write performance on trunk

[Mikeal Rogers <mi...@gmail.com>]

I don't know what we did, but in my tests the write performance on trunk is
about 3x faster than 0.11

http://mikeal.couchone.com/graphs/_design/app/_show/compareWriteReadTest/e69057a29bd6e4ac4ae0115fac018487

The left vertical column is the average response time in ms and the bottom
horizontal line is the duration in seconds.

This is comparing 50 writers and 200 readers using large document. I'm also
going to run the test with small documents to see how big the difference is.
Both are running with delayed_commits = false ;

-Mikeal

Re: write performance on trunk

[Mikeal Rogers <mi...@gmail.com>]

Whoops, i read my graph wrong. We're actually slower for really small docs.

The thing about the small doc test is that the docs are actually so small
that they aren't realistic. I ran the test again with +A 4 added to 0.11 and
it was about twice as bad.

Again tho, these docs are absurdly small so I think that when the documents
size increases the performance of +A balances out. Also, if you had load on
more than one DB +A is going to dramatically improve the performance.

-Mikeal

On Thu, Jul 1, 2010 at 3:00 PM, Mikeal Rogers <mi...@gmail.com>wrote:

> And this is the same test but with much smaller documents:
>
>
> http://mikeal.couchone.com/graphs/_design/app/_show/compareWriteReadTest/e69057a29bd6e4ac4ae0115fac0193c9
>
> The coolest part is that it looks like our read performance has stabilized
> quite a bit.
>
> -Mikeal
>
>
> On Thu, Jul 1, 2010 at 2:51 PM, Damien Katz <da...@apache.org> wrote:
>
>> Awesome! I think most of this is due to Adam and Randall. Nice work guys
>> :)
>>
>> -Damien
>>
>> On Jul 1, 2010, at 2:48 PM, Mikeal Rogers wrote:
>>
>> > I don't know what we did, but in my tests the write performance on trunk
>> is
>> > about 3x faster than 0.11
>> >
>> >
>> http://mikeal.couchone.com/graphs/_design/app/_show/compareWriteReadTest/e69057a29bd6e4ac4ae0115fac018487
>> >
>> > The left vertical column is the average response time in ms and the
>> bottom
>> > horizontal line is the duration in seconds.
>> >
>> > This is comparing 50 writers and 200 readers using large document. I'm
>> also
>> > going to run the test with small documents to see how big the difference
>> is.
>> > Both are running with delayed_commits = false ;
>> >
>> > -Mikeal
>>
>>
>

Re: write performance on trunk

[Mikeal Rogers <mi...@gmail.com>]

And this is the same test but with much smaller documents:

http://mikeal.couchone.com/graphs/_design/app/_show/compareWriteReadTest/e69057a29bd6e4ac4ae0115fac0193c9

The coolest part is that it looks like our read performance has stabilized
quite a bit.

-Mikeal

On Thu, Jul 1, 2010 at 2:51 PM, Damien Katz <da...@apache.org> wrote:

> Awesome! I think most of this is due to Adam and Randall. Nice work guys :)
>
> -Damien
>
> On Jul 1, 2010, at 2:48 PM, Mikeal Rogers wrote:
>
> > I don't know what we did, but in my tests the write performance on trunk
> is
> > about 3x faster than 0.11
> >
> >
> http://mikeal.couchone.com/graphs/_design/app/_show/compareWriteReadTest/e69057a29bd6e4ac4ae0115fac018487
> >
> > The left vertical column is the average response time in ms and the
> bottom
> > horizontal line is the duration in seconds.
> >
> > This is comparing 50 writers and 200 readers using large document. I'm
> also
> > going to run the test with small documents to see how big the difference
> is.
> > Both are running with delayed_commits = false ;
> >
> > -Mikeal
>
>

Re: write performance on trunk

[Damien Katz <da...@apache.org>]

Awesome! I think most of this is due to Adam and Randall. Nice work guys :)

-Damien

On Jul 1, 2010, at 2:48 PM, Mikeal Rogers wrote:

> I don't know what we did, but in my tests the write performance on trunk is
> about 3x faster than 0.11
> 
> http://mikeal.couchone.com/graphs/_design/app/_show/compareWriteReadTest/e69057a29bd6e4ac4ae0115fac018487
> 
> The left vertical column is the average response time in ms and the bottom
> horizontal line is the duration in seconds.
> 
> This is comparing 50 writers and 200 readers using large document. I'm also
> going to run the test with small documents to see how big the difference is.
> Both are running with delayed_commits = false ;
> 
> -Mikeal