Re: /tmp in swap space
Richard Gooch (Richard.Gooch@atnf.CSIRO.AU)
Sun, 24 May 1998 15:19:59 +1000
Larry McVoy writes:
> : Well, I can give you a real-world example. Because I haven't yet found
> : an OS which has useful mmapping semantics for /dev/zero (namely that I
> : can open /dev/zero and pass a FD to another process and both processes
> : can mmap() the FD and share data via the mmapped region), I end up
> : creating a tmpfile in /tmp for mmapping purposes. Under Linux, this is
> : at least 16 MBytes (because under 16 MBytes I can use ordinary SysV
> : SHM). This tmpfile doesn't need to be saved to disc (not when you have
> : gobs of RAM).
> : So here is a real case where large files with a moderate lifetime (at
> : least a few minutes) are created in /tmp.
>
> I agree with you that it would be nice to be able to share anonymous
> memory between processes (which you can do with clone but that's not an
> answer for unrelated processes).
That's right. And in my situation, I don't even have a parent-child
relationship between the two processes.
> However, we were discussing TMPFS vs EXT2FS performance. Do you have
> any reason to believe that creating a mapping in TMPFS is any faster
> than one in EXT2FS?
Not really. I was just pointing out a real-world situation. One thing
to consider may be that while many systems will configure 128 MBytes
of swap, they might not allocate large /tmp partitions. But then, a
small /tmp is not a justification for adding tmpfs to the kernel :-)
Regards,
Richard....
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu