[linux-audio-dev] Disk use causes freezes

New Message Reply About this list Date view Thread view Subject view Author view Other groups

Subject: [linux-audio-dev] Disk use causes freezes
From: Juhana Sadeharju (kouhia_AT_nic.funet.fi)
Date: Wed Aug 01 2001 - 15:19:35 EEST


Hello. Found this:

 ==begin==
I sit behind an NT dual P2/400 256MB for too many hours each day. It is
"slow". One of my (fun) tasks is CD mastering. When i'm loading a 500MB
wav file into the editor, the machine goes useless for two minutes while
loading the wave file. Basically its just reading the file and writing it
to swap. The CPU is at 5%. No Alt-Tab between apps during that time. No
chance of putting the mouse over a button (too many dropped ticks). The
NT kernel is just sitting waiting for I/O completions.

Linux would be embarrassed to be that bad.
 ==end==

I'm a bit worried about it because Debian 2.2 rev 2 is behaving similarly.
If I copy large files, or perform md5sums on large files, it may take even
45 seconds before I get the password prompt on the console or get
a directory listing. Typically it takes 25-30 seconds. Writing text in
Emacs may have 4-5 seconds latency; sometimes larger.

It would be bad if the GUI of audio software freezes similarly. I mean,
what use would be of 3 ms audio latency if other parts of software freeze
for 10 seconds?

Yes I have a swap space, and it looks like it is taking all the power of
Linux under disk use (entire memory is filled with md5summed file and no
room is left to other processes). It could be wise to take swap away if
one is expected to use an audio editor in real-time, but that should not
be the solution to the problem.

How to test if my disk system and Linux is working okay? Just using the
disk and observing the slowness of the system tells actually nothing
(MS Windows works okay).

The memory is filled with file data, so, would it help if read()/write()
buffers would be locked? If yes, then all file accesses of audio programs
should be handled through locked memory buffers (and use of any other
programs should be avoided). Simplest way could be that applications would
each run their own disk access plug-in in LAAGA (where plug-ins gets root
privileges and can lock the buffers, or?).

True solution would be to improve Linux's swap code, but that seems
to be a bigger problem.

Best regards,

Juhana


New Message Reply About this list Date view Thread view Subject view Author view Other groups

This archive was generated by hypermail 2b28 : Wed Aug 01 2001 - 15:20:45 EEST