On Tue, Dec 19, 2000 at 12:16:03PM +0100, Holger Kiehl wrote:
> Does it also include the raid1 readbalance patch from Mika?
it seems raid code in 2.2.18aa2 does rebalancing during reads in raid1:
/*
* read balancing logic:
*/
mirror = conf->mirrors + last_used;
bh->b_rdev = mirror->dev;
sectors = bh->b_size >> 9;
if (bh->b_blocknr * sectors == conf->next_sect) {
conf->sect_count += sectors;
if (conf->sect_count >= mirror->sect_limit)
switch_disks = 1;
} else
switch_disks = 1;
conf->next_sect = (bh->b_blocknr + 1) * sectors;
/*
* Do not switch disks if full resync is in progress ...
*/
if (switch_disks && !conf->resync_mirrors) {
conf->sect_count = 0;
last_used = conf->last_used = mirror->next;
/*
* Do not switch to write-only disks ...
* reconstruction is in progress
*/
while (conf->mirrors[last_used].write_only)
conf->last_used = conf->mirrors[last_used].next;
}
bh_req = &r1_bh->bh_req;
memcpy(bh_req, bh, sizeof(*bh));
bh_req->b_end_io = raid1_end_request;
bh_req->b_dev_id = r1_bh;
map_and_make_request (rw, bh_req);
return 0;
Of course you can't do rebalancing during writes.
Is Mika's patch more advanced than the above?
> And why is the 'shared' value of free or /proc/meminfo always 0. Under
> plain 2.2.18 this is not the case. I saw this under aa1 and aa2.
And you will see this under 2.4.0 too ;). It will avoid to lose clock ticks and
latency loss while reading /proc/meminfo if you have some good ram (starting
from 256M). On a 9G alpha machine xosview rended the machine kind of unusable
due such evil O(N) loop ;).
Andrea
-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to [EMAIL PROTECTED]
