On Sun, Jan 12, 2014 at 6:27 PM, Daniel Farina <[email protected]> wrote:

> On Sun, Jan 12, 2014 at 6:09 PM, Joe Van Dyk <[email protected]> wrote:
> > On Sunday, January 12, 2014 5:57:16 PM UTC-8, Daniel Farina wrote:
> >>
> >> On Sun, Jan 12, 2014 at 5:46 PM, Joe Van Dyk <[email protected]> wrote:
> >> > On Sunday, January 12, 2014 5:45:49 PM UTC-8, Joe Van Dyk wrote:
> >> >>
> >> >> On Sunday, January 12, 2014 5:24:41 PM UTC-8, Daniel Farina wrote:
> >> >>>
> >> >>>
> >> >>> On Jan 12, 2014 4:54 PM, "Joe Van Dyk" <[email protected]> wrote:
> >> >>> >
> >> >>> > When doing a backup-fetch, I get about 6.5-8 megabytes per second
> >> >>> > download from S3. I was expecting it to be a bit faster. Measured
> it
> >> >>> > through
> >> >>> > iftop and nethogs.
> >> >>> >
> >> >>> > I'm noticing that wal-e is using close to 100% of cpu, top shows
> it
> >> >>> > hovering at about 97%.
> >> >>>
> >> >>> Python 2.6 you say? This version cannot block 3DES encipherment
> >> >>> because
> >> >>> of a missing feature of the ssl module.  This fix helped my
> >> >>> surprisingly
> >> >>> slow cases a lot.
> >> >>>
> >> >>> To confirm, try running Linux 'perf top' and see what symbols crop
> up.
> >> >>> The 3DES symbols in openssl were perspicuously named, which is how
> >> >>> this bug
> >> >>> was diagnosed.
> >> >>
> >> >>
> >> >> I upgraded to Python 2.7.6, no changes in speed. I'll check out perf
> >> >> tomorrow.
> >>
> >> Unfortunately.  The 3DES blocking helped my cases a lot.
> >>
> >> > Should wal-e be using such a high cpu percentage?
> >>
> >> I haven't tuned WAL-E very much for CPU usage, but 6 megabytes per
> >> second is suspicious.  Can you get a faster rate if you send /dev/zero
> >> to the disk?
> >
> > I can write at about 68 MB/s.
>
> Not so good.
>
> I have measured WAL-E going faster in the past, but maybe there was a
> regression.
>
> Unfortunately you may have to take point figuring out why for a little
> while if you want to see progress Real Soon, because I'm pretty sure
> there's a nasty bug in backup-fetch right now and that's going to be
> holding my attention.  Maybe one of the other list denizens/github
> issue-watchers with an eye for optimization can help.
>
> I think this level of performance is worth a Github bug report also,
> if only to widen the distribution in hopes someone will make time to
> profile.  I bet there's some howlers in the code right now, as it has
> not been tuned for CPU usage modulo that 3DES change to my
> knowledge...ever.
>

from "perf top":

75.38%  [kernel]               [k] hypercall_page
 3.39%  python2.7              [.] 0x179be3
3.18%  libc-2.15.so           [.] 0x140931
2.86%  libcrypto.so.1.0.0     [.] 0x66dbf
2.78%  python2.7              [.] PyEval_EvalFrameEx
1.77%  liblzo2.so.2.0.0       [.] lzo1x_decompress_safe
1.02%  [kernel]               [k] copy_user_generic_string
0.73%  liblzo2.so.2.0.0       [.] lzo_adler32
0.59%  python2.7              [.] _PyObject_GenericGetAttrWithDict
0.28%  libc-2.15.so           [.] epoll_ctl

-- 
You received this message because you are subscribed to the Google Groups 
"wal-e" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/groups/opt_out.

Reply via email to