[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: SEUL: Partitioning

On Mon, 30 Jun 1997, Jay Bloodworth wrote:

> How big of a performance hit do you take going through a loopback device?
> I haven't looked at the code, but I assume every io syscall requires two
> trips through the VFS layer: one for the initial call and one to perform
> the operation on the underlying file system.

The file is fixed in size+position, so it can be treated exactly like a
partition on the drive - this may require some kernel changes, but they
should be minor. I have never personally tried loopback, so I'll give it a
go tonight (I saw it when kernel v2.0 came out, thought "Cool!" and forgot
about it). I would not have thought the extra VFS step would cause too much
of a performance hit, but I will try it and report the findings later. I use
NFS over ethernet and find it fast enough.

> While others have made pretty convincing arguments for using loopback, my
> gut says that's not the way to go.  The performance issues raised above,
> the possibility of users deleting or corrupting their LINUX.DRV file, and
> figuring out how to 'right size' the initial partition are the biggest
> practical concerns, but I think my real worry is philosophical.

If someone does corrupt LINUX.DRV, then this will happen under Windows or
DOS rather than Linux, and will probably damage something else in the
process. e2fsck will fix it. The point is that after the user has loaded
Linux and spent the introductory half hour learning how to get around, they
will be so impressed that they won't ever go back to DOS/Windows. The size
problem would be exactly the same for a fixed partition.

> Trial mode, while a nice idea on the surface, makes the assumption that
> 'ordinary' users spend lots of time trying things on their computers.  I
> think it is hard for us a Linux users to imagine not wanting to tinker
> with one's computer to get the best possible performance out of it.  If
> you make installation a two stage process (trial mode and full install),
> the set of people who perform the 'safe' trial mode install may be larger
> than if you had forced them do deal with repartitioning initially, but I
> think the subset of that group that goes on to the second stage will be a
> good bit smallers since most will have gotten fed up with the install by
> then.  Put more simply, I think it is better to get the complexity of the
> install out of the way in one fell swoop rather than trying to dilute it
> by spreading it over multiple stages.

This is a valid point. However, this should not be described as a "trial
mode" - it should be one of the steps in the installation process:

"Do you know what a disk partition is?"
 Yes:  give option of repartitioning (recommended) or DOS file
 No:   DOS file

I think a file-based system (as opposed to a partition-based system) would
work as a permanent install option, but I don't know about the performance.

> I think we should remember we are competing with NT Workstation, not
> Windows 95.  Our pontential users are those that are savvy enough to know
> that they can get more from their machines than Win95 can provide, but who
> don't know exactly how to get at it.  Based on general good press, these
> people may go into their local software retailer and buy a copy of NT
> Workstation (maybe OS/2 Warp) to replace the copy of Windows 95 they came
> preloaded on their machine.  Our job is to make SEUL a viable alternative
> for these people who have already made up their mind that Windows 95
> doesn't give them everything they want.

I don't know why people would want to install SEUL, but I would personally
be tempted to aim it at:

* programmers looking for a more efficient programming platform
* companies (large or small) looking for an inexpensive server
* students who want (or need) UN*X for free

I think a lot could be done with older systems and SEUL as well.

> Our ultimate goal should be for major computer manufactures to preload
> SEUL on their commodity systems.  To accomplish that goal, we must create
> a significant body of users who want SEUL and nothing else.  I have argued
> previously that the best place to generate such a body is among
> corporate/organizational users.  Barring that, we should concentrate on
> users really savvy enough to appreciate what they are getting with SEUL.

Yet again, this comes down to defining the end user. We need to have a
fixed, carved-in-stone definition of *who this project is aimed at* before
much more progress can be made...

Thomas Molesworth            (thomas@bass.almac.co.uk)

Simple End User Linux Mailing list
To be removed from this mailing list send a message to majordomo@txcc.net
with the line
unsubscribe seul-project
in the body of the letter.