| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 53.1 | | rasmodem27.reo.dec.com::eurich | Bruno Eurich, NSIS UK | Fri May 16 1997 12:13 | 22 |
| > A customer has experience a dramatic performance drawback when
> moving remote NT SNA Server to a central location, keeping SNA client
> remote. They were trying to take out SNA traffic from the wide are network
> in order to use just TCP/IP and keep SNA locked up local.
Can you give more details about the "before" and "after" scenarios ?
e.g. are they running SNA _and_ IP directly over the WAN?
Is this 3270 or 5250?
Is the performance degradation on terminal emulation or file transfer?
If, before, they had SNA Servers local to the clients (SNA Server connecting to
the host over the WAN using SNA), then I would expect more WAN traffic
when the servers moved to a central location, as all SNA traffic would be
encapsulated in TCP/IP.
MS' suggestion in this case is to use distributed links between local and remote
SNA Servers. This will compress and multiplex several client connections over
a single logical IP connection.
Bruno Eurich
NSIS UK.
|
| 53.2 | Mainly 3270 | IB002::16.190.32.123::WorkBenchUser | | Mon May 19 1997 08:56 | 16 |
|
Thanks Bruno for your reply.
The customer is a financial institution having more than 250
remote branches using mainly LU0 and 3270.
Initially they were testing MS SNA Server on remote branches
(local to the clients) moving SNA traffic on the WAN; then they tried
to move the SNA Server to the central location (local to the host).
Performance degradation was dramatic on 3270 terminal
emulation, where response time decreased by a factor of 7 as customer
said.
Emilio Garcia-Pozuelo
|
| 53.3 | | rasmodem9.reo.dec.com::eurich | Bruno Eurich, NSIS UK | Mon May 19 1997 12:53 | 19 |
| >
> Initially they were testing MS SNA Server on remote branches
> (local to the clients) moving SNA traffic on the WAN; then they tried
> to move the SNA Server to the central location (local to the host).
>
> Performance degradation was dramatic on 3270 terminal
> emulation, where response time decreased by a factor of 7 as customer
> said.
The main differences between "client to SNA Server" and "SNA Server to
Host" traffic are that the client traffic will be more fragmented and (in total)
slightly larger due to some TCP/IP overhead (approx. up to 10%) so I
certainly wouldn't expect this large performance hit.
I would suggest some form of problem analysis ensuring that there are no
timeouts occurring on the WAN etc.
Bruno Eurich
NSIS UK
|
| 53.4 | Sna Rpc Service runs amok | EDSCLU::JAYAKUMAR | | Mon May 19 1997 20:00 | 5 |
| On a related note, have you noticed in Ver 3.0 that "Sna Rpc Service"
(Snarpcsv.exe) consumes 100% CPU time all the time. Isn't this process which
enables the client talk to the server?
-Jay
|