From: Duncan <1i5t5.duncan@cox.net>
To: linux-btrfs@vger.kernel.org
Subject: Re: kernel 3.17-rc3: task rsync:2524 blocked for more than 120 seconds
Date: Wed, 3 Sep 2014 05:37:43 +0000 (UTC) [thread overview]
Message-ID: <pan$d7111$755bc381$f59147dd$8ff671d2@cox.net> (raw)
In-Reply-To: C5F63696-2ADD-4CB6-A049-71B71A9D083E@colorremedies.com
Chris Murphy posted on Tue, 02 Sep 2014 20:44:06 -0600 as excerpted:
> On Sep 2, 2014, at 12:40 AM, Duncan <1i5t5.duncan@cox.net> wrote:
>>
>> Mkfs.btrfs used to default to 4 KiB node/leaf sizes; now days it
>> defaults to 16 KiB as that's far better for most usage. I wonder if
>> USB sticks are an exception…
>
> USB sticks > 1 GB get 16KB nodesize also.
Seems you read into that a meaning I didn't even consider when I wrote
it. =:^/
I /meant/ that AFAIK mkfs.btrfs did the usual 16K thing, but that perhaps
(some, depending on erase-block size) USB sticks are an exception to 16K
being better than 4K thing.
> At <= 1 GB, mixed-bg is
> default as is 4KB nodesize. Probably because queue/rotational is 1 for
> USB sticks, they mount without ssd or ssd_spread which may be
> unfortunate (I haven't benchmarked it but I suspect ssd_spread would
> work well for USB sticks).
I did mention ssd_spread somewhere in my replies, due to the same
suspicion. Good to see you have the same suspicion I do. =:^)
> It was suggested a while ago that maybe mixed-bg should apply to larger
> volumes, maybe up to 8GB or 16GB?
Indeed. Considering the default data chunk size is 1 GiB, that as the
cutoff for default mixed-bg mode seems kinda low. If anything, I think
16 GiB is still a low cutoff value, at least as long as the only way to
reclaim out-of-balance data/metadata assigned chunks is via a manual
balance. I'd suggest a 32 GiB cutover by default.
Tho if btrfs gets a good auto-balance-trigger mechanism that say triggers
when allocated (fi show device figure) is within say 10% of total
filesystem space (rounded up to the nearest GiB, minimum 2 GiB) AND data
or metadata has more than say 10% spread between used and allocated (fi
df, again rounded up, 2 GiB minimum), that can arguably come down a
notch, to say 16 GiB.
--
Duncan - List replies preferred. No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master." Richard Stallman
next prev parent reply other threads:[~2014-09-03 5:37 UTC|newest]
Thread overview: 19+ messages / expand[flat|nested] mbox.gz Atom feed top
2014-09-01 13:33 kernel 3.17-rc3: task rsync:2524 blocked for more than 120 seconds john terragon
2014-09-01 16:02 ` Chris Mason
2014-09-01 16:36 ` john terragon
2014-09-02 5:20 ` Duncan
2014-09-02 6:12 ` john terragon
2014-09-02 6:40 ` Duncan
2014-09-02 19:56 ` john terragon
2014-09-02 20:10 ` Chris Mason
2014-09-02 20:23 ` john terragon
2014-09-02 20:48 ` john terragon
2014-09-03 1:31 ` john terragon
2014-09-03 12:36 ` Chris Mason
2014-09-03 14:11 ` john terragon
2014-09-03 15:02 ` Chris Murphy
2014-09-03 2:44 ` Chris Murphy
2014-09-03 5:37 ` Duncan [this message]
2014-09-03 6:03 ` john terragon
2014-09-03 9:14 ` Liu Bo
2014-09-03 3:32 ` Zygo Blaxell
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='pan$d7111$755bc381$f59147dd$8ff671d2@cox.net' \
--to=1i5t5.duncan@cox.net \
--cc=linux-btrfs@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).