determining the cause of data transfer weirdness

hi all,

I have 2 servers that talk over the internet. server1 and server2.

Server1 has a file with a size of 5MB that needs to be transferred
to server2.

Im using sftp to transfer data.

Weird problem:

* if i initiate sftp from server1 to server2 and send the file its
very slow
* if i initiate sftp from server2 to server1 and get the file its
very fast

I was also wondering if routes are symmetric?


jm,

On Mar 6, 2:58*pm, Maxwell Lol <nos...@com.invalid> wrote:
> jm <jerom...@gmail.com> writes:
> > hi all,
>
> > * I have 2 servers that talk over the internet. *server1 and server2.
>
> > * Server1 has a file with a size of 5MB that needs to be transferred
> > to server2.
>
> > * Im using sftp to transfer data.
>
> > * Weird problem:
>
> > * ** if i initiate sftp from server1 to server2 and send the file its
> > very slow
> > * ** if i initiate sftp from server2 to server1 and get the file its
> > very fast
>
> > * I was also wondering if routes are symmetric?
>
> Routes and bandwidth are not always symmetric.
>
> There is also variations in applications, and OS parameters (default
> TCP window size)
>
> Try ping, and also try using different packet sizes.
>
> Traceroute will also show you the route.

hmm.. application is not going to an issue since im using the same app
to transfer data on both sides.
OS parameters are the same on both machines..

the only difference is the route is take..

from server1 it takes path A, B, C to get to server2

from server2 it takes path C, D, E, F, G, B, A to get to server1.

im assuming that when server1 connects to server2 the return packet
will use the same path, (path used to initiate the connection)...

by hunch.. is that there is a QoS running but i don't have access on
the router on the other end since its being managed by the ISP...

On Mar 9, 12:47*pm, jm <jerom...@gmail.com> wrote:

> im assuming that when server1 connects to server2 the return packet
> will use the same path, (path used to initiate the connection)...

No. That's not only impossible but would cause network providers to
violate their contracts.

DS

David Schwartz <(E-Mail Removed)> writes:

>On Mar 9, 12:47=A0pm, jm <jerom...@gmail.com> wrote:

>> im assuming that when server1 connects to server2 the return packet
>> will use the same path, (path used to initiate the connection)...

>No. That's not only impossible but would cause network providers to
>violate their contracts.

??? No idea what this means. But, there is no reason the packet should take
the same route. The packet has no idea of what the route was on the
outgoing leg, so has not idea what the "same" route is. All it has is an
IP address, and it tries to get the packet to that address. The path it
takes will depend entirely on the routing tables along the way.



>DS

On Mar 9, 4:01*pm, Unruh <unruh-s...@physics.ubc.ca> wrote:

> David Schwartz <dav...@webmaster.com> writes:

> >No. That's not only impossible but would cause network providers to
> >violate their contracts.

> ??? No idea what this means.

It means that a network provider would have to violate their contract
to allow a packet to take the same route back that it took out.

For example, suppose you are on the West Coast and use ProviderX. I am
on the East Coast and use ProviderY. The contract between ProviderX
and ProviderY specifies nearest exit.

If I send a packet to your server, my packet will cross between
ProviderX and ProviderY on the East Coast, as their contract requires.
If you send a reply along the same path, you would be causing your
provider to violate its contract with my provider by dumping the
traffic on the wrong coast.

Or suppose you are in Japan and I am in Singapore, and we have
different providers. Typically, one provider will be responsible for
the long haul in one direction and one would be responsible for the
long haul in the other direction. If reply packets took the same path,
that would be abusing whichever provider's long haul circuits the
packet happened to take by making them do the long haul in both
directions.

Same path in both directions is simply not equitable on the Internet.

DS