Effects of remote machine entries in local host files on RPC calls

Effects of remote machine entries in local host files on RPC calls

Post by Dvai » Thu, 14 Nov 2002 01:42:43



Hi,

I am in the process of solving a lingering problem with telnet. It
takes for ever to establish a telnet connection from some machines to
other machines on my LAN (same network segment, no firewall ...),
however from other machines in the same network segment the telnet
session responds almost immediatly. From searching around I believe
that this can be solved by bypassing the DNS server and having entries
in the local host files of the different machines (UNIX and VMS). The
problem appears to be related to reverse lookups.

Arranging the use of local host files will take time on some of the
machines, but I beleive that the process can be sped up if I could
sell the change as a performance improvment. We only use the RPC
component of DCE, which maked similar calls to Microsofts COM/DCOM
technology. Our DCE servers are on the VMS server and we have a
central registry on a UNIX machine. The central registry is actually a
DCE server. The clients are running on NT servers. Calls are initiated
from the NT machine to the UNIX machine to find a server, once a
server is located, calls from the NT clients, go directly to the VMS
machines. Each time a call is made from the client the whole localion
process begins again. Although timings dont suggest that the RPC calls
take as long as the telnet connections, I am wondering if we can
reduce the call timings by adding entries to the local hosts files.

Any ideas would be appreciated...

 
 
 

Effects of remote machine entries in local host files on RPC calls

Post by Jakob Erbe » Thu, 14 Nov 2002 06:53:59



Quote:> Hi,

> I am in the process of solving a lingering problem with telnet. It
> takes for ever to establish a telnet connection from some machines to
> other machines on my LAN (same network segment, no firewall ...),
> however from other machines in the same network segment the telnet
> session responds almost immediatly. From searching around I believe
> that this can be solved by bypassing the DNS server and having entries
> in the local host files of the different machines (UNIX and VMS). The
> problem appears to be related to reverse lookups.

What has telnet got to do with DCE? Are you using telnet with DCE security?

Jakob

 
 
 

Effects of remote machine entries in local host files on RPC calls

Post by Ga M » Thu, 14 Nov 2002 16:33:49


Why bother maintaining all those host files.  Why don't you get the DNS
server(s) to the point where they are properly resolving forward and
reverse lookups?

Greg


> Hi,

> I am in the process of solving a lingering problem with telnet. It
> takes for ever to establish a telnet connection from some machines to
> other machines on my LAN (same network segment, no firewall ...),
> however from other machines in the same network segment the telnet
> session responds almost immediatly. From searching around I believe
> that this can be solved by bypassing the DNS server and having entries
> in the local host files of the different machines (UNIX and VMS). The
> problem appears to be related to reverse lookups.

> Arranging the use of local host files will take time on some of the
> machines, but I beleive that the process can be sped up if I could
> sell the change as a performance improvment. We only use the RPC
> component of DCE, which maked similar calls to Microsofts COM/DCOM
> technology. Our DCE servers are on the VMS server and we have a
> central registry on a UNIX machine. The central registry is actually a
> DCE server. The clients are running on NT servers. Calls are initiated
> from the NT machine to the UNIX machine to find a server, once a
> server is located, calls from the NT clients, go directly to the VMS
> machines. Each time a call is made from the client the whole localion
> process begins again. Although timings dont suggest that the RPC calls
> take as long as the telnet connections, I am wondering if we can
> reduce the call timings by adding entries to the local hosts files.

> Any ideas would be appreciated...

--
Pursuant to U.S. code,title 47, Chapter 5, Subchapter II, Section 227,
and consistent with Oregon State Law, any and all unsolicited commercial
E-mail sent to this address is subject to a consulting fee of $500 USD.
E-Mailing denotes acceptance of these terms.  Consult
http://www.law.cornell.edu/uscode/47/227.html for details.
 
 
 

Effects of remote machine entries in local host files on RPC calls

Post by Dvai » Fri, 15 Nov 2002 02:18:55



> Hi,

> I am in the process of solving a lingering problem with telnet. It
> takes for ever to establish a telnet connection from some machines to
> other machines on my LAN (same network segment, no firewall ...),
> however from other machines in the same network segment the telnet
> session responds almost immediatly. From searching around I believe
> that this can be solved by bypassing the DNS server and having entries
> in the local host files of the different machines (UNIX and VMS). The
> problem appears to be related to reverse lookups.

