network driver cannot accept on hp-ux 11.00

I'm having a problem with Informix 7.31UC2 running on HP-UX 11.00. I'm getting the following in the Informix log file several times per day: listener-thread: err = -25573: oserr = 233: errstr = : Network driver cannot accept a connection on the port. When this happens, our users can no longer open any new connections to the database. Ones that are already on are ok, but new connections are not accepted. This condition clears up by itself in a few minutes. The application is making multiple connections to the database per user. There are no messages in any HP-UX logs to correspond to the outages or messages in the Informix logs. HP is saying it's most likely an Informix problem and Informix is blaming HP. Error 233 is ENOBUFS. In the context of a socket not accepting a connection, HP says this is a "generic" type message and that Informix should recover and retry the connection. Informix asked us to bump up some parameters (like shmmax). But, the parameters they gave us were for a large V class and our T class won't accept parameters that large (we're maxxed out on the T). We did not have this issue under 10.20. Does anyone have any suggestions at what to look at from here? Thank You, Nick

Well, not having ever encountered this error, I'll ask the stupid questions: 1) How many connections are you able to make before the failure? 2) What are the values for NETTYPE? Just checking for the obvious... Doug "Nicholas Pieter Heesters Jr." <heesters@copland.udel.edu> wrote in message news:8rdaut$eln$1@copland.udel.edu... > I'm having a problem with Informix 7.31UC2 running on HP-UX 11.00. > I'm getting the following in the Informix log file several times > per day: > > listener-thread: err = -25573: oserr = 233: errstr = : Network > driver cannot accept a connection on the port. > > When this happens, our users can no longer open any new connections > to the database. Ones that are already on are ok, but new connections > are not accepted. This condition clears up by itself in a few > minutes. The application is making multiple connections to the > database per user. There are no messages in any HP-UX logs to > correspond to the outages or messages in the Informix logs. > > HP is saying it's most likely an Informix problem and Informix is > blaming HP. Error 233 is ENOBUFS. In the context of a socket not > accepting a connection, HP says this is a "generic" type message > and that Informix should recover and retry the connection. > > Informix asked us to bump up some parameters (like shmmax). But, > the parameters they gave us were for a large V class and our T > class won't accept parameters that large (we're maxxed out on the T). > We did not have this issue under 10.20. > > Does anyone have any suggestions at what to look at from here? > > Thank You, > Nick

I need to make a correction. I was looking at the wrong server when I said the version was 7.31UC2. It is really Informix 9.21FC1 that is giving me the problems. The value of NETTYPE is "soctcp,5,30,CPU". As for connections, that really hasn't changed. It's been between 500 - 600. Nick In article <l%qC5.5299$%f7.33529@news2.atl>, Doug Agnew <dagnew@charlottepipe.com> wrote: >Well, not having ever encountered this error, I'll ask the stupid questions: > >1) How many connections are you able to make before the failure? >2) What are the values for NETTYPE? > >Just checking for the obvious... > >Doug

Move to Informix 7.31.UC5? Check network is not saturated/ showing errors netstat -i? ENOBUFS means either a) lack of memory hence no more buffers can be allocated OR b) too much is being buffered i.e. network is not fast enough OR c) streams buffer network parameters are too low OR d) a kernel/network driver bug. I would say it is an HP-UX bug i.e. c) or d) but check a) and b) first Nicholas Pieter Heesters Jr. wrote in message <8rdaut$eln$1@copland.udel.edu>... >I'm having a problem with Informix 7.31UC2 running on HP-UX 11.00. >I'm getting the following in the Informix log file several times >per day: > >listener-thread: err = -25573: oserr = 233: errstr = : Network >driver cannot accept a connection on the port. > >When this happens, our users can no longer open any new connections >to the database. Ones that are already on are ok, but new connections >are not accepted. This condition clears up by itself in a few >minutes. The application is making multiple connections to the >database per user. There are no messages in any HP-UX logs to >correspond to the outages or messages in the Informix logs. > >HP is saying it's most likely an Informix problem and Informix is >blaming HP. Error 233 is ENOBUFS. In the context of a socket not >accepting a connection, HP says this is a "generic" type message >and that Informix should recover and retry the connection. > >Informix asked us to bump up some parameters (like shmmax). But, >the parameters they gave us were for a large V class and our T >class won't accept parameters that large (we're maxxed out on the T). >We did not have this issue under 10.20. > >Does anyone have any suggestions at what to look at from here? > >Thank You, >Nick

