[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Nmh-workers] IMAP/nmh, again
From: |
valdis . kletnieks |
Subject: |
Re: [Nmh-workers] IMAP/nmh, again |
Date: |
Fri, 27 Oct 2017 18:27:13 -0400 |
On Fri, 27 Oct 2017 22:29:32 +0100, Ralph Corderoy said:
> What does `scan -version' give?
> Is 1-10426 actually 10,426 emails? :-)
Linux-kernel is a firehose. 800 to 1200 emails a day (depending on whether or
not Greg KH drops a -stable review patchbomb series, those can be 300 mails
right there :). And I probably subscribed to the list around kernel 2.5.47
(Nov 2002), at which time it was still a garden hose :)
> If you have grep(1) read the files up to the end of the headers instead,
> does it still take as long? Something like
(as root) echo 3 >| /proc/sys/vm/drop_caches to get cache-cold first run
[~/Mail/linux-kernel] time grep -m1 '^$' `seq 1 10426` > /dev/null
real 0m17.566s
user 0m0.141s
sys 0m2.087s
[~/Mail/linux-kernel] time grep -m1 '^$' `seq 1 10426` > /dev/null
real 0m0.187s
user 0m0.052s
sys 0m0.129s
So cache-cold is within 0.5%. I'm pretty sure what we're measuring there is
how many rotations at 5400rpm are needed (and given that on a cache-cold
it will need at least 1 i/O to read the inode and get the extent list, and a
second
to read the first 4K of the file, that's 20,852 I/O minimum. And in 17.5
seconds
there's only 94,500 rotations. And assuming a 1ms track-to-track seek time,
each one
eats the equivalent of 5 rotations, 4 if it's smart enough to do rotational
positioning during
the seek - which eats up the rest of the available rotations and then some
unless
we've got some read-ahead going....
(several tries later)
Well, cache-cold time changed. Now is a consistent:
time grep -m1 '^$' `seq 1 10426` > /dev/null
real 0m10.781s
Conclusion 1: The disk is known to be a hybrid (with a small 16G or so flash
on the front), and after 4 or 5 references, they've been promoted to flash. Not
sure
that 10.7s makes sense, I would have expected more speedup.
Conclusion 2: I don't understand ext4 as well as I thought I did. ;)
pgpD1bdRwJ9Bx.pgp
Description: PGP signature
- Re: [Nmh-workers] IMAP/nmh, again, (continued)
Re: [Nmh-workers] IMAP/nmh, again, Ralph Corderoy, 2017/10/26
- Re: [Nmh-workers] IMAP/nmh, again, Ken Hornstein, 2017/10/26
- Re: [Nmh-workers] IMAP/nmh, again, Ralph Corderoy, 2017/10/27
- Re: [Nmh-workers] IMAP/nmh, again, Ken Hornstein, 2017/10/27
- Re: [Nmh-workers] IMAP/nmh, again, valdis . kletnieks, 2017/10/27
- Re: [Nmh-workers] IMAP/nmh, again, Ken Hornstein, 2017/10/27
- Re: [Nmh-workers] IMAP/nmh, again, Ralph Corderoy, 2017/10/27
- Re: [Nmh-workers] IMAP/nmh, again,
valdis . kletnieks <=
Re: [Nmh-workers] IMAP/nmh, again, Ralph Corderoy, 2017/10/27
Re: [Nmh-workers] IMAP/nmh, again, Michael Richardson, 2017/10/26