> Arranging the use of local host files will take time on some of the
> machines, but I beleive that the process can be sped up if I could
> sell the change as a performance improvment. We only use the RPC
> component of DCE, which maked similar calls to Microsofts COM/DCOM
> technology. Our DCE servers are on the VMS server and we have a
> central registry on a UNIX machine. The central registry is actually a
> DCE server. The clients are running on NT servers. Calls are initiated
> from the NT machine to the UNIX machine to find a server, once a
> server is located, calls from the NT clients, go directly to the VMS
> machines. Each time a call is made from the client the whole localion
> process begins again. Although timings dont suggest that the RPC calls
> take as long as the telnet connections, I am wondering if we can
> reduce the call timings by adding entries to the local hosts files.

> Any ideas would be appreciated...

I am just wondering if fixing my telnet problem will also improve the
performance of my RPC calls. I believe that adding host names to local
host files will reduce the time taken from when a telnet command is
issued to when the remote machine responds with a username prompt.

Regarding DCE, will the addition of names in local host files, speed
up the binding between the client and server? If a server recieves a
connection from a client, does the server have to do a reverse lookup
on the client's name/IP address? Surely if a lookup were performed on
client initialisation calls, response times would be quicker if a
local host file were used instead of using a DNS server?

Configuring our DNS server to handle reverse lookups, i am told, is
not practical in the short/medium term.

 
 
 

Effects of remote machine entries in local host files on RPC calls

Post by Ga M » Fri, 15 Nov 2002 02:48:05


    <<...snip...>>

Quote:> Configuring our DNS server to handle reverse lookups, i am told, is
> not practical in the short/medium term.

Having DNS working properly is not practical..?  Sounds like your DNS
admin is clueless and/or unorganized.  Good luck maintaining those hosts
files!

Greg

--
Pursuant to U.S. code,title 47, Chapter 5, Subchapter II, Section 227,
and consistent with Oregon State Law, any and all unsolicited commercial
E-mail sent to this address is subject to a consulting fee of $500 USD.
E-Mailing denotes acceptance of these terms.  Consult
http://www.law.cornell.edu/uscode/47/227.html for details.

 
 
 

Effects of remote machine entries in local host files on RPC calls

Post by Phil Hirsc » Fri, 15 Nov 2002 02:54:35


Quote:> Regarding DCE, will the addition of names in local host files, speed
> up the binding between the client and server? If a server recieves a
> connection from a client, does the server have to do a reverse lookup
> on the client's name/IP address?

No. DCE uses DNS only when locating a foreign cell to make cross-cell
RPCs, and sometimes when a new machine is being configured into a cell.
Even cross-cell RPCs don't require DNS after the foreign cell is found,
because the information about the foreign cell will be cached; but
"normal" intra-cell RPCs don't make any use of DNS at all.
 
 
 

Effects of remote machine entries in local host files on RPC calls

Post by Dvai » Fri, 15 Nov 2002 20:49:02



> > Regarding DCE, will the addition of names in local host files, speed
> > up the binding between the client and server? If a server recieves a
> > connection from a client, does the server have to do a reverse lookup
> > on the client's name/IP address?

> No. DCE uses DNS only when locating a foreign cell to make cross-cell
> RPCs, and sometimes when a new machine is being configured into a cell.
> Even cross-cell RPCs don't require DNS after the foreign cell is found,
> because the information about the foreign cell will be cached; but
> "normal" intra-cell RPCs don't make any use of DNS at all.

We only use the RPC components of DCE, some of the machines are
configured in their own DCE cell and others are not associated with
any cell. Would DNS be used if the server existed in a DCE cell and
the client was outside the cell on a machine that is not in any cell?
If we are making cross-cell calls would the use of local host files
negate the requirement to do a DNS lookup?

We have DCE clients/Servers running on OVMS, Digital Unix, Solaris and
NT4.0. When DCE was installed we were only interested in the run-time
components, consequently only some of the machines were added to their
own DCE cell. No DCE configuration work has ever been done on any of
the machines with DCE cells, as we are only interested in linking
against the DCE libraries. Should we consider moving all of our
machiens to 1 DCE cell? If we should what would be involved and what
should we watch out for? Just a cuple of brief points.