In comp.sys.hp.hpux Nicholas Pieter Heesters Jr. <heesters@copland.udel.edu> wrote: > HP is saying it's most likely an Informix problem and Informix is > blaming HP. Error 233 is ENOBUFS. In the context of a socket not > accepting a connection, HP says this is a "generic" type message > and that Informix should recover and retry the connection. That is correct - it could simply mean that the connection was closed prior to the accept() completing. It is a non-fatal error. > Informix asked us to bump up some parameters (like shmmax). But, > the parameters they gave us were for a large V class and our T class > won't accept parameters that large (we're maxxed out on the T). We > did not have this issue under 10.20. Different stack, different timing windows. rick jones -- these opinions are mine, all mine; HP might not want them anyway... :) feel free to email, OR post, but please do NOT do BOTH... my email address is raj in the cup.hp.com domain...

"Nicholas Pieter Heesters Jr." wrote: > > I need to make a correction. I was looking at the wrong > server when I said the version was 7.31UC2. It is really > Informix 9.21FC1 that is giving me the problems. > > The value of NETTYPE is "soctcp,5,30,CPU". As for connections, > that really hasn't changed. It's been between 500 - 600. Umm, but that NETTYPE is allowing for only 150 polling slots! Also it is very inefficient, despite what the manual states, to have tcp poll threads running on CPU VPs, run them as listener threads in NET VPs instead you will get better responsiveness. Also try this: NETTYPE soctcp,3,250,NET Keep the CPU VPs for shared memory polling. Art S. Kagel > Nick > > In article <l%qC5.5299$%f7.33529@news2.atl>, > Doug Agnew <dagnew@charlottepipe.com> wrote: > >Well, not having ever encountered this error, I'll ask the stupid questions: > > > >1) How many connections are you able to make before the failure? > >2) What are the values for NETTYPE? > > > >Just checking for the obvious... > > > >Doug

"Art S. Kagel" wrote: > "Nicholas Pieter Heesters Jr." wrote: > > > > I need to make a correction. I was looking at the wrong > > server when I said the version was 7.31UC2. It is really > > Informix 9.21FC1 that is giving me the problems. > > > > The value of NETTYPE is "soctcp,5,30,CPU". As for connections, > > that really hasn't changed. It's been between 500 - 600. > > Umm, but that NETTYPE is allowing for only 150 polling slots! Also it is > very inefficient, despite what the manual states, to have tcp poll threads > running on CPU VPs, run them as listener threads in NET VPs instead you > will get better responsiveness. Also try this: > > NETTYPE soctcp,3,250,NET> > Keep the CPU VPs for shared memory polling. > > Art S. Kagel Hi, I would like to point out that everybody should make his/her own measurements to decide if poll threads are more efficient running on NET VPs or CPU VPs. In the environments I have worked with so far (mostly Baan and SAP) it is much more efficient to have poll threads configured to run on the CPU VPs (inlined) as you save one context switch for each network message. With respect to the NETTYPE parameters: if there is an actual problem with the number of sessions configured it should show up in 'onstat -g ntu'. You should increase number of users if the q-exceed values are > 0 (for shared memory connections you have to be much more careful to chose a good value for number of sessions). If you have continously 500-600 connections to the database server you may want to check out MaxConnect to offload connection management to a different machine (see http://www.informix.com/informix/products/servers/ids2000/22177_74.pdf ). Hope this helps, Heiko

