tech-net archive

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

Re: abysmally low scp performance (Edgar =?iso-8859-1?B?RnXf?=) writes:

>> A common good value is 920-940Mbit/s for TCP alone.
>That's approximately what I'd expect.

>> ssh then depends on CPU performance.
>But as long as all CPUs are significantly idle, all should be OK, no?

Unfortunately not, the encryption can cause signficant delays so that
TCP will throttle.

>wm0 at pci2 dev 0 function 0: Intel i82574L (rev. 0x00)
>wm0 at pci0 dev 4 function 0: Intel i82541GI 1000BASE-T Ethernet (rev. 0x05)

Here it is I211 vs. I217.

>Anything suspicious about that?

Not to me.

In the iperf test you will see a 'Retr' column on the sender. That's
TCP retries, usually caused by dropped packets. Any number > 0 is bad,
unfortunately it doesn't tell why or where the packets were dropped.

>> That HPN patch is old, the current (patch, not NetBSD) one is better.
>> But for most people, disabling is much easier.
>So I guess it should be disabled by default?

It was recently disabled (in /etc/ssh/sshd_config) for -current.

Home | Main Index | Thread Index | Old Index