NetBSD-Users archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: rsync over ssh/nfs (was: server crashing)



On Aug 31, 2011, at 1:01 PM, David Brownlee wrote:
>> Are you using SSH+HPN patches?  SSH by default doesn't buffer enough data to 
>> saturate a high bandwidth * delay connection; also, of course, the crypto 
>> overhead can cause you to be CPU bound.
>> 
>>  http://www.psc.edu/networking/projects/hpn-ssh/
> 
> This was between two boxes, each with 1.3Ghz Athlon II Neo N36L Dual-Core 
> CPUs, so not particularly powerful.
> 
> They were both into the same cheap gigabit switch. Now they are connected 
> across ADSL so transfer rate (across ssh) averages 90KB/s, so I wouldn't have 
> a chance to retry with the HPN patches. At first glance they would not be 
> relevant for a LAN?

Agreed-- the receive window changes probably won't help much for a LAN, where 
there is probably/hopefully very little delay.  But, if your CPUs are the 
bottleneck, then switching to a less-expensive crypto algorithm, or even HPN's 
"none" might make a more significant difference.

[The "none" crypto algorithm from HPN uses SSH normal crypto to handle the 
login, but then doesn't encrypt the data stream once the session is 
established.  It's definitely not suitable for all purposes, but if you're 
willing to send the data over NFS in plaintext, then using this might be 
reasonable... ]

Regards,
-- 
-Chuck



Home | Main Index | Thread Index | Old Index