On 12/16/2013 02:30 PM, Zbigniew Jędrzejewski-Szmek wrote:
On Mon, Dec 16, 2013 at 08:31:46AM +0100, Holger Winkelmann [TP] wrote:
Hi,
Is there any particular reason? I think thresold for runtime journal
size can lower much because in initramfs it's not supposed to have much
logs.
First, there are some data strcutures which are allocated when the file
is created, and if the file was very small, relatively more space would
wasted. Second, repeated fields are not stored, just referenced, so things
become more efficient when the file is not too small. But neither is
fundamental reason, and with some tweaking the journal could be made
to work much smaller files.
I understand. These are really good points when logs are relatively
large, ie. the journal is stored on a real disk.
However when it's in initramfs context, journal is stored in tmpfs which
is using the real memory resource as it's backend. 4 MB seems a little
bit overkill especially when memory is quite limited case, like kdump.
To be more specific, I think 512 KB or 1 MB is a fairly large enough
nubmer when journal is stored to a volatile backend.
We totally agree that a minimum size must be below 1MB either on flash or
ramfs for embedded devices. otherwise you end up with two solutions for smaller
and bigger devices. Is there any reference about the overhead if you use smaller
file size? Is there technical limitation for a minimum size?
No, there's no real technical limitation. Except some hero should go through
src/jounal/journal-file.c and adjust all the constants that they also work
with very small files.
Zbyszek
Hi,
the minimum file size changed in version 203 from 64KiB to 4MiB. Is
there a special reason for this? The relevant code line seems to be
#define JOURNAL_FILE_SIZE_MIN (4ULL*1024ULL*1024ULL) /* 4 MiB */
which did not change in 208.
Sascha
_______________________________________________
systemd-devel mailing list
systemd-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/systemd-devel
