home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!auspex-gw!guy
- From: guy@Auspex.COM (Guy Harris)
- Newsgroups: comp.sys.sun.apps
- Subject: Re: Yes, f77 can be taught not to use /tmp
- Message-ID: <16565@auspex-gw.auspex.com>
- Date: 22 Jan 93 05:48:04 GMT
- References: <1993Jan20.185627.19507@mapsut.einstein.com> <1993Jan22.014443.17806@mapsut.einstein.com>
- Sender: news@auspex-gw.auspex.com
- Organization: Auspex Systems, Santa Clara
- Lines: 30
- Nntp-Posting-Host: auspex.auspex.com
-
- >Use virtual memory (tmpfs) for the tmp directory. I did not try this,
- >but several respondents said it does.
-
- Does what? :-)
-
- Basically, what it does is gives you a file system that you might think
- of as living on the "b" partition of your root disk. I.e., its
- "metadata", including directories, are (wired-down) data structures in
- the kernel's address space, and its data are "anonymous" pages backed by
- swap space.
-
- If you have the swap space to spare, it lets you use it for temporary
- files. (If you don't, you run the risk of getting ENOSPC errors, just
- as you do on a "/tmp" directory on a regular file system.)
-
- It also should run faster than does a "/tmp" directory on a normal file
- system (and seems to have done so when I've tried it).
-
- For one thing, metadata operations don't have to touch the disk at all -
- creating or deleting files just means updating some data structures in
- the kernel.
-
- For another thing, "update" or other users of "sync()" won't push dirty
- temporary-file pages out to their backing store, so you may avoid some
- UFS or NFS writes to the disk or the network. (If the file is deleted
- before the pages get pushed, they'll *never* get pushed to the backing
- store.)
-
- However, it does mean that anything that must survive a reboot had
- better not be in "/tmp"....
-