Solving the ext3 latency problem [LWN.net]

Solving the ext3 latency problem

Posted Apr 14, 2009 20:10 UTC (Tue) by drag (guest, #31333)
In reply to: Solving the ext3 latency problem by malor
Parent article: Solving the ext3 latency problem

Personally I would think that multiuser actually means having multiple people with users using the computer at the same time.

That is they are logged onto the machine and are doing something on them.. programming, editing, web browsing, etc.

So your dealing with computers with multiple monitors with multiple users logged in at once, or LTSP, or people that still sell shell accounts. All of which are fairly rare compared to personal desktops, embedded systems, or most server systems.

to post comments

Solving the ext3 latency problem

Posted Apr 14, 2009 23:24 UTC (Tue) by ktanzer (guest, #6073) [Link] (1 responses)

Multi-user systems may be relatively rare, but as someone who was responsible for multiple LTSP deployments, I shudder at the thought that security in such cases would be cast aside or relegated to an afterthought. LTSP is a great way to use old hardware. For that matter, any home machine used by the whole family would also be multi-user (assuming people set up their own logins).

On a more general note, Linux has inherited rich multi-user capabilities from Unix, and I hate to see that atrophy over time. As one example, it is very easy to find information on how to configure popular programs such as Firefox, KDE, OpenOffice, etc. for a single user, but often maddeningly hard to determine how to configure on a system-wide basis. The fact that FF can't run multiple instances from a single profile is not technically a multi-user issue, but also drives me up the wall...

Solving the ext3 latency problem

Posted Apr 15, 2009 1:25 UTC (Wed) by drag (guest, #31333) [Link]

Well I can agree with that.

I don't know exactly why, but I have a feeling that multiuser systems will become increasingly important in the future.

Something will come along... like the acceptance of IPv6 and the decline of the "Personal Computer"-ing inflicted client-server relationship and the internet will return to it's P2P roots. (for reasons of scalability, robustness, expense). If something like that were to happen and people realized that having mobile computers could become essentially disposible if they turned into little more then terminals for the 'big' computer at home or clusters at work... Then multiuser systems could become common place again.

Weirder things have happenned.

Solving the ext3 latency problem

Posted Apr 15, 2009 4:35 UTC (Wed) by malor (guest, #2973) [Link]

Typically, different services run under different user accounts, and rely on the system's security features to keep the two separate.

It would be bad if a bug in the mail server gave access to, say, deleted .htaccess files, or part of a SQL database.

All Unix systems are inherently multiuser, and sabotaging inter-account security features is deliberately cutting away one layer of the net that can catch you if a bug exposes an attack vector.

Solving the ext3 latency problem

Posted Apr 15, 2009 5:55 UTC (Wed) by hawk (subscriber, #3195) [Link]

It may not be "multi-user" by that definition but the exact definition of "multi-user" isn't really that interesting for the actual argument, is it?

Isn't the point that as soon as soon as there are multiple users (no matter if they are logged in using a system account or accessing the system through some other means, eg HTTP, authenticated in some way or even anonymously), there would be a chance that one user's data (or data "belonging to the system") could leak into a file which will be accessible by another user.

So in the case of the the system crashing, a file publicly available on the web, or some logged in user(s), might end up containing anything that has previously been deallocated if that file was being modified, be it by the site administrator, or by a random user on a site where you can for instance upload an image to include in your content.

I would think this definitely falls within the "common usage"-realm for Linux systems and that whoever made the argument may not have really thought it through.

(Or I'm just not understanding in what scenarios something like this could actually happen.)