You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@ignite.apache.org by Arthur Naseef <ar...@apache.org> on 2022/08/03 23:10:59 UTC

Performance of cancelling services (related to IGNITE-17274)

Testing with 60,000+ services, found that cancelling large numbers of
services takes significant time.  Found another linear lookup that can be
optimized with a map.

Here is a commit that fixes the problem in local testing:

https://github.com/artnaseef/ignite/commit/4ca87f7f6e7aae193faf1404736e0b4ee7168752


The difference, on my local system, with and without the patch:

CANCEL 60,000 *WITHOUT PATCH* = 67s 358ms

CANCEL 60,000 *PATCHED* = 1s 323ms


Note that, with the IGNITE-17274 fix, startup of the same 60,000 services
only takes ~ 9s.

Appreciate any feedback, and I hope we can push this fix forward.  I can
create a new Jira ticket and pull request.

Art

Re: Performance of cancelling services (related to IGNITE-17274)

Posted by Arthur Naseef <ar...@amlinv.com>.
Thank you Pavel - I applied your recommended edit, created a ticket and
submitted the PR.


https://github.com/apache/ignite/pull/10196
https://issues.apache.org/jira/browse/IGNITE-17524

Cheers!

Art


On Thu, Aug 4, 2022 at 12:11 AM Pavel Tupitsyn <pt...@apache.org> wrote:

> The change looks good to me, good catch.
>
> >  I can create a new Jira ticket and pull request.
> Please do, let's merge it.
>
> On Thu, Aug 4, 2022 at 2:11 AM Arthur Naseef <ar...@apache.org> wrote:
>
> > Testing with 60,000+ services, found that cancelling large numbers of
> > services takes significant time.  Found another linear lookup that can be
> > optimized with a map.
> >
> > Here is a commit that fixes the problem in local testing:
> >
> >
> >
> https://github.com/artnaseef/ignite/commit/4ca87f7f6e7aae193faf1404736e0b4ee7168752
> >
> >
> > The difference, on my local system, with and without the patch:
> >
> > CANCEL 60,000 *WITHOUT PATCH* = 67s 358ms
> >
> > CANCEL 60,000 *PATCHED* = 1s 323ms
> >
> >
> > Note that, with the IGNITE-17274 fix, startup of the same 60,000 services
> > only takes ~ 9s.
> >
> > Appreciate any feedback, and I hope we can push this fix forward.  I can
> > create a new Jira ticket and pull request.
> >
> > Art
> >
>

Re: Performance of cancelling services (related to IGNITE-17274)

Posted by Pavel Tupitsyn <pt...@apache.org>.
The change looks good to me, good catch.

>  I can create a new Jira ticket and pull request.
Please do, let's merge it.

On Thu, Aug 4, 2022 at 2:11 AM Arthur Naseef <ar...@apache.org> wrote:

> Testing with 60,000+ services, found that cancelling large numbers of
> services takes significant time.  Found another linear lookup that can be
> optimized with a map.
>
> Here is a commit that fixes the problem in local testing:
>
>
> https://github.com/artnaseef/ignite/commit/4ca87f7f6e7aae193faf1404736e0b4ee7168752
>
>
> The difference, on my local system, with and without the patch:
>
> CANCEL 60,000 *WITHOUT PATCH* = 67s 358ms
>
> CANCEL 60,000 *PATCHED* = 1s 323ms
>
>
> Note that, with the IGNITE-17274 fix, startup of the same 60,000 services
> only takes ~ 9s.
>
> Appreciate any feedback, and I hope we can push this fix forward.  I can
> create a new Jira ticket and pull request.
>
> Art
>