Ivan Zhakov <i...@visualsvn.com> writes:
> On 4 January 2016 at 16:25, Philip Martin <philip.mar...@wandisco.com> wrote:
>>
>> When I checkout Subversion trunk from my local mirror I cannot measure a
>> client gain, but I can measure better server efficiency:
>>
>> - The CPU used by Apache goes down from 1.2s to 1.1s.
>>
>> - The number of system calls made by Apache goes down
>>
>> 68822 to 50664 for hot FSFS cache
>> 178885 to 161722 for cold FSFS cache
>>
> I'm getting similar results on Windows.
Using strace on Linux it is possible to see how much data is returned by
the reads. SVNCacheFullTexts affects how much gets cached and thus how
significant the uncached revprops become. Looking at the total amout of
data read by apache during a checkout:
Initial checkout Second Checkout
unpatched
fulltext cache 150MB 100MB
no fulltext cache 155MB 105MB
patched
fulltext cache 105MB 35MB
no fulltext cache 120MB 50MB
I haven't separated disk reads from network reads but I suspect the
network reads are only a few MB. These numbers show how much repository
data has to be read from disk, or from the OS cache, to serve a
checkout. In the best case the patch removes 2/3 of the data: 35MB
instead of 100MB.
--
Philip Martin
WANdisco
