Context:

- x86 (P4 HT 512 MB ram)
- Linux: Suse 9.3 (or 10)
- FB SS 2.0RC1 NPTL

Under load (large, io-intensive queries) and many simultaneous
requests we are logging frequent read errors:

INET/inet_error: select in packet_receive errno = 9

INET/inet_error: read errno = 9

You can check this running something like:

echo "select-*-from-big-table;" | /opt/firebird/bin/isql \
localhost:dbalias -u user -p passw &

in parallel, with about 30+ instances (notice that the loopback address
should remove doubts about network hw/quality/bandwidth)