You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by solo turn <so...@yahoo.com> on 2003/02/10 18:56:42 UTC
faster client pre-1.0: neon prefetching, multithreading
i tried to profile also the https client, the results are:
full update, https, server/client on same machine, duration 728sec
(reminder: ra_local took 280sec):
time percent fct
svn_cl__update
...
reporter_finish_reports
svn_ra_dav__parsed_request
ne_request_dispatch
434 61 ne_read_response_block
gz_reader,ne_xml_parse_v, ne_xml_parse, xmlParseChunk,
xmlParseTryOrFinish,
87 xmlParseEndeTag (615sec)
end_element, end_element
269 44 add_node_props
216 80 svn_ra_dav__get_props_resource
svn_ra_dav__getprops, ne_propfind_allprop,propfind,
ne_request_dispatch,
433 61 ne_read_response_block
275 39 ne_begin_request
53 20 addProps
137 22 close_file
71 12 simple_fetch_file
61 10 set_special_wc_prop
40 7 close_directory
26 4 close_edit
13 xmlParseStartTag
275 39 ne_begin_request, send_request, read_status_line,
ne_sock_readline, read_ossl, readable_ossl, select,
select, _poll
(times are not always according to percentage, cause functions may be
called from different callees)
total times:
272 _poll
259 _rename
119 __open
23 mkdir
20 unlink
9 _write
the following points could be time-consuming:
- there is a lot of disk access and network wait time
which seems not parallel, ie. multithreaded.
- especially urls and their properties seem to be fetched
in order, without any function which gets an url,
with the small properties, and the entries, and the
small properties of the entries in one fetch.
__________________________________________________
Do you Yahoo!?
Yahoo! Mail Plus - Powerful. Affordable. Sign up now.
http://mailplus.yahoo.com
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: faster client pre-1.0: neon prefetching, multithreading
Posted by Daniel Berlin <db...@dberlin.org>.
> the following points could be time-consuming:
> - there is a lot of disk access and network wait time
> which seems not parallel, ie. multithreaded.
>
Easier to use AIO when it's available.
You can just aio_read the sockets/disk stuff as we discover what we
need, then aio_suspend in the routine that actually needs the data to
wait for it all to get there (or a timeout to occur).
> - especially urls and their properties seem to be fetched
> in order, without any function which gets an url,
> with the small properties, and the entries, and the
> small properties of the entries in one fetch.
>
Once again, you can just aio_read this stuff, then wait for it all to
finish at once, rather than wait on them one at a time.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org