 
 
 

Effects of remote machine entries in local host files on RPC calls

Post by Phil Hirsc » Fri, 15 Nov 2002 22:28:35


Quote:> We only use the RPC components of DCE, some of the machines are
> configured in their own DCE cell and others are not associated with
> any cell. Would DNS be used if the server existed in a DCE cell and
> the client was outside the cell on a machine that is not in any cell?
> If we are making cross-cell calls would the use of local host files
> negate the requirement to do a DNS lookup?

If your clients are not members of a cell, then they must be using
string bindings to contact their servers; that would mean that DNS
is not involved at all, at any time in the RPC conversation (nor is
the hosts file).

I do wonder though about a comment you made previously, when you asked:

Quote:> will the addition of names in local host files, speed
> up the binding between the client and server

If your clients are indeed using string bindings, then the act of
binding should be almost instantaneous. Is that not the case?

Quote:> We have DCE clients/Servers running on OVMS, Digital Unix, Solaris and
> NT4.0. When DCE was installed we were only interested in the run-time
> components, consequently only some of the machines were added to their
> own DCE cell. No DCE configuration work has ever been done on any of
> the machines with DCE cells, as we are only interested in linking
> against the DCE libraries. Should we consider moving all of our
> machiens to 1 DCE cell? If we should what would be involved and what
> should we watch out for? Just a cuple of brief points.

If things are working OK as they are, then I can't see any reason to
change the configuration.
 
 
 

Effects of remote machine entries in local host files on RPC calls

Post by Phil Hirsc » Fri, 15 Nov 2002 23:54:48



> If your clients are not members of a cell, then they must be using
> string bindings to contact their servers; that would mean that DNS
> is not involved at all, at any time in the RPC conversation (nor is
> the hosts file).

After drinking my first cup of coffee, it occurred to me that you might
be using string bindings of the form "ncadg_ip_udp:machineName" rather
than bindings like "ncadg_ip_udp:10.11.12.13". If you are using string
bindings that contain machine names, then DCE does indeed have to use
DNS or the hosts file (or whatever you have configured for name
resolution) during the binding process; and if your name-resolution
setup is not working well, that could slow down the DCE binding
procedure
in this case.
 
 
 

Effects of remote machine entries in local host files on RPC calls

Post by Dvai » Mon, 18 Nov 2002 21:34:40




> > If your clients are not members of a cell, then they must be using
> > string bindings to contact their servers; that would mean that DNS
> > is not involved at all, at any time in the RPC conversation (nor is
> > the hosts file).

> After drinking my first cup of coffee, it occurred to me that you might
> be using string bindings of the form "ncadg_ip_udp:machineName" rather
> than bindings like "ncadg_ip_udp:10.11.12.13". If you are using string
> bindings that contain machine names, then DCE does indeed have to use
> DNS or the hosts file (or whatever you have configured for name
> resolution) during the binding process; and if your name-resolution
> setup is not working well, that could slow down the DCE binding
> procedure
> in this case.

Thanks Phil,

I checked the binding strings out in rpccp (show mapping). We are
using IP addresses. Any ideas as to why I get multiple listings for
each server? Surely I should have only 1 listing per server.

 
 
 

Effects of remote machine entries in local host files on RPC calls

Post by Phil Hirsc » Tue, 19 Nov 2002 23:25:03


Quote:> I checked the binding strings out in rpccp (show mapping). We are
> using IP addresses. Any ideas as to why I get multiple listings for
> each server? Surely I should have only 1 listing per server.

Two things:

1. You'll always see IP addresses in rpccp show mapping; but your
clients
   can still use machine names (rather than IP addresses) in their
string
   bindings. And if clients are indeed using names in string bindings,
   then DNS or the hosts file (or whatever) will be used to translate
the
   names to IP addresses.

2. There will be one entry in the map for each combination of protseq/
   interface/object. So if your server is accessible via both TCP and
UDP,
   and offers 3 different interfaces and 2 different objects, you'd see
   12 entries for the server in the map (2 * 3 * 2). If you look at your
   endpoint map, you should see that the multiple entries for the same
   server do all differ in some respect -- protseq or interface or
object,
   so that all 12 (or whatever) possible combinations are listed.