDs are *the* reason that our computing facilities' staff prefer Microsoft Windows NT or Solaris in computing labs (ie: the #1 reason wasn't mindshare, ease of use, application availability, ...) The reason is that WinNT and Solaris allow lab staff to pass the load of user and password management onto the central NDS/LDAP/NIS+/Kerberos-enabled facility.
A time-use survey here last year found that about 10% of our computing lab staff (ie, about an hour a day) was spent managing user accounts when they didn't use the central facility. This is a cost-of-ownership penalty of about US$50 per machine per year.
About 30% of lab staff time was spent installing and managing software, something that is much simpler for most Linux distributions. Hopefully, software bundling will become harmonised across Linux distributions with the LSB project. This would give Linux a significant running-cost advantage over Windows.
These two steps would allow our lab staff to do the thing that the survey said they wanted to do most: help users (currently <10% of time, listed as the activity that gave most job satisfaction).
-- Glen Turner Network Specialist Tel: (08) 8303 3936 Information Technology Services Fax: (08) 8303 4400 The University of Adelaide 5005 Email: glen.turner@adelaide.edu.au South Australia
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |