Symptom:
Slow memory leak in small/middle I/O buffers. This can be identified by looking at the output of "show buffer" and "show buffer usage" commands

You'll see the number of small and middle buffers incrementing to very high values

VG224-1#sh buffer | inc peak
Small buffers, 104 bytes (total 1116, permanent 50, peak 1242 @ 00:00:17):
Middle buffers, 600 bytes (total 1937, permanent 25, peak 2217 @ 00:00:16):

The output of 'show buffer usage' will show the SCCP Application as a Resource User of the buffers and increasing until memory is exhausted. 

Caller pc    : 0x6238D4C8 count:     4454
Resource User: SCCP Appli count:     4455

Once memory is exhausted, telnet sessions will fail to establish. Console access may still be available.

Conditions:
The issue was first found on VG224 registered to CUCM and defined as a SCCP controlled gateway. This is seen when the CUCM rejects the registration attempts of the VG224 FXS ports due to it reaching the " Maximum Number of Registered Devices" value as defined in the CUCM Service Parameters. 

This can occur when devices fail-over from the primary to secondary CUCM and the proper device sizing has not been followed as per the CUCM SRND. Too many devices attempt to register and CUCM starts to reject their attempts.