www.chriscoffee.com: quality & service, second to none

Slow Load? - Page 2

Postby HB on Tue Oct 09, 2007 11:42 pm

From my corner of the world, the performance is much better. To collect some objective data, I've added basic statistics of the site's last hour to the bottom right of the page:

overall speed: OK*, current page: 0.68s, avg: 1.5s, max: 8.3s

The overall speed criteria are:
  • fast - 85% of the pages are returned in 1.5 seconds or less
  • OK - 80% of the pages are returned in 3 seconds or less
  • slow - 80% of the pages are returned in 6 seconds or less
  • very slow - the site is slower than "slow" :shock:
I've noticed some spikes of slow performance. This is indicated by an "*" next to the overall status whenever more than 3% of the pages are returned in more than 6 seconds. For example, "OK*" means 80% of the pages were returned in 3 seconds or less, but there's a small segment of the membership potentially experiencing timeouts.
Dan Kehn
User avatar
HB
 
Posts: 13170
Joined: Apr 29, 2005
Location: Cary, NC

Postby cannonfodder on Wed Oct 10, 2007 11:11 pm

Don't they have a "coffeegeek" slow rating? :lol:

My observations, much faster responses. The only delays are on threads with lots of graphics.
Dave Stephens
User avatar
cannonfodder
Team HB
 
Posts: 6812
Joined: May 23, 2005
Location: Downingtown PA

Postby HB on Wed Oct 10, 2007 11:46 pm

To be fair, although CoffeeGeek suffered for more than a year with deplorable response times, it's zippy these days.

Anyway, the graphics should be fast since they're static content. But I think there's a problem with perceived slowness caused by the dynamic reformatting. As phpBB is written, the size of uploaded images are not stored with the thread text. So when a thread's HTML loads, the space reserved for a given image is a tiny block, which blooms out when the actual image arrives. If a thread has lots of images, the effect cascades as the images are loaded. If the image sizes were statically defined, the HTML would flow more smoothly (the "current page" time at the bottom of the page represents the HTML generation time, not the image loading time, which depends dramatically on the user's browser settings, cache, connection speed, etc.). But I think the images look better at their actual size than fixed thumbnail dimensions.

I'll have more hard data in a week, but all early indicators say the server upgrade was worth the trouble.
Dan Kehn
User avatar
HB
 
Posts: 13170
Joined: Apr 29, 2005
Location: Cary, NC

Postby HB on Sun Nov 11, 2007 12:14 pm

Below is the post-mortem data on the site's overall performance. It's obvious when the new server came online, and why it was necessary. The trend from August through early October was worrisome; I figured the site would really crawl during the Holiday Wishlist Gift run without faster hardware.

Image

I also added some server-side calculations of image sizes to help the layout of dynamic pages. Now the page refresh appears much faster because the images aren't "fluttering" into place, causing a bunch of redraws. That was a side benefit of the switch to images hosted on HB's server (versus the offsite [img] of the past).

As always, I'll keep my eye on the site performance, but for now, it's "wicked fast" at least 80% of the time from this corner of the world. :D
Dan Kehn
User avatar
HB
 
Posts: 13170
Joined: Apr 29, 2005
Location: Cary, NC

Postby HB on Sat Mar 01, 2008 10:47 am

A few months later and again I'm seeing spotty performance:

overall speed: slow, current page: 1.3s, avg: 2.9s, max: 18.0s

Once the beta of V3 goes live, I'll look into upgrading the server.
Dan Kehn
User avatar
HB
 
Posts: 13170
Joined: Apr 29, 2005
Location: Cary, NC

Previous

Return to News and Suggestion Box