From France - Sorry For bad English... What is your nettype ? Do you try to use the Alias to connect through another port (the number of cnx / port may be close up -> see your system admin) ? How is your sqlhosts ? Here are some ideas where to search May this help you to FIND. Regards - Philippe "smooth1" <smooth1@iclway.co.uk> a 'crit dans le message news: 39da5c5d_3@news2.vip.uk.com... > Move to Informix 7.31.UC5? > Check network is not saturated/ showing errors > > netstat -i? > > ENOBUFS means either > > a) lack of memory hence no more buffers can be allocated OR > b) too much is being buffered i.e. network is not fast enough OR > c) streams buffer network parameters are too low OR > d) a kernel/network driver bug. > > I would say it is an HP-UX bug i.e. c) or d) but check a) and b) > first > > Nicholas Pieter Heesters Jr. wrote in message > <8rdaut$eln$1@copland.udel.edu>... > >I'm having a problem with Informix 7.31UC2 running on HP-UX 11.00. > >I'm getting the following in the Informix log file several times > >per day: > > > >listener-thread: err = -25573: oserr = 233: errstr = : Network > >driver cannot accept a connection on the port. > > > >When this happens, our users can no longer open any new connections > >to the database. Ones that are already on are ok, but new connections > >are not accepted. This condition clears up by itself in a few > >minutes. The application is making multiple connections to the > >database per user. There are no messages in any HP-UX logs to > >correspond to the outages or messages in the Informix logs. > > > >HP is saying it's most likely an Informix problem and Informix is > >blaming HP. Error 233 is ENOBUFS. In the context of a socket not > >accepting a connection, HP says this is a "generic" type message > >and that Informix should recover and retry the connection. > > > >Informix asked us to bump up some parameters (like shmmax). But, > >the parameters they gave us were for a large V class and our T > >class won't accept parameters that large (we're maxxed out on the T). > >We did not have this issue under 10.20. > > > >Does anyone have any suggestions at what to look at from here? > > > >Thank You, > >Nick > >

Hi, Try increasing the physical (primary) memory. You had mentioned that on 10.20 this problem was not there, but it occurs on 11.00. This is because the kernel has become a bigger beast :-) ENOBUFs are returned when kernel subsystems can't allocate memory and they can't block. Best Regards, Suresh. philippe wrote: > > From France - Sorry For bad English... > What is your nettype ? > Do you try to use the Alias to connect through another port > (the number of cnx / port may be close up -> see your system admin) ? > How is your sqlhosts ? > Here are some ideas where to search > May this help you to FIND. > Regards - Philippe > > "smooth1" <smooth1@iclway.co.uk> a écrit dans le message news: > 39da5c5d_3@news2.vip.uk.com... > > Move to Informix 7.31.UC5? > > Check network is not saturated/ showing errors > > > > netstat -i? > > > > ENOBUFS means either > > > > a) lack of memory hence no more buffers can be allocated OR > > b) too much is being buffered i.e. network is not fast enough OR > > c) streams buffer network parameters are too low OR > > d) a kernel/network driver bug. > > > > I would say it is an HP-UX bug i.e. c) or d) but check a) and b) > > first > > > > Nicholas Pieter Heesters Jr. wrote in message > > <8rdaut$eln$1@copland.udel.edu>... > > >I'm having a problem with Informix 7.31UC2 running on HP-UX 11.00. > > >I'm getting the following in the Informix log file several times > > >per day: > > > > > >listener-thread: err = -25573: oserr = 233: errstr = : Network > > >driver cannot accept a connection on the port. > > > > > >When this happens, our users can no longer open any new connections > > >to the database. Ones that are already on are ok, but new connections > > >are not accepted. This condition clears up by itself in a few > > >minutes. The application is making multiple connections to the > > >database per user. There are no messages in any HP-UX logs to > > >correspond to the outages or messages in the Informix logs. > > > > > >HP is saying it's most likely an Informix problem and Informix is > > >blaming HP. Error 233 is ENOBUFS. In the context of a socket not > > >accepting a connection, HP says this is a "generic" type message > > >and that Informix should recover and retry the connection. > > > > > >Informix asked us to bump up some parameters (like shmmax). But, > > >the parameters they gave us were for a large V class and our T > > >class won't accept parameters that large (we're maxxed out on the T). > > >We did not have this issue under 10.20. > > > > > >Does anyone have any suggestions at what to look at from here? > > > > > >Thank You, > > >Nick > > > >