11 октября 2016 г. 21:58:48 CEST, Martin Waldenvik <walden...@gmx.com> пишет: >On 10/11/2016 09:14 PM, Dan McDonald wrote: >> Dumb question - is your FreeBSD frame size jumbo? >> >> Also, try this on OmniOS: >> >> ipadm set-prop -p send_buf=1048576 tcp >> ipadm set-prop -p recv_buf=1048576 tcp >> >> And see if that helps if the frame size isn't smaller on OmniOS. >> >> Dan >> >> Sent from my iPhone (typos, autocorrect, and all) >> >>> On Oct 11, 2016, at 1:53 PM, Martin Waldenvik <walden...@gmx.com> >wrote: >>> >>> Hi >>> >>> I'm wondering if anyone can shed some light on a problem I have. I'm >using lftp sftp a lot and it is very slow. I can work around that >sometimes by using another server. But I am also using zrep (using ssh) >to replicate my pool to a backup server. This is also slow. Transfer is >about 40 megabyte (omnios to omnios). When doing this on freebsd it is >about 100 megabyte and it saturates my gigabit link (freebsd to >freebsd). Same hardware. >>> >>> I'm using the latest Omios LTS release and openssh. Is there any way >to speed this up on Omnios? >>> >>> Regards >>> Martin >>> >>> >>> _______________________________________________ >>> OmniOS-discuss mailing list >>> OmniOS-discuss@lists.omniti.com >>> http://lists.omniti.com/mailman/listinfo/omnios-discuss > >Hi Dan > >Thanks for your input. I have had this problem for a long time, then i >thought is was the old SunSSH but the problem seems to persist with >OpenSSH. > >I have not touched the frame size, the only thing i have is LACP with >L3,L4. Intel NIC on a supermicro board. > >Tried to set send and recv buffer but there was no difference. It's >about the same speed with both sending and recieving ~40-45M/s. > >I have found that zrep is simple enough for me that manually replicates > >my system on an irregular basis. Maybe there is something that simple >around that works with netcat. > >Regards >Martin >_______________________________________________ >OmniOS-discuss mailing list >OmniOS-discuss@lists.omniti.com >http://lists.omniti.com/mailman/listinfo/omnios-discuss
Does this problem hit any ssh transfers, or in particular zfs send/recv? The latter is known to be bursty and benefits a lot from caching by mbuffer on at least one side of the link. Also for curiosity, did you try the hpc ssh patches (build your own), and/or reduced-security protocols (arcfour), and/or inline compression (ssh -z or explicit gzip/gzcat in the pipe)? Jim -- Typos courtesy of K-9 Mail on my Samsung Android _______________________________________________ OmniOS-discuss mailing list OmniOS-discuss@lists.omniti.com http://lists.omniti.com/mailman/listinfo/omnios-discuss