On 2016-01-26 22:39, Lars Nordin wrote:
> On 2016-01-26 21:40, Lars Nordin wrote:
>> On 2016-01-26 20:37, Daniel Stenberg wrote:
>>> On Tue, 26 Jan 2016, Lars Nordin wrote:
>>>
>>>> Have I missed any setting? I know local_echo is an option, but if
>>>> SecureCRT can, why not libssh2?
>>>
>>> Could it be as easy as setsockopt(... TCP_NODELAY ...) ?
>>> _______________________________________________
>>> libssh2-devel
>>> http://cool.haxx.se/cgi-bin/mailman/listinfo/libssh2-devel
>> I added the line, but no change...I'm on Windows and tried both the
>> ioctlsocket(... FIONBIO ... and now the setsockopt(... TCP_NODELAY
>> ...)
>> u_long flags = 1;
>>
>> //ioctlsocket(sock, FIONBIO, &flags); // tried both with and
>> without
>> setsockopt(sock, IPPROTO_TCP, TCP_NODELAY, (char *) &flags,
>> sizeof (u_long));
>>
>> Using my automatic character-sender I discovered a pattern, the ACK
>> is sent after 14 packetets, ie 7 characters sent and received. The
>> delay is 100ms (as programmed) between each sent characters (with a
>> reply in less then 10ms) until the ACK , the ACK is sent efter 200ms
>> delay and then the next character after 100ms more. Summary 200ms of
>> extra delay each 7 characters sent.
>>
>> /Lars
>> _______________________________________________
>> libssh2-devel http://cool.haxx.se/cgi-bin/mailman/listinfo/libssh2-devel
> I added code to print the clock() before and after the call to libssh2
> and discovered that it's the libssh2_write() function that sometime
> takes extre time to complete:
> write<: clock (ms) = 2042
> write>: clock (ms) = 2042
> read <: clock (ms) = 2042
> read >: clock (ms) = 2042
> write<: clock (ms) = 2143
> write>: clock (ms) = 2143
> read <: clock (ms) = 2143
> read >: clock (ms) = 2143
> write<: clock (ms) = 2243
> write>: clock (ms) = 2448
>
> Above explained, the 'write <' is before the call to write and 'write
> >' is after the call. I have a 100ms delay between each write() and
> the function is not using any time until the last one . Checking all
> of the output it's repeated each 7 write.
>
> I dig inte the libssh2_write() tomorrow!
>
> /Lars
>
I started timing functions inside libssh_channel_write and found the
cause of the delays, the RAND_bytes() function in the special
openssl-library I'm using creates the delay. I find a solution tomorrow
and maybe I propose a fix to the libssh2, I think the problem is that my
special openssl requires seed:ing the random generater and that is not
done in libssh2.
/Lars
_______________________________________________
libssh2-devel http://cool.haxx.se/cgi-bin/mailman/listinfo/libssh2-devel
Received on 2016-01-26