Rackspace’s Mosso Hosting Cloud Review

Mosso is Rackspace’s (NYSE: RAX) hosting cloud which is a cloud computing platform for hosting websites and web applications. I tested Mosso’s ability to respond to load and scale automatically, which is an important, definitive feature of a cloud computing platform. Some cloud computing platforms require that the user intervene or program their own process for automatically scaling applications to capacity. Users only pay for the computing resources used and hosting plans start at $100 per month. I found Mosso to be an all around great cloud computing platform for hosting as the load testing data reflects. Mosso supports a wide variety of web applications for deployment on their cloud such as PHP, Ruby on Rails, Perl, Python, and .NET applications.

The Mosso Administration Console

How was the Mosso Cloud’s Performance evaluated?

The Application:
PHP-based e-commerce web application: Oscommerce 2.2rc2a default installation
MySQL 5

Load Testing Environment Snapshot:
Load Testing Generator
AMD 64 Athlon 3200+
2GB RAM
Load Generated Apache-Bench
100Mbit dedicated WAN

Load Testing Measurements
Apache-Bench Metrics from the Load Generator
Tamper Data Firefox Plugin for gathering End-user Performance Metrics
Alertra HTTP Check for Site Monitoring

The review was conducted primarily on independent performance tests using a simple load testing environment while hitting a single URL that contained a lot of results. Also, my overall experience while using the service is taken into account, and it’s viability as a true cloud computing platform are also taken into consideration during this review.
Load was generated over the public Internet against my Mosso cloud hosted site using a 100Mbps dedicated WAN link and a lot of apache-bench processes primarily. The test was intended to be simple and simulate various “flash-crowd” scenarios that a production website or startup company might encounter when becoming popular or showcasing popular content. Given the small amount of time and eventually the high load put on the website towards the end of the test (10,000 concurrent users) the Mosso hosted example site, purewebcraft.com, held up exceptionally well.

NOTE: Please ignore the metric that says “failed requests” this is a known issue with apache bench and dynamic content because the size of the page is differing from one request to another.

Alertra was used for monitoring server status. In case the load tests caused an outage of the site, I wanted to know when it happened. Alertra was used to measure the overall availability during all of the load tests and was taken as an aggregate, independent metric. Aletra monitoring data was not factored into the individual test suite results.

Tamper data was used to measure the response times of the site from the end-user perspective so that I had user experience metrics in addition to those metrics generated by apache-bench. The baseline results (no load being generated) for end user response times are as follows:

Cloud Platform Testing Results

Test Suite 1 – 500 Concurrent User
Load Generator Network Utilization/Saturation: ~11%
Apache Bench (ab) Command: ab -n 10000 –c 500 http://www.purewebcraft.com/oscommerce/catalog/index.php?cPath=3_10

Apache Bench Metrics

Server Software: Apache/2.2.3
Server Hostname: www.purewebcraft.com
Server Port: 80
Document Path: /oscommerce/catalog/index.php?cPath=3_10
Document Length: 32896 bytes

Concurrency Level: 500
Time taken for tests: 417.875 seconds
Complete requests: 10000
Failed requests: 9724
(Connect: 0, Receive: 0, Length: 9724, Exceptions: 0)
Write errors: 0
Non-2xx responses: 63
Total transferred: 332399719 bytes
HTML transferred: 327402177 bytes
Requests per second: 23.93 [#/sec] (mean)
Time per request: 20893.750 [ms] (mean)
Time per request: 41.788 [ms] (mean, across all concurrent requests)
Transfer rate: 776.81 [Kbytes/sec] received

Connection Times (ms)
min mean[+/-sd] median max
Connect: 31 41 8.6 47 281
Processing: 1750 20193 2656.7 20641 31406
Waiting: 219 10623 5606.2 10641 30406
Total: 1797 20234 2656.8 20688 31438

Percentage of the requests served within a certain time (ms)
50% 20688
66% 20719
75% 20766
80% 20813
90% 20984
95% 21563
98% 22406
99% 22422
100% 31438 (longest request)

Tamper Data (User Experience Metrics)

Mosso Usage Meter Snapshot Test Suite 1

Mosso Usage View Before Test Suite 1

Mosso Usage View After Test Suite 1

Test Suite 2 – 1000 Concurrent User
Load Generator Network Utilization/Saturation: ~10-15%
Apache Bench (ab) Command: ab -n 10000 –c 1000 http://www.purewebcraft.com/oscommerce/catalog/index.php?cPath=3_10

Apache Bench Metrics

Server Software: Apache/2.2.3
Server Hostname: www.purewebcraft.com
Server Port: 80

Document Path: /oscommerce/catalog/index.php?cPath=3_10
Document Length: 32991 bytes

