In my continued just-getting-started explorations, I tried firing up hello.fan web example. It comes up no problem and serves pages to the browser seemingly quickly. Then I thought I'd fire it up in apachebench (ab) just for fun to see how it performs.
ab -n 1000 127.0.0.1:8080/
results in a super sluggish wisp, that takes something like 10-11 seconds per request. Note that the above command does not issue any concurrent requests either. This is not a huge issue for me, as I was just playing with ab out of curiosity, but indicates that something is probably not happy in wisp.
brianSat 11 Jul 2009
Promoted to ticket #663 and assigned to brian
brianSat 11 Jul 2009
Ticket resolved in 1.0.45
There was a bug in my logic for checking for persistent connections where I was checking to ensure a 1.0 connection was not persistent. That fixes this problem:
C:\dev\fan\src>ab -n 1000 http://localhost:8080/
Server Software: Wisp/1.0.44
Server Hostname: localhost
Server Port: 8080
Document Path: /
Document Length: 619 bytes
Concurrency Level: 1
Time taken for tests: 5.252 seconds
Complete requests: 1000
Failed requests: 0
Write errors: 0
Total transferred: 781000 bytes
HTML transferred: 619000 bytes
Requests per second: 190.39 [#/sec] (mean)
Time per request: 5.252 [ms] (mean)
Time per request: 5.252 [ms] (mean, across all concurrent requests)
Transfer rate: 145.21 [Kbytes/sec] received
Connection Times (ms)
min mean[+/-sd] median max
Connect: 0 2 6.4 0 140
Processing: 0 4 6.5 0 16
Waiting: 0 2 4.6 0 16
Total: 0 5 8.4 0 140
Then I tried the more interesting tests with concurrency and keep-alives. The way apache bench works seems fishy to me - it uses HTTP/1.0 but with a Connection header which doesn't seem correct to me, but I tweaked wisp to make it work. Plus it doesn't appear to support chunked transfer encoding, so your test has to ensure that Content-Length is used on the response. That test on my machine was:
C:\dev\fan\src>ab -n 50000 -c 50 -k http://localhost:8080/
Server Software: Wisp/1.0.44
Server Hostname: localhost
Server Port: 8080
Document Path: /
Document Length: 14 bytes
Concurrency Level: 50
Time taken for tests: 70.745 seconds
Complete requests: 50000
Failed requests: 0
Write errors: 0
Keep-Alive requests: 50000
Total transferred: 8801584 bytes
HTML transferred: 700126 bytes
Requests per second: 706.77 [#/sec] (mean)
Time per request: 70.745 [ms] (mean)
Time per request: 1.415 [ms] (mean, across all concurrent requests)
Transfer rate: 121.50 [Kbytes/sec] received
Connection Times (ms)
min mean[+/-sd] median max
Connect: 0 0 0.5 0 109
Processing: 16 71 14.0 62 218
Waiting: 0 69 13.7 62 218
Total: 16 71 14.1 62 249
Don't know how that stacks up to other web servers, but seems pretty decent to me (wisp runs this site).
liamstaskSat 11 Jul 2009
I'd call this fixed for sure. Updated to latest and greatest and ran the hello.fan example again. Results after a warm up run on my MacBook Pro:
Server Software: Wisp/1.0.44
Server Hostname: localhost
Server Port: 8080
Document Path: /
Document Length: 14 bytes
Concurrency Level: 50
Time taken for tests: 3.555 seconds
Complete requests: 50000
Failed requests: 0
Write errors: 0
Keep-Alive requests: 50000
Total transferred: 8800352 bytes
HTML transferred: 700028 bytes
Requests per second: 14062.75 [#/sec] (mean)
Time per request: 3.555 [ms] (mean)
Time per request: 0.071 [ms] (mean, across all concurrent requests)
Transfer rate: 2417.13 [Kbytes/sec] received
Connection Times (ms)
min mean[+/-sd] median max
Connect: 0 0 0.1 0 3
Processing: 0 4 3.4 3 143
Waiting: 0 3 3.4 3 143
Total: 0 4 3.5 3 145
Percentage of the requests served within a certain time (ms)
50% 3
66% 4
75% 4
80% 4
90% 5
95% 6
98% 9
99% 11
100% 145 (longest request)
It's a simple page, of course, but I think those are pretty damn good numbers. Thanks for the fix!
