T.R | Title | User | Personal Name | Date | Lines |
---|
4169.1 | | VMSNET::P_NUNEZ | | Mon Feb 24 1997 12:56 | 18 |
|
You may be better off using UCX's TCPIPTRACE facility rather than
server debug mode to trace this. Putting the server in debug mode is
just going to slow things down more.
Do $ HELP TCPIPTRACE for usage info.
Might also ask in UPSAR::FDDI regarding FDDI tuning/parameters.
LASSIE::UCX is UCX notes conference and URL for UCX tuning is
http://www-mcs.iso.dec.com/ucx_tuning.html.
Do you have any non-NT clients you could test the transfer with (to
eliminate the NT IP stack as the source of problems)?
I assume you used the same disk device on the PATHWORKS server for all
file copy/ftp tests?
Paul
|
4169.2 | Try this. | CPEEDY::FLEURY | | Mon Feb 24 1997 13:07 | 10 |
| RE: .1
This may be a case where the internal workings of UCX may be playing a
part. Can you have the customer try the following:
$ UCX SET PROTOCOL TCP /NODELACK
$ UCX SET CONFIG PROTOCOL TCP /NODELACK
Dan
|
4169.3 | .2 fixed half of the problem :-) | BULAN::peoto.soo.dec.com::ptorlind | | Tue Feb 25 1997 07:02 | 19 |
| Hi !!
.2 Fixed half of the problem
Now it takes 2-3 seconds to copy the same 5MB file as mentioned in
.0 from NT server to VMS/PW server. Great. Compare that with
previous 15-20 seconds. Customer impressed
But, it still takes 35 seconds from VMS/PW to NT server.
I checked the manuals for NT and kan see that Delayed Acks is
default in NT and there is according to the manuals no configuration
parameter to disable Acks as we did in the VMS/PW server.
Anyone know if there is any undocumented parameter or any other
to do the trick of getting this to work ???
Per Torlind / PATHWORKS support, Sweden
|
4169.4 | Problem fixed | BULAN::peoto.soo.dec.com::ptorlind | | Wed Feb 26 1997 07:30 | 13 |
| Hi !!
Update to .3
Setting TcpWindowSize in Registryt on the NT server fixed
it.
Thanks
Per Torlind / PATHWORKS support, Sweden
|
4169.5 | Details, details... | VMSNET::P_NUNEZ | | Wed Feb 26 1997 09:09 | 6 |
|
What's the old/new values and on what basis did you select the new
value?
Paul
|
4169.6 | 16k default, 64k max | BULAN::peoto.soo.dec.com::ptorlind | | Thu Feb 27 1997 02:28 | 22 |
| Hi !!
The old value "should be" 16k TcpWindowSize for FDDI, default
value in NT. So what we did was to set it first to 64k, maximum
value for fddi in NT. And then it worked so fast that the customer just
went bananas. Then i asked the customer to try lower that value
but he wanted to test the max value for a while to see if it has any
effect on anything else. Otherwise he will run with the max value.
Since the TcpWindowSize is not in there by default, you have to enter
it manually. But that, you probably already knew.
With this customers hardware configuration, AS4000, Prioris 6000 with
dubble CPU's and on FDDI. I would suspect that it would not be any
disadvantage to use the maximum 64k value for TcpWindowSize.
Or ?? Anyone have any input on the value 64k ?? Any side effects to ??
Per Torlind / PATHWORKS support, Sweden
|
4169.7 | UCX not the problem | PCSERV::PTORLIND | | Sun Mar 02 1997 10:04 | 11 |
| Hi !!
I got the customer to set the ucx parameter back to the default..
DEFACK
And then try again and it still works great, so the problem was that
the default TcpWindowSize is too low for this configuration.
Per Torlind / PATHWORKS support, Sweden
|