Incident #5638
closed
Task #5590: DataMiner as Generic Worker
Check data availability through URI Resolver
100%
Description
Sequential access to this file http://goo.gl/FcnUc0 (FishBase taxonomic file) fails. Concurrent access seems to systematically fail. This prevents using the file in experiments and could be affecting the system performance.
Updated by Gianpaolo Coro about 9 years ago
I have done some stress tests, since the link was working again this afternoon.
On the access-d.d4science.org machine, I run these benchmark processes few times on both the short Url and the long url:
ab -n 1000 -c 500 "http://data.d4science.org/smp?fileName=FISHBASE_taxa.taf.gz&contentType=application%2Fx-gzip&smp-uri=smp%3A%2F%2FShare%2F89971b8f-a993-4e7b-9a95-8d774cb68a99%2FWork+Packages%2FWP+6+-+Virtual+Research+Environments+Deployment+and+Operation%2FT6.2+Resources+and+Tools%2FCOMET-Species-Matching-Engine%2FYASMEEN%2F1.2.0%2FData%2FBiOnymTAF%2FFISHBASE_taxa.taf.gz%3F5ezvFfBOLqb3YESyI%2FkesN4T%2BZD0mtmc%2F4sZ0vGMrl0lgx7k85j8o2Q1vF0ezJi%2FxIGDhncO9jOkV1T8u6Db7GZ%2F4ePgMws8Jxu8ierJajHBd20bUotElN0kyA%2Bs3HQuMVYbva9MKgw1wahC7aUCyaItSZIQuKPu4pSjoDP8iox%2FXO2bqsokgB5v1H%2FQUQgN" ab -n 1000 -c 500 "http://goo.gl/FcnUc0"
and several wgets on the same files to check availability. After a few attempts (the benchmarks were always successful) the server began to be non-responding at all to the wgets. After a while, the file was available again (the server returned to be responding).
Updated by Roberto Cirillo about 9 years ago
- Status changed from New to Rejected
At this time we are not able to handle 1000 requests, with 500 requests running concurrently for the same file.
Is there a real case that need 500 concurrent access to the same file?
If the answer is yes (please specify the case), we need to convert our system in a sharding system with a horizontal scalability.
Updated by Gianpaolo Coro about 9 years ago
- Status changed from Rejected to In Progress
ab simulates high traffic but does not download the file. You can test also with less calls and concurrency. I wanted to demonstrate that under certain conditions the uri resolver does not respond.
The concrete case is that concurrent calls from the DataMiners often fail. We need to understand which is this maximum degree of concurrency we support.
If we do not support medium concurrency, then people are justified to use their own services to store and publish data.
Currently, a bug in the ic-client prevents to run tests with DataMiner, but I think the issue reported in this ticket is crucial and we cannot ignore it. We cannot use DataMiner to test the other services in the e-Infrastructure.
Updated by Gianpaolo Coro about 9 years ago
- Status changed from In Progress to Rejected
Since this issue requires further investigation and is possibly related to other issues than concurrency, I close it and a new one will be opened.
Updated by Pasquale Pagano about 9 years ago
I made same tests with three types of files.
The first file is 15 Mb and I made 1000 attempts with concurrency set to 50.
mb-pagano:~ pasqualepagano$ ab -n 1000 -c 50 http://data.d4science.org/dk9oekp4b1ZFajc5Z1ZXYXlUOUtMaUgzYUJOWXE5eDdHbWJQNStIS0N6Yz0
This is ApacheBench, Version 2.3 <$Revision: 1748469 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/
Benchmarking data.d4science.org (be patient)
Finished 1000 requests
Server Software: Apache-Coyote/1.1
Server Hostname: data.d4science.org
Server Port: 80
Document Path: /dk9oekp4b1ZFajc5Z1ZXYXlUOUtMaUgzYUJOWXE5eDdHbWJQNStIS0N6Yz0
Document Length: 14229041 bytes
Concurrency Level: 50
Time taken for tests: 226.883 seconds
Complete requests: 1000
Failed requests: 0
Total transferred: 14229343000 bytes
HTML transferred: 14229041000 bytes
Requests per second: 4.41 [#/sec] (mean)
Time per request: 11344.144 [ms] (mean)
Time per request: 226.883 [ms] (mean, across all concurrent requests)
Transfer rate: 61246.77 [Kbytes/sec] received
Connection Times (ms)
min mean[+/-sd] median max
Connect: 0 2 6.4 2 142
Processing: 4993 10770 2789.5 10396 27958
Waiting: 1354 3824 2176.4 3244 21097
Total: 4996 10772 2789.5 10397 27960
Percentage of the requests served within a certain time (ms)
50% 10397
66% 11657
75% 12407
80% 12814
90% 14275
95% 15271
98% 17536
99% 19947
100% 27960 (longest request)
Note that the amount of data transferred is 14229343000 bytes that means 13 GB; 100% success; 4.41 #/sec
The second test I did is with a smaller file, 500 kb.
