Re: Windows and Maildir
From: Sam (sam_at_email-scan.com)
Date: 07/14/04
- Next message: Paul Attryde: "Re: MapViewOfFile with /3GB switch on WinXP Pro"
- Previous message: Ivan Brugiolo [MSFT]: "Re: MapViewOfFile with /3GB switch on WinXP Pro"
- In reply to: Jonathan de Boyne Pollard: "Re: Windows and Maildir"
- Next in thread: Slava M. Usov: "Re: Windows and Maildir"
- Reply: Slava M. Usov: "Re: Windows and Maildir"
- Reply: Arnaud Debaene: "Re: Windows and Maildir"
- Messages sorted by: [ date ] [ thread ]
Date: Wed, 14 Jul 2004 17:50:27 -0500
Jonathan de Boyne Pollard writes:
> I> Surely that colon is not essential to its basic functionality.
>
> S> Yes it is;
>
> No, it isn't. The point of the colon is to allow for the storage of message
> status information outside of the actual file contents. Maildir stores this
> in the message filename itself. Unix has no further mechanisms available.
> Other operating systems *have*, however. OS/2 Warp has extended attributes
> (which seem tailor made for the sort of things that go into the "info" part of
> a message filename). Windows NT with NTFS has both file streams and extended
> attributes.
If you modify or replace the message status flags with something else,
you're no longer talking about maildir-maildirs. It's now something else.
> (The fact that as a message changes status its filename changes
> does cause and has caused difficulties with parallel uses of Maildirs by
> multiple entities.
Only if those entities are not prepared to handle this eventuality.
> S> With an average-sized mailbox containing thousands of messages,
> S> it means that you will have thousands of small files lying around.
> S> Windows fares very poorly with handling directories with large
> S> numbers of files.
>
> False. Several filesystem formats, such as FAT, fare very poorly with
> handling large directories. But "FAT" is (obviously) not synonymous with
> "Windows", and furthermore there are filesystem formats available for Windows
> NT whose performance with large directories is better than that of FAT.
Is that why Linux+Samba spanks NT's monkey on the same hardware?
> Indeed, one only has to realise that the BSD Unix FFS format has the *exact
> same problem* with large directories, because like FAT it too stores
> directories as simple linear lists, to realise that this is *not* a Windows
> problem.
It's more than just the linear directory. Take FFS with a directory
containing ten thousand files, then take NT with the same. No comparison
whatsoever. And there won't be much of a difference between FAT and NTFS.
> (Failing to realise that poor choice of filesystem format will affect Unix
> softwares too, because the problem is in the filesystem format and not in the
> operating system, is a common mistake made by mail administrators on Unix
> systems, incidentally.
FAT: two decades' worth of revisions, and two decades worth of
backwards-compatibility crap. End result: a turd inside a turd inside a
turd to the n'th generation.
NTFS: one massive hairball. Its primary purpose is to raise the
interoperability bar to prevent access to the filesystem from non-Microsoft
technologies. Unnecessarily complicated, overengineered turkey.
In both cases you have massive amounts of bloat to cut through, before you
get to the files themselves.
Unix/Linux filesystems, on the other hand, are models of simplicity. I'll
take a filesystem benchmark of Linux ext3 against NT or XP, on the same
hardware, any time.
> S> Linux, and Unix, handles directories containing large number
> S> of files much better than Windows.
>
> False. As explained above. Poor handling for large directories is just as
> much a Linux and a Unix problem as it is a Windows problem, because it is a
> *filesystem format* problem.
There are “filesystem format” problems, and there are “filesystem format”
problems caused by decades' worth of backward-compatibility hacks (MICROS~1
anyone?) and artificially-overengineered bloat.
> MSDN is, as is all too often the case, oversimplified to the point of being
> wrong. NTFS has an inode analogue, the file's index in the MFT. With a
> little work, it is possible to retrieve a volume-wide unique identifier for a
> file on NTFS in a Windows NT program.
“A little work” versus a plain, garden-variety, system call. Multiply by a
few hundred thousand, and you'll get the answer of why Samba is a better SMB
server than NT/XP itself.
- application/pgp-signature attachment: stored
- Next message: Paul Attryde: "Re: MapViewOfFile with /3GB switch on WinXP Pro"
- Previous message: Ivan Brugiolo [MSFT]: "Re: MapViewOfFile with /3GB switch on WinXP Pro"
- In reply to: Jonathan de Boyne Pollard: "Re: Windows and Maildir"
- Next in thread: Slava M. Usov: "Re: Windows and Maildir"
- Reply: Slava M. Usov: "Re: Windows and Maildir"
- Reply: Arnaud Debaene: "Re: Windows and Maildir"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|