liamstask Fri 10 Jul 2009
In my continued just-getting-started explorations, I tried firing up hello.fan web example. It comes up no problem and serves pages to the browser seemingly quickly. Then I thought I'd fire it up in apachebench (ab) just for fun to see how it performs.
results in a super sluggish wisp, that takes something like 10-11 seconds per request. Note that the above command does not issue any concurrent requests either. This is not a huge issue for me, as I was just playing with ab out of curiosity, but indicates that something is probably not happy in wisp.
brian Sat 11 Jul 2009
Promoted to ticket #663 and assigned to brian
brian Sat 11 Jul 2009
Ticket resolved in 1.0.45
There was a bug in my logic for checking for persistent connections where I was checking to ensure a 1.0 connection was not persistent. That fixes this problem:
C:\dev\fan\src>ab -n 1000 http://localhost:8080/ Server Software: Wisp/1.0.44 Server Hostname: localhost Server Port: 8080 Document Path: / Document Length: 619 bytes Concurrency Level: 1 Time taken for tests: 5.252 seconds Complete requests: 1000 Failed requests: 0 Write errors: 0 Total transferred: 781000 bytes HTML transferred: 619000 bytes Requests per second: 190.39 [#/sec] (mean) Time per request: 5.252 [ms] (mean) Time per request: 5.252 [ms] (mean, across all concurrent requests) Transfer rate: 145.21 [Kbytes/sec] received Connection Times (ms) min mean[+/-sd] median max Connect: 0 2 6.4 0 140 Processing: 0 4 6.5 0 16 Waiting: 0 2 4.6 0 16 Total: 0 5 8.4 0 140Then I tried the more interesting tests with concurrency and keep-alives. The way apache bench works seems fishy to me - it uses HTTP/1.0 but with a Connection header which doesn't seem correct to me, but I tweaked wisp to make it work. Plus it doesn't appear to support chunked transfer encoding, so your test has to ensure that Content-Length is used on the response. That test on my machine was:
C:\dev\fan\src>ab -n 50000 -c 50 -k http://localhost:8080/ Server Software: Wisp/1.0.44 Server Hostname: localhost Server Port: 8080 Document Path: / Document Length: 14 bytes Concurrency Level: 50 Time taken for tests: 70.745 seconds Complete requests: 50000 Failed requests: 0 Write errors: 0 Keep-Alive requests: 50000 Total transferred: 8801584 bytes HTML transferred: 700126 bytes Requests per second: 706.77 [#/sec] (mean) Time per request: 70.745 [ms] (mean) Time per request: 1.415 [ms] (mean, across all concurrent requests) Transfer rate: 121.50 [Kbytes/sec] received Connection Times (ms) min mean[+/-sd] median max Connect: 0 0 0.5 0 109 Processing: 16 71 14.0 62 218 Waiting: 0 69 13.7 62 218 Total: 16 71 14.1 62 249Don't know how that stacks up to other web servers, but seems pretty decent to me (wisp runs this site).
liamstask Sat 11 Jul 2009
I'd call this fixed for sure. Updated to latest and greatest and ran the hello.fan example again. Results after a warm up run on my MacBook Pro:
Server Software: Wisp/1.0.44 Server Hostname: localhost Server Port: 8080 Document Path: / Document Length: 14 bytes Concurrency Level: 50 Time taken for tests: 3.555 seconds Complete requests: 50000 Failed requests: 0 Write errors: 0 Keep-Alive requests: 50000 Total transferred: 8800352 bytes HTML transferred: 700028 bytes Requests per second: 14062.75 [#/sec] (mean) Time per request: 3.555 [ms] (mean) Time per request: 0.071 [ms] (mean, across all concurrent requests) Transfer rate: 2417.13 [Kbytes/sec] received Connection Times (ms) min mean[+/-sd] median max Connect: 0 0 0.1 0 3 Processing: 0 4 3.4 3 143 Waiting: 0 3 3.4 3 143 Total: 0 4 3.5 3 145 Percentage of the requests served within a certain time (ms) 50% 3 66% 4 75% 4 80% 4 90% 5 95% 6 98% 9 99% 11 100% 145 (longest request)It's a simple page, of course, but I think those are pretty damn good numbers. Thanks for the fix!