I made 1000 attempts with concurrency set to 50.
mb-pagano:~ pasqualepagano$ ab -n 1000 -c 50 http://data.d4science.org/eWlXR1gvM05iZFRWUWhEWktTeVNVdWdTWGF0VTRIcVJHbWJQNStIS0N6Yz0
This is ApacheBench, Version 2.3 <$Revision: 1748469 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/
Benchmarking data.d4science.org (be patient)
Finished 1000 requests
Server Software: Apache-Coyote/1.1
Server Hostname: data.d4science.org
Server Port: 80
Document Path: /eWlXR1gvM05iZFRWUWhEWktTeVNVdWdTWGF0VTRIcVJHbWJQNStIS0N6Yz0
Document Length: 574815 bytes
Concurrency Level: 50
Time taken for tests: 119.583 seconds
Complete requests: 1000
Failed requests: 0
Total transferred: 575046000 bytes
HTML transferred: 574815000 bytes
Requests per second: 8.36 [#/sec] (mean)
Time per request: 5979.173 [ms] (mean)
Time per request: 119.583 [ms] (mean, across all concurrent requests)
Transfer rate: 4696.04 [Kbytes/sec] received
Connection Times (ms)
min mean[+/-sd] median max
Connect: 0 1 5.7 0 151
Processing: 2401 5863 1456.0 5708 10291
Waiting: 2382 5777 1437.1 5606 10279
Total: 2401 5865 1456.0 5709 10292
Percentage of the requests served within a certain time (ms)
50% 5709
66% 6329
75% 6783
80% 7062
90% 7937
95% 8677
98% 9192
99% 9405
100% 10292 (longest request)
Even in this case the percentage of success is 100%. The total transfer is 0.5 GB; the medium number of request per second are 8.36 [#/sec].
Finally, the third attempt was done with an even smaller file, 150 Kb, but with 100 concurrency accesses.
mb-pagano:~ pasqualepagano$ ab -n 1000 -c 100 http://data.d4science.org/LzhNd1h4c2VuVEo4YW5oVVRHbTBpcWhkeDhTUWRDeWxHbWJQNStIS0N6Yz0
This is ApacheBench, Version 2.3 <$Revision: 1748469 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/
Benchmarking data.d4science.org (be patient)
Finished 1000 requests
Server Software: Apache-Coyote/1.1
Server Hostname: data.d4science.org
Server Port: 80
Document Path: /LzhNd1h4c2VuVEo4YW5oVVRHbTBpcWhkeDhTUWRDeWxHbWJQNStIS0N6Yz0
Document Length: 147082 bytes
Concurrency Level: 100
Time taken for tests: 116.561 seconds
Complete requests: 1000
Failed requests: 0
Total transferred: 147360000 bytes
HTML transferred: 147082000 bytes
Requests per second: 8.58 [#/sec] (mean)
Time per request: 11656.074 [ms] (mean)
Time per request: 116.561 [ms] (mean, across all concurrent requests)
Transfer rate: 1234.60 [Kbytes/sec] received
Connection Times (ms)
min mean[+/-sd] median max
Connect: 0 2 9.1 0 147
Processing: 4516 11298 2868.0 11350 18630
Waiting: 4513 11168 2828.1 11204 18626
Total: 4517 11301 2867.6 11352 18633
Percentage of the requests served within a certain time (ms)
50% 11352
66% 12708
75% 13399
80% 13864
90% 14958
95% 16149
98% 16927
99% 17364
100% 18633 (longest request)
Again 100% success with 8.58 requests served per second.
It seems to me that overall the system works well. We need more details on the average volume to transfer and an average number of reasonable concurrent accesses to serve.
Updated by Roberto Cirillo about 9 years ago
I've created another ticket for the specific problem with the old smp uri reported by @gianpaolo.coro@isti.cnr.it : #5646
