Magento came to us with a challenge, they asked us if Magento is stripped down to a basic store, how well does it perform with and without CloudMaestro? We took the challenge! Here are the findings that surprised us.
2. TEST GOALS
Show performance results for each of the Magento store
scenarios* with and without Lagrange Systems’
CloudMaestro
Demonstrate the positive impact of a distributed solution
* Magento store scenarios were provided by eBay/Magento and were ‘vanilla’ stores
(no design or assets, only code)
3. WHAT IS SITE PERFORMANCE?
The ability to seamlessly scale servers in and out
without any impact to end-users, while maintaining
optimal site speed during a traffic influx.
4. REFERENCE ARCHITECTURE
ENTERPRISE
2 x 60GB app server, 1 x
30db server*
Scaled from 4 to 12 servers
Total: 48 Cores, 180GB
RAM
25k to 120k users/hour
55.1 to 224.7 page views
per second
Small Business
Infrastructure
Medium Business Infrastructure
Enterprise Infrastructure
MEDIUM
2 x 8GB app server, 1 x
8GB db server
Scaled from 2 to 8 servers
Total: 32 Cores, 32GB RAM
10k to 90k users/hour
22.1 to 198.5 page views
per second
SMALL
1 x 8GB app server, 1 x
8GB db server
Scaled from 1 to 4 servers
Total: 16 Cores, 16GB RAM
4k to 40k users/hour
8.8 to 88.2 page views per
second
* NOTE: For the ‘with Lagrange’ enterprise infrastructure
testing we started with 8 x 8GB servers and scaled out to
16 x 8GB servers in order to provide a more cost
effective (and highly available) solution
5. WHAT WAS MEASURED
Each testing combination of architecture
type and store size was measured for
- page load speed
- checkouts completed
- autoscaling by CloudMaestro to eliminate
downtime
6. TEST PROTOCOL
Option and Description % of Traffic
view_product_add_to_cart_percent:
Percentage of users which are browsing, adding items to cart
and abandoning the cart.
62%
view_catalog_percent:
Percentage of users who are just browsing the site.
30%
guest_checkout_percent:
Percentage of users who are browsing, adding items to a cart
and checking out as guests.
4%
customer_checkout_percent:
Percentage of users who are browsing, adding items to a cart
and checking out as registered customers.
4%
7. SMALL BUSINESS ARCHITECTURE
Jmeter Load Generation
Application Delivery Controller Tier
ADC ADC ADC
Rackspace Cloud
AS
Application Cluster
AS AS
Magento Store Environment
AS
AS
Application Scaling
Scenarios:
• without CloudMaestro, the
servers tested were
• 1 x 8GB (app)
• 1 x 8GB (db)
• With CloudMaestro, the
servers tested were
• 1 – 4 x 8GB (app)
• 1 x 8GB (db)
• 16 Cores, 16GB RAM
8. SMALL BUSINESS ARCH. RESULTS:
LOAD TEST TO FAILURE
Data where the Magento Store has
user checkouts beginning to timeout
Magento
Store
Scenario
Pageviews per
Second Arrival Rate
Average Page Load
Time without
Lagrange
Average Page Load
Time with
Lagrange
Small 35.3 16,000
users/hr
16.7 sec 1.2 sec
Medium 30.9 14,000
users/hr
13.1 sec 1.2 sec
Large 26.5 12,000
users/hr
11.5 sec 1.2 sec
NOTES:
• Testing stopped at 4 servers as the performance continued to stay the same for the ‘with Lagrange’ at 1.2sec
• With 4 servers, the application cluster was able to handle 40,000 users/hour or 88.2 page views per second
9. SMALL BUSINESS ARCH. RESULTS:
AVG. PAGE LOAD TIME
Without CloudMaestro
With CloudMaestro
10. SMALL BUSINESS ARCH. RESULTS:
CHECKOUTS COMPLETED
Without CloudMaestro
With CloudMaestro
11. MEDIUM BUSINESS ARCHITECTURE
Jmeter Load Generation
Application Delivery Controller Tier
ADC ADC ADC
Rackspace Cloud
AS
AS AS AS AS
Application Cluster
Magento Store Environment
AS AS
AS
AS
Application Scaling
Scenarios:
• without CloudMaestro, the
servers tested were
• 2 x 8GB (app)
• 1 x 8GB (db)
• With CloudMaestro, the
servers tested were
• 2 – 8 x 8GB (app)
• 1 x 8GB (db)
• 32 Cores, 32GB RAM
12. MEDIUM BUSINESS ARCH. RESULTS:
LOAD TEST TO FAILURE
Data where the Magento Store has
user checkouts beginning to timeout
Magento
Store
Scenario
Page views per
second Arrival Rate
Average Page Load
Time without
Lagrange
Average Page Load
Time with
Lagrange
Small 55.1 25,000 6.4 sec 0.9 sec
Medium 44.1 20,000 3.3 sec 0.4 sec
Large 44.1 20,000 3.3 sec 0.4 sec
NOTES:
• Testing stopped at 8 servers as the performance continued to stay the same for the ‘with Lagrange’ under 1 sec
• With 8 servers, the application cluster was able to handle 90,000 users/hour or 198.5 page views per second
13. MEDIUM BUSINESS ARCH. RESULTS:
AVG. PAGE LOAD TIME
Without CloudMaestro
With CloudMaestro
14. MEDIUM BUSINESS ARCH. RESULTS:
CHECKOUTS COMPLETED
Without CloudMaestro
With CloudMaestro
15. ENTERPRISE ARCHITECTURE
Rackspace Cloud
Scenarios:
• without CloudMaestro, the
servers tested were
• 2 x 60GB (app)
• 1 x 30GB (db)
• With CloudMaestro, the
servers tested were
• 4 – 12 x 8GB (app)
servers
• 1 x 30GB (db)
• 48 Cores, 180GB RAM
Jmeter Load Generation
Application Delivery Controller Tier
ADC ADC ADC
AS AS AS AS
Application Cluster
Magento Store Environment
AS
AS AS
AS
Application Scaling
AS
AS AS AS AS
16. ENTERRPISE ARCH. RESULTS:
LOAD TEST TO FAILURE
Data where the Magento Store has
user checkouts beginning to timeout
Magento
Store
Scenario
Page views per
second Arrival Rate
Average Page Load
Time without
Lagrange
Average Page Load
Time with
Lagrange
Small 97 44,000 4.6 sec .34 sec
Medium 97 44,000 7.1 sec .54 sec
Large 88.2 40,000 5.1 sec .61 sec
NOTES:
• Testing stopped at 12 servers as the performance continued to stay the same for the with ‘Lagrange’ under 1 sec
• With 12 servers, the application cluster was able to handle 120k users/hour or 264.7 page views per second
19. TEST TAKEAWAYS
19
Lagrange Systems’ CloudMaestro is able to
demonstrate significant performance improvement to
variously sized Magneto stores
In all cases without CloudMaestro, the Magento
stores had timeouts for checkouts at low user arrival
rates, that would negatively impact merchant revenue
Through auto-scaling, CloudMaestro was able to
provide consistently low page load times throughout
all arrival rate scenarios and process all checkout
activity
20. NOW….TO SUMMARIZE
In all testing scenarios, Magento eCommerce sites
experienced lower page load times, faster and more
checkouts while eliminating timeouts when used with
CloudMaestro