[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[orion-dev] UX: Performance - Page Speed Analysis
We ended up talking performance at yesterday's UX meeting. I took a task to run page speed on some of Orion's pages. I ran these tests on orion.eclipse.org.
I ran the first set late last night from home on my MacBook. It's a machine from last year - off the shelf, not slow, but not the fastest computer either. My internet connection at home is wireless.
Page Speed was unable to capture the initial load time after logging in. I counted 1-second, 2-second and got to 24 seconds human experience time after logging in the first time before seeing the Navigator page.
Bug 381063 - Initial login to Orion takes way too long
On my iPad this morning I reached 36 seconds - with no feedback at all during that time, but for a brief glimpse of the browser progress wheel. On my iPad the login page took 6 seconds to load.
Simon showed me yesterday that my use of Orion ( in using my hosted site as my dev tool ) was not the best way of using it - that I should use orion.eclipse.org for development. So I based measurements on that - most of observations prior to that were based on anton.orion.eclipse.org.
Here's a summary of each of the pages I accessed afterwards, in general it didn't feel too bad.
* I have improvements for the Login - Bug 381064 - Login page assets can be optimized here's my data with those improvements from my computer this morning:
My feeling is that this is not a constant - and that sometimes page loads can take much longer. I don't know if that feeling is right or wrong. So here are some measurements made from my ThinkPad this morning, which show a slower variation in time. Also the first time I accessed the editor page I counted seconds again and it took 7 seconds in human time. Faster after that at just over three seconds. 9 seconds ( white screen ) for Repositories, 10 seconds for Git Status. Page Speed isn't able to capture those first time experiences - I can only run it when the page has loaded.
I've made general bugs for the Navigator, Editor and Git Status page peformances and attached the json data that I have.
Bug 381066 - Navigator page speed
Bug 381067 - Editor Page Speed
Bug 381068 - Git Status Page Speed
You can find Page Speed information here:
In my experience it is best used in FireFox as the FF version lets you export the data. This is not human experience data, but machine data. It's a starting point. I'll run the tests again next week and make human experience measurements then too.
- I don't know why it is slower at work this morning than it was from home last night. Reasons that I can suppose - different network, more users, slower computer, different time, different geographic location. The transfer and request counts are very similar of course.
- Latency - the more requests the more latency. Latency is a big factor. If you can imagine if you were geographically more distant from the server the tax on those requests could be higher because each has a bigger network to travel through. So keeping the requests to a minimum is better - also latency comes into play in the device as well as geography - so a mobile phone from a rural access point say somewhere in Ireland ( I'm Irish ) may be significantly slower than my MacBook in Ottawa. I'd like to ask Susan and Szymon to run Page Speed on orion.eclipse.org to see their data - so we could develop a take on Latency.
- Size - some of the pages are moderately sized - getting on for half a megabyte. Maybe I'm old fashioned, but a megabyte seems a threshold to me for a web page. Maybe I need to wake up to this century a bit :)