Re: Time delay on ipcshm connection versus tlitcp
Posted in 2003
Topics: Stored Procedures & SPL, Server Administration, Networking & sqlhosts Configuration, Versions, Editions & End-of-Life
How many CPU VPs do you have? What is the RESIDENT setting? Have you tried
using Jonathan Leffler's sqlcmd in benchmark mode (-B on the commandline or
'benchmark on;' at the prompt) to get timings of the SQL itself independent of
connect time?
Art S. Kagel
----- Original Message -----
From: Tracy Sweeney <sweentra@pentamation.com>
At: 3/11 13:42
> Hello all:
>
> I have a UnixWare 7.1.1 System running IDS 7.31.UD5. I have two
> connections set up as follows:
>
> #TCP/IP Link For GSMS and COGNOS
> aspdb03 ontlitcp aspdb03 sqlexec>
> #Interprocess communications link - For Open Series
> ipcdb03 onipcshm aspdb03 db03sql2>
> When I time a simple sql command with dbaccess I get a significant delay
> on the ipc connection in the amount of real time used. I thought shared
> memory connections were supposed to be faster? Am I missing something in
> my configuration? It looks like the actual command runs quickly and the
> display prints the Database closed line, but then a delay occurs before
> returning to the command line and the real time is very much higher that
> with the tlitcp connection.
>
> support: echo $INFORMIXSERVER
> ipcdb03
> support: time dbaccess pmpfinance /peishare/hosting/sql/sql1.sql
>
> Database selected.
> (count(*))
> 1135
> 1 row(s) retrieved.
> Database closed.
>
> real 0m1.05s <-----
> user 0m0.01s
> sys 0m0.03s
>
> support: echo $INFORMIXSERVER
> aspdb03
> support: time dbaccess pmpfinance /peishare/hosting/sql/sql1.sql
>
> Database selected.
> (count(*))
> 1135
> 1 row(s) retrieved.
> Database closed.
>
> real 0m0.06s <-------
> user 0m0.03s
> sys 0m0.00s
>
> Here are my onconfig entries in the nettype:
> NETTYPE tlitcp,2,200,NET # Configure poll thread(s) for nettype
> NETTYPE ipcshm,2,200,CPU # Configure poll thread(s) for nettype
If you did the test exactly as noted below, without bouncing the
instance
between tests, then maybe you got better results the second time because the
rows selected were already in the buffers?
Paul Mosser
-----Original Message-----
From: ART KAGEL, .... [mailto:KAGEL@bloomberg.net]
Sent: Tuesday, March 11, 2003 11:50 AM
To: ids@iiug.org
Subject: Re: Time delay on ipcshm connection versus tlitcp [647]
How many CPU VPs do you have? What is the RESIDENT setting? Have you tried
using Jonathan Leffler's sqlcmd in benchmark mode (-B on the commandline or
'benchmark on;' at the prompt) to get timings of the SQL itself independent
of
connect time?
Art S. Kagel
----- Original Message -----
From: Tracy Sweeney <sweentra@pentamation.com>
At: 3/11 13:42
< snipped >
>
> real 0m1.05s <-----
> user 0m0.01s
> sys 0m0.03s
>
< snipped >
>
> real 0m0.06s <-------
> user 0m0.03s
> sys 0m0.00s
>
> Here are my onconfig entries in the nettype:
> NETTYPE tlitcp,2,200,NET # Configure poll thread(s) for nettype
> NETTYPE ipcshm,2,200,CPU # Configure poll thread(s) for nettype