[
https://issues.apache.org/jira/browse/SSHD-1169?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17459657#comment-17459657
]
zhaizeyu commented on SSHD-1169:
--------------------------------
The data in the table are the experimental data recorded. The comparison
between No.1 and No.3 shows that in the case of 100 MB files and 10 cycles,
whether to restart the system has little impact on the throughput. And also in
the case of No.9 and No.10, even when the number of cycles is increased. We
need to further analyze the other sets of experimental data to find out the key
influencing factors. Next, we're going to try to do 2000 cycles. The test
project address will be added to github after I optimize it.
|o.|threads|cycles|file size（M)|Number of files|Total Execution Time|Execution
duration when no file is transferred|Actual SFTP
Duration|Throughput（M/s）|Throughput/threads|Remarks|
|1|20|10|109|200|121.1|23.2|97.9|222.68|11.1338| |
|3|20|10|109|200|117.6|21.5|96.1|226.85|11.3424|After the SFTP server is
restarted|
|9|20|100|109|2000|1229.4|242.7|986.7|220.94|11.0469| |
|10|20|100|109|2000|1230.2|241.3|988.9|220.45|11.0223|After the SFTP server is
restarted|
> Sftp Server throughput is lagging for concurrent threads
> --------------------------------------------------------
>
> Key: SSHD-1169
> URL: https://issues.apache.org/jira/browse/SSHD-1169
> Project: MINA SSHD
> Issue Type: Improvement
> Affects Versions: 2.5.1
> Reporter: Susmit Sarkar
> Priority: Blocker
> Attachments: image-2021-05-19-10-27-20-261.png
>
>
> Hello Team,
> We are seeing a considerably low throughput for large-sized file (100Mb sized
> file during concurrent execution)
> We did some load testing and the throughputs sometimes increase (when we do a
> VM reboot for 1st run) and sometimes its pretty low compare to commercial
> servers
>
> *2000 cycle runs | 20 concurrent threads | 100mb files == Throughput is 0.94
> documents per seconds*
> For a commercial server, it's around 1.17 documents per second.
> Now if we do a server reboot and re-execute the test then the throughput
> increases by leaps and bounds
> *2000 cycle runs | 20 concurrent threads | 100 MB files == Throughput is 1.90
> documents per seconds (More than 100 percent)*
> *cycle means 100 MB file upload times, 2000 means 2000times*100 MB which got
> distributed by 20 concurrent sftp sessions*
> I am using IO Factory as NIO2, and below are the combinations for different
> NIO worker threads
> !image-2021-05-19-10-27-20-261.png!
> We are pretty confused and need help. For small files, Apache performed very
> well the throughput was 30 percent better than commercial server, but for
> large files, we are seeing these issues sometimes (mostly with reboot it
> performs good, but after that run its starts lagging)
> nio-workers ===> *varies from the chart above*
> max-auth-requests ===> 3
> Compression ===> NONE
> max-concurrent-sessions ===> 2147483647
> packet-size ===> 65536
> idle-timeout ===> 300000
> server-identification ===> SFTP Server
> window-size ===> 4194304
> we are setting currently all this properties to *SshServer*, is there any
> performance tuning parameter can be used?
>
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
