You are on page 1of 6

Walled Garden Performance Test Results Purpose of Test

The purpose of the performance test is to verify that the new Walled Garden architecture can support peak loads of both desirable and undesirable traffic.

Details of the Test


Tester(s): Ram Nagarajan and John Kurceba We attempted to run the target load of 1500/sec HTTP requests to Walled Garden servers, 600/sec TCP requests to Walled Garden servers and 2000/sec HTTP requests to Activation web servers. This request rate was attempted for duration of one hour for the tests performed. Lower rates were attempted when the responses that were returned were not as expected.

Test Conditions for the Test


1. 2. 3. 4. A 100% (4100/sec) load attempt was performed on 01/12. A 50% (2050/sec) load attempt was performed on 01/13. A 20% (820/sec) load attempt was performed on 01/14. An 80% (3280/sec) load attempt was performed on 01/15.

All the above load was generated by two servers (50% load on each) using multiple threads and with a sleep time between transactions flat at 25ms. The HTTP load was sent with Keep-Alive set in the header to mimic browser functionality. The TCP load was sent with a 30 sec connection hold time as per the requirement of the test. The Walled Garden load (HTTP and TCP) was sent to the Walled Garden VIP 68.87.122.93 and the Activation HTTP load was sent to the Activation Web Server VIP 68.87.122.110. The HTTP load (both Walled Garden and Activation) are split into the following HTTP methods:
HTTP Breakdown HEAD 2% PUT 3%

POST GET

20% 75%

Results

Response Type Breakdown:


20% load HTT P (WG) 86% 14% 0% 0% HTTP (ActS vr) 82% 18% 0% 0% 50% load HTTP (ActS vr) 84% 16% 0% 0% 80% load HTTP (ActS vr) 88% 12% 0% 0% 100% load HTTP (ActS vr) 0% 0% 100% 0%

Response Type Activation null blank 302 Found

TCP 0% 23% 39% 38%

HTTP (WG) 78% 22% 0% 0%

TCP 0% 49% 35% 16%

HTTP (WG) 78% 22% 0% 0%

TCP 0% 58% 31% 11%

HTTP (WG) 71% 29% 0% 0%

TCP 0% 31% 6% 3%

Response Times (in seconds) were:


20% load HTTP (WG) Method HEAD GET Count 21584 81025 Avg. RT 0.05 1 0.05 Std. Dev. 0.50 6 0.52 Medi an 0.003 0.003 90th %ile 0.04 4 0.04 Count 53559 20156 Avg. RT 0.16 1 0.15 Std. Dev. 1.18 5 1.16 Medi an 0.003 0.003 90th %ile 0.11 4 0.11 Count 86527 32397 Avg. RT 0.24 2 0.25 Std. Dev. 2.34 3 2.74 Medi an 0.017 0.018 90th %ile 0.18 4 0.18 Count 10783 5 40506 Avg . RT 0.22 5 0.22 Std. Dev. 1.65 8 1.64 Medi an 0.021 0.021 90th %ile 0.201 0.203 50% load 80% load 100% load

POST PUT TCP Method TCP HTTP (ActSvr) Method HEAD GET POST PUT

6 21573 6 32424

3 0.05 3 0.05 4

2 0.53 0 0.53 8

0.003 0.003

4 0.04 4 0.04 4

18 53843 4 80389

9 0.15 8 0.15 4

3 1.13 8 1.11 0

0.003 0.003

4 0.11 5 0.11 4

30 86398 9 12975 4

2 0.24 4 0.23 6

3 2.54 6 2.24 6

0.018 0.017

6 0.18 6 0.18 5

23 10798 05 16173 7

3 0.22 4 0.22 1

5 1.64 9 1.55 9

0.021 0.020

0.204 0.202

Count 43200 0

Avg. RT 23.2 57

Std. Dev. 12.5 95

Medi an 30.00 7

90th %ile 30.0 77

Count 10800 00

Avg. RT 16.1 37

Std. Dev. 15.9 44

Medi an 30.00 7

90th %ile 30.6 13

Count 17280 00

Avg. RT 15.9 92

Std. Dev. 22.0 90

Medi an 0.000

90th %ile 39.0 08

Count 21600 00

Avg . RT 7.74 7

Std. Dev. 16.0 26

Medi an 0.000

90th %ile 30.18 2

Count 28583 10808 34 28758 9 42994

Avg. RT 0.02 4 0.02 5 0.02 5 0.02 5

Std. Dev. 0.11 7 0.14 7 0.13 3 0.14 8

Medi an 0.025 0.025 0.025 0.025

90th %ile 0.03 4 0.03 4 0.03 4 0.03 4

Count 72026 27012 08 71903 3 10773 3

Avg. RT 0.12 6 0.12 4 0.12 4 0.12 8

Std. Dev. 0.78 2 0.77 8 0.79 0 0.86 0

Medi an 0.040 0.040 0.041 0.040

90th %ile 0.10 4 0.10 4 0.10 4 0.10 4

Count 11485 2 43200 62 11524 54 17263 2

Avg. RT 0.29 9 0.30 1 0.29 9 0.30 5

Std. Dev. 1.62 5 1.68 8 1.68 6 1.81 5

Medi an 0.054 0.054 0.054 0.054

90th %ile 0.16 9 0.16 8 0.16 8 0.16 8

Count 14414 6 53999 60 14402 70 21562 4

Avg . RT 0.29 6 0.29 7 0.29 7 0.29 6

Std. Dev. 1.61 6 1.58 9 1.60 0 1.50 6

Medi an 0.069 0.069 0.069 0.069

90th %ile 0.169 0.170 0.169 0.171

Conclusions
Response times generally increased under the successively higher loads. The TCP transactions are an exception, due the fact that more connections were dropped as the load increased. Response time variance increased as the load was increased. There was lower percentage of activation page response as the load was increased indicating that the system does not yield an inelastic response under different loads. We were not able to achieve the specified 4100 requests/second in the lab. At 100% load, it appears that something catastrophic occurred which caused a blank response for all HTTP Activation Server requests. Legacy Activation flows appeared to be unaffected by any additional load on the Walled Garden. CPU utilization and memory usage was insignificant and barely detectable on the Walled Garden servers.

You might also like