EPICS Controls Argonne National Laboratory

Experimental Physics and
Industrial Control System

1994  1995  1996  1997  1998  1999  2000  2001  2002  2003  <20042005  2006  2007  2008  2009  2010  2011  2012  2013  2014  2015  2016  2017  2018  2019  2020  2021  2022  2023  2024  Index 1994  1995  1996  1997  1998  1999  2000  2001  2002  2003  <20042005  2006  2007  2008  2009  2010  2011  2012  2013  2014  2015  2016  2017  2018  2019  2020  2021  2022  2023  2024 
<== Date ==> <== Thread ==>

Subject: RE: Buffer problems
From: "Jeff Hill" <[email protected]>
To: "'Andrew Johnson'" <[email protected]>, "'Mark Rivers'" <[email protected]>
Cc: <[email protected]>
Date: Wed, 24 Mar 2004 16:21:08 -0700

> I'm not quite sure what causes ENOBUFS errors when
> sending CA beacons; I have seen similar errors on IOCs here that have
> never run out of DataPool or SysPool buffers, and I don't yet 
> know what they're complaining about (any ideas Jeff?).  

The particular type of PPC SBC used at the SNS as been observed to go into
an ENOBUF error producing state when there is shortage of driver buffer
pool. Dave Thompson has observed that an IOC can get into this state by
sending large ping messages. CA routinely sends UDP messages with a 1472
byte payload. Here is an excerpt from his message on the subject.

----------------------snip-snip--------------------------------------

(You) don't need to write a test program, you can do it with ping and some
flags. The number of 1520 byte buffers available to the driver is small in
comparison to the number of buffers available to the IP stack, both inbound
and outbound.  I think tNetTask has to look at the inbound packets in order
to route them into the IP stack so I would guess, and saw with ping, that
UDP broadcasts that have large contents that span multiple link layer
packets can kill the inbound driver queue.  Anything that blocks interrupts
or the NIC's access to the PCI bus could be a problem also.

----------------------snip-snip--------------------------------------

> They do appear to be benign though, if my memory is correct - the 
> loss of a beacon can cause some unnecessary CA searches to occur, 
> but nothing should break.
> 

In my experience, the ENOBUF message is typically symptomatic of a global
communication problem in the IP kernel. The UDP send attempt tends to fail
out quickly if there are buffering problems whereas the TCP send tends to
block - I suspect.

Jeff


References:
Re: Buffer problems Andrew Johnson

Navigate by Date:
Prev: Re: Buffer problems john sinclair
Next: RE: Buffer problems Mark Rivers
Index: 1994  1995  1996  1997  1998  1999  2000  2001  2002  2003  <20042005  2006  2007  2008  2009  2010  2011  2012  2013  2014  2015  2016  2017  2018  2019  2020  2021  2022  2023  2024 
Navigate by Thread:
Prev: Re: Buffer problems Andrew Johnson
Next: RE: Buffer problems Mark Rivers
Index: 1994  1995  1996  1997  1998  1999  2000  2001  2002  2003  <20042005  2006  2007  2008  2009  2010  2011  2012  2013  2014  2015  2016  2017  2018  2019  2020  2021  2022  2023  2024 
ANJ, 10 Aug 2010 Valid HTML 4.01! · Home · News · About · Base · Modules · Extensions · Distributions · Download ·
· Search · EPICS V4 · IRMIS · Talk · Bugs · Documents · Links · Licensing ·