Closed ankitsingla closed 9 years ago
The error may been caused by sorting issue we have. I think it has been fixed
We need to be sure about this one. Can't you do a "replay" on the data, using your compute code? First remove all the tests from Chandigarh. Then add those data items one by one, running the computations each time and seeing the output for Chandigarh.
I don't see how the sorting would affect things like "How's the median in Chandigarh 128.2s? One test had 112.9s, the other had 192.9s (as is clear from the earlier screenshot in the other issue I've filed)."
Ankit
Ankit Singla PhD Candidate, Computer Science University of Illinois at Urbana-Champaign (UIUC) http://web.engr.illinois.edu/~singla2/
On 28 November 2014 at 08:10, Xuefeng Zhu notifications@github.com wrote:
The error may been caused by sorting issue we have. I think it has been fixed
— Reply to this email directly or view it on GitHub https://github.com/Xuefeng-Zhu/cSpeed/issues/12#issuecomment-64846953.
Let me inspect the data.
'Airtel Broadband': { median: [ 13932, 16513, 17212, 17343, 18269, 18793, 18860, 19907, 19951, 21740, 22903, 23864, 26408 ], count: 13 } }
`
This is the data in database for Chandigarh. It seems that there is even data reaching 100s..
Then there are even more problems than before ... because the screenshots I posted clearly say Chandigarh, and clearly say 100+ seconds more than once.
Ankit
Ankit Singla PhD Candidate, Computer Science University of Illinois at Urbana-Champaign (UIUC) http://web.engr.illinois.edu/~singla2/
On 28 November 2014 at 08:22, Xuefeng Zhu notifications@github.com wrote:
'Airtel Broadband': { median: [ 13932, 16513, 17212, 17343, 18269, 18793, 18860, 19907, 19951, 21740, 22903, 23864, 26408 ], count: 13 } }
This is the data in database for Chandigarh. It seems that there is even data reaching 100s..
— Reply to this email directly or view it on GitHub https://github.com/Xuefeng-Zhu/cSpeed/issues/12#issuecomment-64847666.
Is it based on test database or the production one? The data above is from test database
I'm talking about production results only. The test database will have small numbers because it's using fewer Websites.
In my screenshots, you can see at least three results for Chandigarh:
72.2s ( https://cloud.githubusercontent.com/assets/6640600/5047178/9f56ab42-6c38-11e4-9cba-9664377507e4.png )
192.9s ( https://cloud.githubusercontent.com/assets/6640600/5046643/1b5912c6-6c33-11e4-9530-f490912da875.png )
112.9s (same screenshot as above, see the "1 test" in the median results for Chandigarh)
Ankit
Ankit
'Airtel Broadband': { median: [ 78184, 112903, 115370, 128175 ], count: 4 } }
This is the data I got.
I feel it is also caused by the sorting median issue.
Hmmm ... do see a problem here? There are inconsistencies between the data and the screenshots. So 112.9 is there, but there's not 192.9 or 72.2
Ankit Singla PhD Candidate, Computer Science University of Illinois at Urbana-Champaign (UIUC) http://web.engr.illinois.edu/~singla2/
On 28 November 2014 at 08:48, Xuefeng Zhu notifications@github.com wrote:
'Airtel Broadband': { median: [ 78184, 112903, 115370, 128175 ], count: 4 } } This is the data I got.
— Reply to this email directly or view it on GitHub https://github.com/Xuefeng-Zhu/cSpeed/issues/12#issuecomment-64849249.
Is there timeout test for the 192.9. For the database in the database, it may treat timeout result differently from client side.
So at client side the timeout will be treat by
uTotal += $scope.total[value.name] * timeoutRatio;
At server side, I use $scope.total[value.name] directly.
You should use the same value ($scope.total[value.name] * timeoutRatio;) on the server.
Remember though, exclude any results with more than 50% timeouts. Don't show a total number even on the client.
Ankit
Ankit Singla PhD Candidate, Computer Science University of Illinois at Urbana-Champaign (UIUC) http://web.engr.illinois.edu/~singla2/
On 28 November 2014 at 09:19, Xuefeng Zhu notifications@github.com wrote:
So at client side the timeout will be treat by uTotal += $scope.total[value.name] * timeoutRatio; At server side, I use $scope.total[value.name] directly.
— Reply to this email directly or view it on GitHub https://github.com/Xuefeng-Zhu/cSpeed/issues/12#issuecomment-64851190.
I have made the change
The first test I ran from Chandigarh (my hometown in India), I got the result (without charts, because this is still the Chrome store version) saying 312% slower than the typical user, etc. Now, with the second test, the results are in the below screenshot:
Now, the median here is 50.3 seconds. The last test's result was 112.9s per this screenshot (because that's the only other test, as "1 test" confirms). So how could that have been 312% slower? If the median was still roughly 50s, 312% slower means 207s. The median certainly shouldn't have been 27s (which would make 112.9s slower by 312%). There's something fishy here.