Hi,

Am I missing something obvious? Why is it that ssh2 on my openbsd 3.1 sparc machine is so incredibly
slow? I've tried using both RSA and DSA. When i say slow I mean it's slow to get a prompt to enter
my username and password, but after I connect it's fine. Someone suggested that it's because these
sparcs don't have hardware divide... is this the case?

uname says:
OpenBSD moron 3.1 GENERIC#27 sparc

ssh -V says:
OpenSSH_3.2, SSH protocols 1.5/2.0, OpenSSL 0x0090602f

Thanks,
Andrew

Andrew Eberbach wrote:

> Hi,
>
> Am I missing something obvious? Why is it that ssh2 on my openbsd 3.1
> sparc machine is so incredibly slow? I've tried using both RSA and
> DSA. When i say slow I mean it's slow to get a prompt to enter my
> username and password, but after I connect it's fine. Someone
> suggested that it's because these sparcs don't have hardware divide...
> is this the case?
>
> uname says:
> OpenBSD moron 3.1 GENERIC#27 sparc
>
> ssh -V says:
> OpenSSH_3.2, SSH protocols 1.5/2.0, OpenSSL 0x0090602f
>

Recompile for your specific machine. It will speed up with approx. a
factor of 4.

EJ
--
Remove the obvious part (including the dot) for my email address.
http://www.vanwesten.net for examples of ipf and pf.

Hi,

Should I be passing some special flags to the compiler? I just recompiled and it's still slow (no
noticeable difference).

Andrew

erik wrote:
> Andrew Eberbach wrote:
>
>
>>Hi,
>>
>>Am I missing something obvious? Why is it that ssh2 on my openbsd 3.1
>>sparc machine is so incredibly slow? I've tried using both RSA and
>>DSA. When i say slow I mean it's slow to get a prompt to enter my
>>username and password, but after I connect it's fine. Someone
>>suggested that it's because these sparcs don't have hardware divide...
>>is this the case?
>>
>>uname says:
>>OpenBSD moron 3.1 GENERIC#27 sparc
>>
>>ssh -V says:
>>OpenSSH_3.2, SSH protocols 1.5/2.0, OpenSSL 0x0090602f
>>
>
>
> Recompile for your specific machine. It will speed up with approx. a
> factor of 4.
>
> EJ

Andrew Eberbach wrote:

> Hi,
>
> Should I be passing some special flags to the compiler? I just
> recompiled and it's still slow (no noticeable difference).
>

If memory serves me well you should also recompile ssl. Again, use the
flags specific for your type of machine. Not all suns are created
equally.

EJ
--
Remove the obvious part (including the dot) for my email address.
http://www.vanwesten.net for examples of ipf and pf.

On Fri, 12 Dec 2003, Andrew Eberbach wrote:

> Am I missing something obvious? Why is it that ssh2 on my openbsd 3.1 sparc machine is so incredibly
> slow? I've tried using both RSA and DSA. When i say slow I mean it's slow to get a prompt to enter
> my username and password, but after I connect it's fine. Someone suggested that it's because these
> sparcs don't have hardware divide... is this the case?

1. ssh2 really is a lot slower. a lot lot slower.
2. the first sparcs didn't have hardware multiply/divide. so the
compiler can't use those instructions. compiling with
COPTS=-march=supersparc in /etc/mk.conf works on some machines. i don't
know about what cpu is in ss1, but i think it's one without the
instructions, so this won't work.


--
"People have criticized me because my security detail is larger
than the president's. But you must ask yourself: are there more
people who want to kill me than who want to kill the president?
I can assure you there are."
- M. Barry, Mayor of Washington, DC

On Sat, 13 Dec 2003 17:01:39 -0800, Ted Unangst <tedu@stanford.edu> wrote:

>On Fri, 12 Dec 2003, Andrew Eberbach wrote:
>
>> Am I missing something obvious? Why is it that ssh2 on my openbsd 3.1 sparc machine is so incredibly
>> slow? I've tried using both RSA and DSA. When i say slow I mean it's slow to get a prompt to enter
>> my username and password, but after I connect it's fine. Someone suggested that it's because these
>> sparcs don't have hardware divide... is this the case?
>
>1. ssh2 really is a lot slower. a lot lot slower.
>2. the first sparcs didn't have hardware multiply/divide. so the
>compiler can't use those instructions. compiling with
>COPTS=-march=supersparc in /etc/mk.conf works on some machines. i don't
>know about what cpu is in ss1, but i think it's one without the
>instructions, so this won't work.

The SS1 has a SparcV7 CPU, which doesn't have multiply/divide. These
instructions are emulated, which is slow. -mcpu=cypress should be better.

Mit freundlichen Grüßen

Dipl.-Ing. Frank-Christian Krügel

Hi,

I set CFLAGS=" -mcpu=cypress"

and recompiled it. No gains. Still takes a little over three minutes to get a prompt.

DEBUG3 reports:
Dec 14 10:26:40 moron sshd[31908]: debug3: mm_request_receive entering
Dec 14 10:29:20 moron sshd[31908]: debug3: monitor_read: checking request 4
Dec 14 10:29:20 moron sshd[31908]: debug3: mm_answer_sign

Cheers,
Andrew

Frank-Christian Kruegel wrote:

> On Sat, 13 Dec 2003 17:01:39 -0800, Ted Unangst <tedu@stanford.edu> wrote:
>
>
>>On Fri, 12 Dec 2003, Andrew Eberbach wrote:
>>
>>
>>>Am I missing something obvious? Why is it that ssh2 on my openbsd 3.1 sparc machine is so incredibly
>>>slow? I've tried using both RSA and DSA. When i say slow I mean it's slow to get a prompt to enter
>>>my username and password, but after I connect it's fine. Someone suggested that it's because these
>>>sparcs don't have hardware divide... is this the case?
>>
>>1. ssh2 really is a lot slower. a lot lot slower.
>>2. the first sparcs didn't have hardware multiply/divide. so the
>>compiler can't use those instructions. compiling with
>>COPTS=-march=supersparc in /etc/mk.conf works on some machines. i don't
>>know about what cpu is in ss1, but i think it's one without the
>>instructions, so this won't work.
>
>
> The SS1 has a SparcV7 CPU, which doesn't have multiply/divide. These
> instructions are emulated, which is slow. -mcpu=cypress should be better.
>
> Mit freundlichen Grüßen
>
> Dipl.-Ing. Frank-Christian Krügel