Concurrency Level: 1000
Time taken for tests: 422.938 seconds
Complete requests: 10000
Failed requests: 9752
(Connect: 0, Receive: 0, Length: 9752, Exceptions: 0)
Write errors: 0
Non-2xx responses: 302
Total transferred: 324428254 bytes
HTML transferred: 319531591 bytes
Requests per second: 23.64 [#/sec] (mean)
Time per request: 42293.750 [ms] (mean)
Time per request: 42.294 [ms] (mean, across all concurrent requests)
Transfer rate: 749.10 [Kbytes/sec] received

Connection Times (ms)
min mean[+/-sd] median max
Connect: 31 41 9.4 47 344
Processing: 750 39996 8067.0 40938 51281
Waiting: 297 19864 11853.9 19719 50438
Total: 797 40037 8067.2 40984 51313

Percentage of the requests served within a certain time (ms)
50% 40984
66% 41391
75% 41828
80% 43531
90% 46734
95% 46797
98% 47438
99% 50656
100% 51313 (longest request)

Tamper Data (User Experience Metrics)

Mosso Usage Meter Snapshot Test Suite 2

Mosso Usage View Before Test Suite 2

Mosso Usage After Run 2

Test Suite 3 – 10000 Concurrent User
Load Generator Network Utilization/Saturation: ~20-30%
Apache Bench (ab) Command: (10x Processes) ab -n 10000 –c 1000 http://www.purewebcraft.com/oscommerce/catalog/index.php?cPath=3_10

Apache Bench Metrics (multiplied by number of processes)

Server Software:
Server Hostname: www.purewebcraft.com
Server Port: 80

Document Path: /oscommerce/catalog/index.php?cPath=3_10
Document Length: 25 bytes

Concurrency Level: 1000
Time taken for tests: 468.859 seconds
Complete requests: 10000
Failed requests: 1628
(Connect: 0, Receive: 0, Length: 1628, Exceptions: 0)
Write errors: 0
Non-2xx responses: 8372
Total transferred: 55308213 bytes
HTML transferred: 53845795 bytes
Requests per second: 21.33 [#/sec] (mean)
Time per request: 46885.938 [ms] (mean)
Time per request: 46.886 [ms] (mean, across all concurrent requests)
Transfer rate: 115.20 [Kbytes/sec] received

Connection Times (ms)
min mean[+/-sd] median max
Connect: 31 45 106.7 47 3031
Processing: 2438 42814 22461.0 36859 136953
Waiting: 47 21016 10000.0 20797 68719
Total: 2484 42859 22467.5 36906 136984

Percentage of the requests served within a certain time (ms)
50% 36906
66% 38703
75% 40547
80% 41594
90% 86297
95% 100281
98% 108328
99% 115328
100% 136984 (longest request)

Tamper Data (User Experience Metrics)

Mosso Usage Meter Snapshot Test Suite3

Mosso Usage View Before Test Suite 3

Mosso Usage After Test Suite 3

Server Availability and Performance

The site had no problem staying up. Not once was service interrupted or availability affected as reflected in the monitoring report.

Although there appeared to be a slightly noticeable lag time once I hit the 10,000 concurrent user mark, this should be expected as the bandwidth usage was 1Mbps or 8 Megabytes per second, which turns out to be 675 Gigabytes per day of traffic. The fact I tested in short bursts must be taken into consideration as the Mosso system may need a little more time than 5 minutes to scale the capacity up to such a massive amount of traffic. Although at the maximum benchmark response times began to slow down to less than desireable levels, it was still serving traffic to users and service availability was not impacted.

Aggregate Monitoring Report During all Test Suites

Overall Summary

Pro’s
- Easy to deploy and scale applications
- Automatic scaling in capacity of applications, no manual application deployments to meet demand
- Supports PHP,Ruby on Rails, Perl, Python, .NET, MySQL, SQL Server
- The option to manage and bill client websites from a single console
- If a client need support with the Mosso hosted site, Mosso provides tech support
- Well-organized and relevant knowledge base
- Automatic scaling of database, storage and bandwidth

Con’s
- Access to the forums must be manually enabled by a Mosso representative
- Java web applications are not supported
- Ruby on Rails support is limited to one site before incurring additional cost
- No API, but it is in the works
- Minimum monthly fee of $100 per month

Conclusions

Mosso’s cloud scales extremely well and with absolutely no user intervention, which allows it to claim the title of a true cloud computing platform. I found the cost to be reasonable, provided that the user would use at least the computing resources included in the $100 minimum monthly access fee, but any application that would require a cloud computing platform will be expected to scale beyond this and should not be a problem for most cloud shoppers. For Uncle Jerry’s Golf Blog, it will definitely be better to go with a simple shared hosting plan.

Although it was not mentioned whether or not the Mosso cloud is geographically dispersed, I would not doubt it given Rackspace’s massive computing resources and expanse of data centers.

Mosso makes the grade and lives up to it’s claims.

My Rating: A-
User Ratings: [flash http://thebitsource.com/ratings/rating_id.swf?id=102 p={id;102|embed;102}]