Web Performance - random lessons learnt from delivering WPO, Load testing and APM consulting in the UK. PLus a bit about WebPageTest Private Instances etc
1 of 32
More Related Content
London web performance WPO Lessons from the field June 2013
1. CloudOpsGuys Your Software | Our Service TM
1
CloudOpsGuys
Your Software | Our Service TM
WPO Lessons from the Field
2. CloudOpsGuys Your Software | Our Service TM
(1) Some quick & dirty ways to do a WPO
PoC using automation
(2) 5 reasons why having your own
WebPageTest instance rocks!
(3) The most common WPO mistakes I see
(time and time again)
(4) Why I love APM tools and you're nuts if
you don't use one!
Agenda
3
3. CloudOpsGuys Your Software | Our Service TM
PoC - Do a quick example video
There are two easy ways to get a rough
before/after web optimisation video
Using WebPageTest Compare
http://www.webpagetest.org/compare
Use PageSpeedInsights
https://developers.google.com/speed/pagespeed/in
sights
(yes, I know one uses the other!)
5. CloudOpsGuys Your Software | Our Service TM
And you can watch a video
https://developers.google.com/speed/pagespeed/insights#url=http_3A_2F_2Fwww.marksandspencer.com_2FFormal
-Shirts-Shirts-Mens_2Fb_2F2559784031_3Fie_3DUTF8_26intid_3Dgnav__men__formal-shirts&mobile=false
8. CloudOpsGuys Your Software | Our Service TM
Build your own
Ive been building my own PoC tool using
Riverbed Stingray Aptimizer and aiCache.com
Cache static and dynamic optimise resources
Just use CNAMES and some header re-
writing to glue it together
Browser
aiCache
Dynamic
Stingray
Aptimizer
Origin
Server
9. CloudOpsGuys Your Software | Our Service TM
Some gotchas
Multiple levels of caching
Browser cache
aiCache
Aptimizer cache (optimized static objects)
Making sure that you send the right headers at each
hop
Right user-agent
Right hostname
Right GZIP / Deflate headers
FQDN names in source (particularly in top/side nav)
FFS whats wrong with a relative /link?
10. CloudOpsGuys Your Software | Our Service TM
Other lessons
Load time is useless
Render start is useless
Filmstrip, Video or Visual Progress is KEY
More on this later
12. CloudOpsGuys Your Software | Our Service TM
5 reasons why WPT Private
instance rocks!
1. NO QUEUE!
2. Bulk Testing Mode
3. Its private
4. Its inside your firewall
5. Easier access to history of testing
6. You will learn more about features!
1. Visual Progress
2. Connection View
3. Scripting etc etc
19. CloudOpsGuys Your Software | Our Service TM
Lessons learnt from consulting
20/06/2013 19
Generic Description
Reduce Page
Size
Reduce Number of
Requests
Implement
GZIP
Compression
Optimise
Loading of
external files
Defer
JavaScript
loading
Optimise
images
Optimize
HTTP Caching
Domain
Sharding
Cookie-
free
Domains
Back-end
performance
tuning
Remove
Unnecessary
CSS
High Street Retailer Yes Yes No No No No Yes No Yes No Yes
Budget Hotel Chain No Yes No No No Yes Yes Yes Yes Yes No
Professional Body No Yes Yes Yes Yes Yes Yes Yes No Yes No
High Street Retailer No Yes No Yes Yes Yes Yes Yes Yes Yes No
VOD provider No Yes No No Yes Yes Yes Yes Yes Yes Yes
Consumer Electronics Brand Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Fashion Retailer No Yes No No Yes Yes Yes No Yes Yes Yes
Travel Sector No Yes Yes Yes Yes no Yes Yes Yes Yes No
Fashion Retailer Yes Yes Yes Yes Yes No Yes Yes No Yes No
Fashion Retailer Yes Yes No No Yes No Yes Yes Yes Yes No
Home Improvement Retail No Yes No No Yes No Yes Yes Yes Yes Yes
Electrical Retailer Yes Yes no Yes Yes No Yes No Yes Yes No
Electrical Retailer Yes Yes No No Yes Yes Yes Yes No Yes No
Online Gambling Yes Yes Yes Yes No Yes Yes No Yes Yes Yes
Mobile Phone Retailer Yes Yes No Yes Yes Yes Yes Yes Yes No Yes
Insurance Provider No Yes No No Yes No Yes Yes No Yes Yes
Emergency Repair Service No Yes No Yes No Yes Yes Yes No Yes Yes
Mobile Phone Retailer Yes Yes No Yes Yes Yes Yes Yes No Yes Yes
Car Manufacturer No Yes No No No Yes Yes Yes Yes Yes Yes
Financial Services No Yes Yes No Yes No Yes Yes No Yes Yes
Travel Sector Yes Yes No No Yes No Yes Yes No No No
Publisher No Yes Yes No Yes Yes Yes Yes No No Yes
Publisher No Yes No No Yes No Yes Yes Yes No Yes
High Street Retailer Yes Yes No Yes Yes Yes Yes Yes No No Yes
Newspaper Yes Yes No Yes Yes Yes Yes No Yes Yes Yes
48% 100% 28% 48% 80% 60% 100% 80% 60% 76% 64%
20. CloudOpsGuys Your Software | Our Service TM20/06/2013 20
Performance Rule
Frequency
Recommended
Reduce Number of Requests 100%
Optimize HTTP Caching 100%
Defer JavaScript loading 80%
Domain Sharding 80%
-Back end performance tuning 76%
Remove Unnecessary CSS 64%
Optimise images 60%
-Cookie free Domains 60%
Reduce Page Size 48%
Optimise Loading of external
files
48%
Implement GZIP Compression 28%
21. CloudOpsGuys Your Software | Our Service TM
What else?
It takes ages for anything to change
(hence why I am focussing on automation stuff)
Videos are a killer tool
Analysis of synthetic monitoring can tell you
an lot about back-end performance
Use analogies (I wrote a whole whitepaper
using shopping at Tesco as an analogy for web
performance)
22. CloudOpsGuys Your Software | Our Service TM
WHY I LOVE APM TOOLS
Deeper insights into back-end performance
23. CloudOpsGuys Your Software | Our Service TM
4 quick examples
Know your dependencies!
And dont assume its your code that sucks
You can focus on application transactions
How many transactions to load a page?
And why ORMs can suck
And classic Application metrics
The heap is doing WHAT?
31. CloudOpsGuys Your Software | Our Service TM
THANKYOU
www.cloudopsguys.com
CloudOpsGuys
Your Software | Our Service TM
20
32. CloudOpsGuys Your Software | Our Service TM
Get in Touch www.CloudOpsGuys.com
0208 798 3079info@cloudopsguys.com
twitter.com/cloudopsguys
facebook.com/cloudopsguysskype: cloudopsguys
plus.google.com/cloudopsguys
Social Media Social Media
Editor's Notes
Based on 25 consulting reports across a wide range of e-commerce and publishing sites
100% of the time there were too many requests100% of the time they werent using caching to its full advantage, and hence repeat visits were slower than they needed to be76% of the sites exhibited noticable slowing down during peak time of day/week~50% of the time the pages were bloated (often due to images that on average can be optimised ~15%) Frighteningly 村 of the sites analysed either didnt have Gzip compression enabled or it had problems e.g. failing under load!