I've split up the Media Viewer limn dashboard into a global one and one dashboard per pilot site. We'll shortly be adding more pilot site dashboards. For an up-to-date list of Media Viewer's limn dashboards, visit https://www.mediawiki.org/wiki/Multimedia/Metrics
Also in this update were new actions (no data yet since the change to record those actions just hit master) and key percentiles, following Nuria's recommendation, which I was able to generate with MariaDB thanks to a horrid trick described here: http://web.performancerasta.com/metrics-tips-calculating-95th-99th-or-any-pe...
This looks really good -- nice presentation of the graphs. And that really is a horrid SQL trick!
-Toby
On Tue, Apr 22, 2014 at 4:48 PM, Gilles Dubuc gilles@wikimedia.org wrote:
I've split up the Media Viewer limn dashboard into a global one and one dashboard per pilot site. We'll shortly be adding more pilot site dashboards. For an up-to-date list of Media Viewer's limn dashboards, visit https://www.mediawiki.org/wiki/Multimedia/Metrics
Also in this update were new actions (no data yet since the change to record those actions just hit master) and key percentiles, following Nuria's recommendation, which I was able to generate with MariaDB thanks to a horrid trick described here: http://web.performancerasta.com/metrics-tips-calculating-95th-99th-or-any-pe...
Analytics mailing list Analytics@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/analytics
Gilles,
This is pretty awesome. Thanks for sharing :-)
Best, Alolita
Alolita Sharma आलोलिता शर्मा Director of Engineering Internationalization & Localization Wikimedia Foundation
On Tue, Apr 22, 2014 at 5:02 PM, Toby Negrin tnegrin@wikimedia.org wrote:
This looks really good -- nice presentation of the graphs. And that really is a horrid SQL trick!
-Toby
On Tue, Apr 22, 2014 at 4:48 PM, Gilles Dubuc gilles@wikimedia.orgwrote:
I've split up the Media Viewer limn dashboard into a global one and one dashboard per pilot site. We'll shortly be adding more pilot site dashboards. For an up-to-date list of Media Viewer's limn dashboards, visit https://www.mediawiki.org/wiki/Multimedia/Metrics
Also in this update were new actions (no data yet since the change to record those actions just hit master) and key percentiles, following Nuria's recommendation, which I was able to generate with MariaDB thanks to a horrid trick described here: http://web.performancerasta.com/metrics-tips-calculating-95th-99th-or-any-pe...
Analytics mailing list Analytics@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/analytics
Analytics mailing list Analytics@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/analytics
Thank you, Gilles!
This is a really, really helpful, particularly the overall network performance data broken by percentiles.
If I understand this correctly, it looks like we’re well within our goals for images shown in Media Viewer: http://multimedia-metrics.wmflabs.org/graphs/mmv_performance_image_global
This global graph suggests that most images load within a second up to the 50th percentile, and only in the 90th percentile do we start seeing 5-second load times, which seems acceptable. Should we be adding API loads or any other data to that number, to get the complete load times?
I’m also impressed by the comparison of the time it takes to open an image with Media Viewer and on the File: page: http://multimedia-metrics.wmflabs.org/dashboards/mmv#media_viewer_vs_file_pa...
Could you share your views on why the File page mean (~3 seconds) appears to take longer than the Media Viewer load (~2.6) on a cold cache? I would have expected the reverse to be true, since Media Viewer is loading larger images files, plus the thumbnails.
And I love that you broke down the dashboards for different sites, which will make it a lot easier for our community champions to track the data for each project.
Nicely done, my friend — we all send you a big round of virtual applause for this important breakthrough :)
All the best,
Fabrice
On Apr 22, 2014, at 5:08 PM, Alolita Sharma asharma@wikimedia.org wrote:
Gilles,
This is pretty awesome. Thanks for sharing :-)
Best, Alolita
Alolita Sharma आलोलिता शर्मा Director of Engineering Internationalization & Localization Wikimedia Foundation
On Tue, Apr 22, 2014 at 5:02 PM, Toby Negrin tnegrin@wikimedia.org wrote: This looks really good -- nice presentation of the graphs. And that really is a horrid SQL trick!
