Test multideque

Test multideque

BUG=552228
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel

[prashant.n, 2015-11-06 04:46:13]

Description was changed from

==========
Test multideque

BUG=
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel
==========

to

==========
Test multideque

BUG=552228
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel
==========

[prashant.n, 2015-11-06 04:46:14]

prashant.n@samsung.com changed reviewers:
+ danakj@chromium.org, enne@chromium.org, reveman@chromium.org,
vmpstr@chromium.org

[prashant.n, 2015-11-06 04:47:12]

Kindly check and do let me know if we should implement in resource pool.

[danakj, 2015-11-09 19:32:51]

https://codereview.chromium.org/1410833005/diff/20001/cc/resources/resource_p...
File cc/resources/resource_pool_unittest.cc (right):

https://codereview.chromium.org/1410833005/diff/20001/cc/resources/resource_p...
cc/resources/resource_pool_unittest.cc:238: return resource;
don't you need to erase from from unused_ here?

https://codereview.chromium.org/1410833005/diff/20001/cc/resources/resource_p...
cc/resources/resource_pool_unittest.cc:271: return resource;
and erase from dq and possibly from unused_ here?

https://codereview.chromium.org/1410833005/diff/20001/cc/resources/resource_p...
cc/resources/resource_pool_unittest.cc:326: tdq.AppendToUnused(resource[i]);
Your results appear to include the cost of creating and appending all the items
too. Is that intentional?

I suggest base::TimeTicks::Now() before and after the acquire loop and LOG the
delta.

[prashant.n, 2015-11-10 04:10:50]

https://codereview.chromium.org/1410833005/diff/20001/cc/resources/resource_p...
File cc/resources/resource_pool_unittest.cc (right):

https://codereview.chromium.org/1410833005/diff/20001/cc/resources/resource_p...
cc/resources/resource_pool_unittest.cc:238: return resource;
On 2015/11/09 19:32:51, danakj wrote:
> don't you need to erase from from unused_ here?

No. Basically test is -

1. create n resources
2. Append those to unused
3. Then actual test of traversing resources from unused_. (which is done 1m
times). Here the cost is computed for the resource which would not be there at
all, e.g. resource of size 1024 x 1024.

In this test Acquire() is used for creating and traversing purpose only.

4. Finally removing all.

https://codereview.chromium.org/1410833005/diff/20001/cc/resources/resource_p...
cc/resources/resource_pool_unittest.cc:326: tdq.AppendToUnused(resource[i]);
It was not intentional. I thought creating and appending mechanisms are  needed
in both tests so kept there. But I did not add the test for those. Basically for
creating, there is no cost, but for appending to queue, multi-deque is costlier
than deque, (cost of finding the deque of required size is extra).

As suggested by you, we can measure the cost of traversing only, as that is the
major concern to be solved by multi-deque.

[prashant.n, 2015-11-10 05:41:09]

Updated results in bug, on the latest patch set.

[danakj, 2015-11-10 22:12:06]

https://codereview.chromium.org/1410833005/diff/20001/cc/resources/resource_p...
File cc/resources/resource_pool_unittest.cc (right):

https://codereview.chromium.org/1410833005/diff/20001/cc/resources/resource_p...
cc/resources/resource_pool_unittest.cc:238: return resource;
On 2015/11/10 04:10:50, prashant.n wrote:
> On 2015/11/09 19:32:51, danakj wrote:
> > don't you need to erase from from unused_ here?
> 
> No. Basically test is -
> 
> 1. create n resources
> 2. Append those to unused
> 3. Then actual test of traversing resources from unused_. (which is done 1m
> times). Here the cost is computed for the resource which would not be there at
> all, e.g. resource of size 1024 x 1024.
> 
> In this test Acquire() is used for creating and traversing purpose only.

But the actual acquire does erase.

> 4. Finally removing all.

[prashant.n, 2015-11-11 03:04:05]

> 
> But the actual acquire does erase.
> 

In worst case if we don't get any resource, we don't erase. If in worst case MDQ
is faster then definitely if we get the element it is faster. Because the cost
is for finding the queue for the required size and erasing last element from the
queue found.