Geralmente, o TCP/IP verifica, de tempos em tempos, se uma conexão está quebrada ou não. Subentende-se por conexões quebradas, aquelas que foram interrompidas de forma anormal, como por exemplo, um cabo de rede sendo desconectado, uma queda de energia desligando os access points de uma rede wi-fi, etc.

Na maioria dos sistemas operacionais, o TCP/IP faz essa checagem a cada 2h. Em muitas situações, esse é um tempo longo demais. Imagine uma transação que estava realizando alguma alteração em um registro do banco de dados, quando a conexão foi interrompida (antes de commitar). Nesse caso, o Firebird só seria notificado da quebra da conexão 2h depois, e durante todo esse tempo, o registro ficaria "travado" para alterações.

A solução é configurar o TCP/IP para fazer a checagem mais frequentemente. Para isso, cada sistema operacional tem sua forma de configuração do TCP/IP. Abaixo segue um artigo extraído do endereço www.gnugk.org/keepalive.html, onde é ensinado como configurar o parâmetro KeepAlive do TCP/IP para diversos sistemas operacionais:

Listamos também outros links que falam sobre esses parâmetros:
No Linux: tldp.org/HOWTO/TCP-Keepalive-HOWTO/usingkeepalive.html
No Windows: learn.microsoft.com/en-us/previous-versions/windows/it-pro/windows-server-2003/cc782936(v=ws.10)

Using TCP keepalive to Detect Network Errors

To detect network errors and signaling connection problems, you can enable tcp keep alive feature. It will increase signaling bandwidth used, but as bandwidth utilized by signaling channels is low from its nature, the increase should not be significant. Moreover, you can control it using keep alive timeout.

The problem is that most system use keep alive timeout of 7200 seconds, which means the system is notified about a dead connection after 2 hours. You probably want this time to be shorter, like one minute or so. On each operating system, the adjustment is done in a different way.

After settings all parameters, it's recommened to check whether the feature works correctly - just make a test call and unplug a network cable at either side of the call. Then see if the call terminates after the configured timeout.

Here are some hints:

Linux systems

Use sysctl -A to get a list of available kernel variables
and grep this list for net.ipv4 settings (sysctl -A | grep net.ipv4).
There should exist the following variables:
- net.ipv4.tcp_keepalive_time - time of connection inactivity after which
                                the first keep alive request is sent
- net.ipv4.tcp_keepalive_probes - number of keep alive requests retransmitted
                                  before the connection is considered broken
- net.ipv4.tcp_keepalive_intvl - time interval between keep alive probes

You can manipulate with these settings using the following command:

sysctl -w net.ipv4.tcp_keepalive_time=60 net.ipv4.tcp_keepalive_probes=3 net.ipv4.tcp_keepalive_intvl=10

This sample command changes tcp keepalive timeout to 60 seconds with 3 probes,
10 seconds gap between each. With this, your application will detect dead tcp
connections after 90 seconds (60 + 10 + 10 + 10).

FreeBSD

For the list of available tcp settings (FreeBSD 4.8 an up and 5.4):

sysctl -A | grep net.inet.tcp

net.inet.tcp.keepidle - Amount of time, in milliseconds, that the (tcp) 
connection must be idle before keepalive probes (if enabled) are sent.

net.inet.tcp.keepintvl - The interval, in milliseconds, between 
keepalive probes sent to remote machines. After TCPTV_KEEPCNT (default 
8) probes are sent, with no response, the (tcp)connection is dropped.

net.inet.tcp.always_keepalive - Assume that SO_KEEPALIVE is set on all 
TCP connections, the kernel will periodically send a packet to the 
remote host to verify the connection is still up.

therefore formula to calculate maximum TCP inactive connection time is 
following:

net.inet.tcp.keepidle + (net.inet.tcp.keepintvl x 8)

the result is in milliseconds.

therefore, by setting
net.inet.tcp.keepidle = 10000
net.inet.tcp.keepintvl = 5000
net.inet.tcp.always_keepalive =1 (must be 1 always)

the system will disconnect a call when TCP connection is dead for:
10000 + (5000 x 8) = 50000 msec (50 sec)

To make system remember these settings at startup, you should add them 
to /etc/sysctl.conf file

Solaris

For the list of available tcp settings:

ndd /dev/tcp \?

Keepalive related variables:
- tcp_keepalive_interval - idle timeout

Example:
ndd -set /dev/tcp tcp_keepalive_interval 60000

Windows

Search Knowledge Base for article ID 120642:
http://support.microsoft.com/kb/120642/EN-US

Basically, you need to tweak some registry entries under
HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters

O Firebird também possui uma parâmetro similar, chamado DummyPacketInterval, que pode ser alterado no arquivo firebird.conf. No entanto, diversas pessoas recomendam utilizar o KeepAlive ao invés de alterar o firebird.conf.