-Toby
On Tue, Apr 22, 2014 at 4:48 PM, Gilles Dubuc gilles@wikimedia.org wrote: I've split up the Media Viewer limn dashboard into a global one and one dashboard per pilot site. We'll shortly be adding more pilot site dashboards. For an up-to-date list of Media Viewer's limn dashboards, visit https://www.mediawiki.org/wiki/Multimedia/Metrics
Also in this update were new actions (no data yet since the change to record those actions just hit master) and key percentiles, following Nuria's recommendation, which I was able to generate with MariaDB thanks to a horrid trick described here: http://web.performancerasta.com/metrics-tips-calculating-95th-99th-or-any-pe...
Analytics mailing list Analytics@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/analytics
Analytics mailing list Analytics@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/analytics
Multimedia mailing list Multimedia@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/multimedia
_______________________________
Fabrice Florin Product Manager Wikimedia Foundation
This global graph suggests that most images load within a second up to the 50th percentile, and only in the 90th percentile do we start seeing 5-second load times, which seems acceptable. Should we be adding API loads or any other data to that number, to get the complete load times?
Instead of doing a straight addition of the existing data, I would prefer to find time this week to add a new metric measuring time from first thumb click to sharp image appearing, as part of this card of mine which has been left on the backburner for a while: https://wikimedia.mingle.thoughtworks.com/projects/multimedia/cards/430 I might not do preloading on proximity, but I'll at least do hover.
Repeating some information I brought up orally during yesterday's team meeting, for list readers' benefit:
Could you share your views on why the File page mean (~3 seconds) appears
to take longer than the Media Viewer load (~2.6) on a cold cache? I would have expected the reverse to be true, since Media Viewer is loading larger images files, plus the thumbnails.
What the "versus" test sets out to compare is the following, measured visiting mediawiki.org on a cloudbees instance:
Given an article where the page is fully loaded, thumbnails included, we measure the time it takes to do these two things: - opening the file description page (File: page, usually on commons) and seeing the image fully displayed, as if the user command-clicked on the thumbnail to open the file description page in a new tab - clicking on the file's thumbnail and having Media Viewer load the image and fully display it
This essentially compares the pre-Media Viewer status quo, to the Media Viewer experience. Now, the reason why I think the file description page is taking so long can be counter-intuitive, because when you open one yourself, it feels like the page appears instantaneously. However, what really happens, and we easily become blind to, is that the tab/window for that page is actually blank for a considerable amount of time before the content appears almost all at once. The cause for this blank time, which is probably experienced on a lot of our production wiki pages, is that we have too much synchronous javascript and css in the <head> of the document. Even if these are pulled from the browser cache (as is the case in this "versus" test), it takes time for the browser to load them up and interpret them in their entirety before moving on to rendering the page. Media Viewer doesn't suffer from this problem, because opening it doesn't induce a new pageload. In terms of performance, not inducing a new pageload is Media Viewer's biggest performance edge against the status quo of opening the file description page.
Now, there are known issues at play that make our production wikis bad with the amount of things in the head, one of which is on our team's plate: Timed Media Handler putting a ridiculous amount of stuff in the head. And our engineers focusing on general mediawiki performance are always actively looking to reduce the amount of synchronous js/css in the head. We should expect that if we collectively do our homework and improve the <head> situation, the file description page's performance should improve over time, and this "versus" test will actually let us see that.
On Wed, Apr 23, 2014 at 2:31 AM, Fabrice Florin fflorin@wikimedia.orgwrote:
Thank you, Gilles!
This is a really, really helpful, particularly the overall network performance data broken by percentiles.
If I understand this correctly, it looks like we’re well within our goals for images shown in Media Viewer: http://multimedia-metrics.wmflabs.org/graphs/mmv_performance_image_global
This global graph suggests that most images load within a second up to the 50th percentile, and only in the 90th percentile do we start seeing 5-second load times, which seems acceptable. Should we be adding API loads or any other data to that number, to get the complete load times?
I’m also impressed by the comparison of the time it takes to open an image with Media Viewer and on the File: page:
http://multimedia-metrics.wmflabs.org/dashboards/mmv#media_viewer_vs_file_pa...
Could you share your views on why the File page mean (~3 seconds) appears to take longer than the Media Viewer load (~2.6) on a cold cache? I would have expected the reverse to be true, since Media Viewer is loading larger images files, plus the thumbnails.
And I love that you broke down the dashboards for different sites, which will make it a lot easier for our community champions to track the data for each project.
