You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by "C. Michael Pilato" <cm...@collab.net> on 2009/03/12 15:28:40 UTC
Newsflash! HTTP v2 reduces network turnarounds! (As intended.)
Here are some digits I just posted to IRC. I was basically running
commit_tests.py over HTTP in various ways, and then using 'wc -l' on the
access_log to see how many requests were fired off to complete those tests.
Here are the numbers:
serf, v2 enabled: 3601 requests
serf, v2 disabled: 4581 requests
So, the addition of HTTP v2 changes to ra-serf has resulted in a 20%
reduction in the number of requests aimed at the server. Those are almost
certainly all PROPFIND and CHECKOUT requests.
neon, trunk today: 2724 requests
Much fewer requests for ra-neon (which has a subset of the HTTP v2 changes
in it), but this isn't a fair comparison because ra-neon uses full REPORTs
for its update operations, whereas ra-serf uses skelta REPORTs with
subsequent GETs and PROPFINDs. Fortunately, as of 1.5, we can tell the
server not to allow those full REPORTs, which forces ra-neon to behave as
ra-serf does in that manner, and gives us a more fair comparison:
neon, skelta updates: 4104 requests
Cool stuff.
--
C. Michael Pilato <cm...@collab.net>
CollabNet <> www.collab.net <> Distributed Development On Demand
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=1313346
Re: Newsflash! HTTP v2 reduces network turnarounds! (As intended.)
Posted by Justin Erenkrantz <je...@apache.org>.
On Thu, Mar 12, 2009 at 8:28 AM, C. Michael Pilato <cm...@collab.net> wrote:
> So, the addition of HTTP v2 changes to ra-serf has resulted in a 20%
> reduction in the number of requests aimed at the server. Those are almost
> certainly all PROPFIND and CHECKOUT requests.
w00t. -- justin
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=1313616
Re: Newsflash! HTTP v2 reduces network turnarounds! (As intended.)
Posted by Daniel Rall <dl...@finemaltcoding.com>.
On Thu, Mar 12, 2009 at 8:28 AM, C. Michael Pilato <cm...@collab.net> wrote:
> Here are some digits I just posted to IRC. I was basically running
> commit_tests.py over HTTP in various ways, and then using 'wc -l' on the
> access_log to see how many requests were fired off to complete those tests.
>
> Here are the numbers:
>
> serf, v2 enabled: 3601 requests
> serf, v2 disabled: 4581 requests
>
> So, the addition of HTTP v2 changes to ra-serf has resulted in a 20%
> reduction in the number of requests aimed at the server. Those are almost
> certainly all PROPFIND and CHECKOUT requests.
>
> neon, trunk today: 2724 requests
>
> Much fewer requests for ra-neon (which has a subset of the HTTP v2 changes
> in it), but this isn't a fair comparison because ra-neon uses full REPORTs
> for its update operations, whereas ra-serf uses skelta REPORTs with
> subsequent GETs and PROPFINDs. Fortunately, as of 1.5, we can tell the
> server not to allow those full REPORTs, which forces ra-neon to behave as
> ra-serf does in that manner, and gives us a more fair comparison:
>
> neon, skelta updates: 4104 requests
>
> Cool stuff.
Nice work, Mike!
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=1313570
Re: Newsflash! HTTP v2 reduces network turnarounds! (As intended.)
Posted by Mark Phippard <ma...@gmail.com>.
On Thu, Mar 12, 2009 at 4:37 PM, Ben Collins-Sussman
<su...@red-bean.com> wrote:
> On Thu, Mar 12, 2009 at 10:28 AM, C. Michael Pilato <cm...@collab.net> wrote:
>> Here are some digits I just posted to IRC. I was basically running
>> commit_tests.py over HTTP in various ways, and then using 'wc -l' on the
>> access_log to see how many requests were fired off to complete those tests.
>>
>> Here are the numbers:
>>
>> serf, v2 enabled: 3601 requests
>> serf, v2 disabled: 4581 requests
>>
>> So, the addition of HTTP v2 changes to ra-serf has resulted in a 20%
>> reduction in the number of requests aimed at the server. Those are almost
>> certainly all PROPFIND and CHECKOUT requests.
>
> So um.... was it faster? How much faster?
Wouldn't the tests have to be run in a high latency setting to really
become noticeable? Shaving 9 PROPFINDS in a LAN environment or
localhost is not going to be measurable is it?
--
Thanks
Mark Phippard
http://markphip.blogspot.com/
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=1314299
Re: Newsflash! HTTP v2 reduces network turnarounds! (As intended.)
Posted by Ben Collins-Sussman <su...@red-bean.com>.
On Thu, Mar 12, 2009 at 10:28 AM, C. Michael Pilato <cm...@collab.net> wrote:
> Here are some digits I just posted to IRC. I was basically running
> commit_tests.py over HTTP in various ways, and then using 'wc -l' on the
> access_log to see how many requests were fired off to complete those tests.
>
> Here are the numbers:
>
> serf, v2 enabled: 3601 requests
> serf, v2 disabled: 4581 requests
>
> So, the addition of HTTP v2 changes to ra-serf has resulted in a 20%
> reduction in the number of requests aimed at the server. Those are almost
> certainly all PROPFIND and CHECKOUT requests.
So um.... was it faster? How much faster?
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=1314290
Re: Newsflash! HTTP v2 reduces network turnarounds! (As intended.)
Posted by Ben Collins-Sussman <su...@red-bean.com>.
On Thu, Mar 12, 2009 at 4:00 PM, David Glasser <gl...@davidglasser.net> wrote:
> ... we implemented a feature allowing server administators to force clients
> to make more roundtrips?
>
Long ago. It allows admins to force tighter logging and authz. When
a single REPORT is used to fetch a tree all at once, nothing goes into
the apache log. Some admins want to see a separate GET for every path
in the log.
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=1314372
Re: Newsflash! HTTP v2 reduces network turnarounds! (As intended.)
Posted by David Glasser <gl...@davidglasser.net>.
... we implemented a feature allowing server administators to force clients
to make more roundtrips?
I think I need to just stop reading DAV-related threads since I am finding
it difficult not to be boggled by everything I learn about it.
--dave
On Mar 12, 2009 8:29 AM, "C. Michael Pilato" <cm...@collab.net> wrote:
Here are some digits I just posted to IRC. I was basically running
commit_tests.py over HTTP in various ways, and then using 'wc -l' on the
access_log to see how many requests were fired off to complete those tests.
Here are the numbers:
serf, v2 enabled: 3601 requests
serf, v2 disabled: 4581 requests
So, the addition of HTTP v2 changes to ra-serf has resulted in a 20%
reduction in the number of requests aimed at the server. Those are almost
certainly all PROPFIND and CHECKOUT requests.
neon, trunk today: 2724 requests
Much fewer requests for ra-neon (which has a subset of the HTTP v2 changes
in it), but this isn't a fair comparison because ra-neon uses full REPORTs
for its update operations, whereas ra-serf uses skelta REPORTs with
subsequent GETs and PROPFINDs. Fortunately, as of 1.5, we can tell the
server not to allow those full REPORTs, which forces ra-neon to behave as
ra-serf does in that manner, and gives us a more fair comparison:
neon, skelta updates: 4104 requests
Cool stuff.
--
C. Michael Pilato <cm...@collab.net>
CollabNet <> www.collab.net <> Distributed Development On Demand
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=1313346
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=1314350