0

Inter continent data transfer's speed is maximum 2MB/s.

I checked and the SSH server of my server doesn't even use window scaling, and the window itself is very small, around 22KB...

Flags [S], seq 1433200120, win 29200, options [mss 1420,sackOK,TS val 1451891061 ecr 0,nop,wscale 7], length 0 Flags [S.], seq 3549718494, ack 1433200121, win 65535, options [mss 1460,sackOK,TS val 4214039974 ecr 1451891061,nop,wscale 9], length 0 

Server is 65535*9=590 KB

Window scaling is enabled.

$ cat /proc/sys/net/ipv4/tcp_window_scaling 1 

And I've already increase all the parameters to 25MB and 16MB minimum an default in /etc/sysctl.conf:

net.core.wmem_max=25165824 net.core.rmem_max=25165824 net.ipv4.tcp_rmem = 16777216 16777216 25165824 net.ipv4.tcp_wmem = 16777216 16777216 25165824 

I'm using Fedora 39 and RHEL 8, what can I do to force the SSH server to handle much more inflight data?

4
  • 2
    scaling 9 means 2^9 = 512
    – A.B
    CommentedJan 26, 2024 at 11:10
  • This does not look like an issue with the window scaling at your end. If your packet traces suggest this is the issue, then you should check the other end (e.g. AIX still defaults to disabling this the last time I checked)
    – symcbean
    CommentedJan 26, 2024 at 15:42
  • How come? I set the min window at 16MB, I get 0.5MB... I really want to force a much bigger window to improve the bandwidth.
    – None
    CommentedJan 27, 2024 at 0:23
  • scp a file (10s of MB will do) and capture the whole session to a pcap. open that in Wireshark and you can Statistic -> TCP Stream Graphs -> Window Scaling. Would be interesting to see how (if) the window changes over time. If you get a sawtooth pattern, something is causing your client to back off.CommentedOct 4, 2024 at 23:07

1 Answer 1

0

So the point is the latency between the end-points which hits TCP quite bad for bandwidth consumption, and therefore you try to go ahead with the TCP window scaling.

About 10 years back i also had issues transferring TBs of data across the glob and if you can not head for block based replication and therefore stuck with filesystem structure scans and copy, you may checkout bbcp.

http://pcbunn.cithep.caltech.edu/bbcp/using_bbcp.htm (disclaimer: i am not taking responsibility for the content of the link)

Those days it was quite helpful as it also can utilize with the parallel streams the full WAN capacity and works similar as scp.

In the end this was also the reason why i stopped it, as on leased lines one is not alone, you have to anyway align with the network operations center, either on bandwidth limits or on quality of service definitions of your traffic, in the end - technical feasible - but not sustainable.

1
  • Please provide a relevant quote from the link, in case it goes offline or is otherwise unavailable.
    – Hack Saw
    CommentedOct 6, 2024 at 21:24

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.