Nicely done, my friend — we all send you a big round of virtual applause for this important breakthrough :)
All the best,
Fabrice
On Apr 22, 2014, at 5:08 PM, Alolita Sharma asharma@wikimedia.org wrote:
Gilles,
This is pretty awesome. Thanks for sharing :-)
Best, Alolita
Alolita Sharma आलोलिता शर्मा Director of Engineering Internationalization & Localization Wikimedia Foundation
On Tue, Apr 22, 2014 at 5:02 PM, Toby Negrin tnegrin@wikimedia.orgwrote:
This looks really good -- nice presentation of the graphs. And that really is a horrid SQL trick!
-Toby
On Tue, Apr 22, 2014 at 4:48 PM, Gilles Dubuc gilles@wikimedia.orgwrote:
I've split up the Media Viewer limn dashboard into a global one and one dashboard per pilot site. We'll shortly be adding more pilot site dashboards. For an up-to-date list of Media Viewer's limn dashboards, visit https://www.mediawiki.org/wiki/Multimedia/Metrics
Also in this update were new actions (no data yet since the change to record those actions just hit master) and key percentiles, following Nuria's recommendation, which I was able to generate with MariaDB thanks to a horrid trick described here: http://web.performancerasta.com/metrics-tips-calculating-95th-99th-or-any-pe...
Analytics mailing list Analytics@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/analytics
Analytics mailing list Analytics@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/analytics
Multimedia mailing list Multimedia@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/multimedia
Fabrice Florin Product Manager Wikimedia Foundation
http://en.wikipedia.org/wiki/User:Fabrice_Florin_(WMF)
Multimedia mailing list Multimedia@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/multimedia
On Thu, Apr 24, 2014 at 3:27 AM, Gilles Dubuc gilles@wikimedia.org wrote:
This global graph suggests that most images load within a second up to the
50th percentile, and only in the 90th percentile do we start seeing 5-second load times, which seems acceptable. Should we be adding API loads or any other data to that number, to get the complete load times?
Instead of doing a straight addition of the existing data, I would prefer to find time this week to add a new metric measuring time from first thumb click to sharp image appearing, as part of this card of mine which has been left on the backburner for a while: https://wikimedia.mingle.thoughtworks.com/projects/multimedia/cards/430 I might not do preloading on proximity, but I'll at least do hover.
Can we also log when the domready event and the onload event of the page happens relative to the click on the thumbnail? As I mentioned in the thread about slow loading [1], we have a problem on slow connections with pages that contain lots of thumbnails, and it would be nice to know how widespread an issue this is. The simplest way I can think of to measure this sort of performance problem is to compare the time passed from click to thumbnail display and the time passed from click to page onload event.
[1] http://lists.wikimedia.org/pipermail/multimedia/2014-April/000285.html
Filed: https://wikimedia.mingle.thoughtworks.com/projects/multimedia/cards/515
Time between click after page load/replayed click and image load will be measured by https://wikimedia.mingle.thoughtworks.com/projects/multimedia/cards/508
On Thu, Apr 24, 2014 at 8:27 PM, Gergo Tisza gtisza@wikimedia.org wrote:
On Thu, Apr 24, 2014 at 3:27 AM, Gilles Dubuc gilles@wikimedia.orgwrote:
This global graph suggests that most images load within a second up to
the 50th percentile, and only in the 90th percentile do we start seeing 5-second load times, which seems acceptable. Should we be adding API loads or any other data to that number, to get the complete load times?
Instead of doing a straight addition of the existing data, I would prefer to find time this week to add a new metric measuring time from first thumb click to sharp image appearing, as part of this card of mine which has been left on the backburner for a while: https://wikimedia.mingle.thoughtworks.com/projects/multimedia/cards/430I might not do preloading on proximity, but I'll at least do hover.
Can we also log when the domready event and the onload event of the page happens relative to the click on the thumbnail? As I mentioned in the thread about slow loading [1], we have a problem on slow connections with pages that contain lots of thumbnails, and it would be nice to know how widespread an issue this is. The simplest way I can think of to measure this sort of performance problem is to compare the time passed from click to thumbnail display and the time passed from click to page onload event.
[1] http://lists.wikimedia.org/pipermail/multimedia/2014-April/000285.html
Multimedia mailing list Multimedia@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/multimedia