Default TTL Values in TCP/IP
Internet's unbroken growth and recent fundamental topology changes within
the international backbone occasionally cause some nasty connectivity
problems. In most cases, the problem could be identified as hosts sending
out IP packets with too small initial TTL (Time To Live) values.
Worst case maximum
hop counts of up to 40 can be observed nowadays. TCP and UDP initial TTL
values should be set to a "safe" value of at least 60 today.
Look here for more background information.
Two RFC's give very clear guidelines for the
implementation of the TTL field in the Internet Protocol:
- Requirements for Internet Hosts -- Communication
Layers (RFC 1122, R. Braden, Oct 89):
- extracts from paragraph 3.2.1.7:
The IP layer MUST provide a means for the transport layer to
set the TTL field of every datagram that is sent. When a
fixed TTL value is used, it MUST be configurable. The
current suggested value will be published in the "Assigned
Numbers" RFC.
and
A fixed value must be at least big enough for the
Internet "diameter," i.e., the longest possible path.
A reasonable value is about twice the diameter, to
allow for continued Internet growth.
Assigned Numbers (RFC 1700, J.
Reynolds, J. Postel, October 1994):
page 64: IP TIME TO LIVE PARAMETER
The current recommended default time to live (TTL) for the
Internet Protocol (IP) [45,105] is 64.
The 'Assigned Numbers' RFC gets periodical updates going back as far
as 1976. It is interesting to point out that the recommendation for a
TTL default value of 64 is more than three years old now. It has been
first published in Assigned Numbers (RFC
1340, J. Reynolds, J. Postel, July 1992).
Disclaimer:
The following list is a best effort overview of some widely used
TCP/IP stacks. The information was provided by vendors and many
helpful system administrators.
We would like to thank all these contributors for their precious help !
SWITCH cannot, however, take any responsibility
that the provided information is correct. Furthermore,
SWITCH cannot be made liable for any damage that may arise by the use
of this information.
+--------------------+-------+---------+---------+
| OS Version |"safe" | tcp_ttl | udp_ttl |
+--------------------+-------+---------+---------+
AIX n 60 30
DEC Pathworks V5 n 30 30
FreeBSD 2.1R y 64 64
HP/UX 9.0x n 30 30
HP/UX 10.01 y 64 64
Irix 5.3 y 60 60
Irix 6.x y 60 60
Linux y 64 64
MacOS/MacTCP 2.0.x y 60 60
OS/2 TCP/IP 3.0 y 64 64
OSF/1 V3.2A n 60 30
Solaris 2.x y 255 255
SunOS 4.1.3/4.1.4 y 60 60
Ultrix V4.1/V4.2A n 60 30
VMS/Multinet y 64 64
VMS/TCPware y 60 64
VMS/Wollongong 1.1.1.1 n 128 30
VMS/UCX (latest rel.) y 128 128
MS WfW n 32 32
MS Windows 95 n 32 32
MS Windows NT 3.51 n 32 32
MS Windows NT 4.0 y 128 128
- AIX
- TCP TTL is 60 but UDP has a too low TTL of 30. The TTL initial
values can be printed out and modified using the command "no". See the
man page for more instructions. Further measures might be required to
make the changes permanent.
- DEC PATHWORKS V5 for DOS and Windows
- TCP and UDP use too low TTLs of 30. These defaults can be modified
in file
C:\PW\PWTCP.INI add line "IPTTL = 64"
FreeBSD 2.1R
TCP and UDP are set to a safe default of 64. The sysctl command can
be used to show and modify (if required) the defaults:
% /usr/sbin/sysctl net.inet.ip.ttl # show parameter
# /usr/sbin/sysctl -w net.inet.ip.ttl=64 # set parameter
Note that the change isn't permanent. One has to put this command
into the /etc/rc or /etc/rc.local scripts to set this parameter after
every reboot.
HP/UX 9.0x
TCP and UDP use too low TTLs of 30. HP Support Center provided the
set_ttl tool to modify these defaults.
HP/UX 10.0x
TCP and UDP values are set to 64.
Irix 5.3, 6.2, 6.3, 6.4
TCP and UDP values are set to 60. The defaults can be modified by
changing the file /var/sysgen/master.d/bsd and rebuilding the kernel.
Irix 6.5
TCP and UDP values are set to 60. The defaults can be modified using
the 'systune' command.
Linux
At least after V1.1.59 (and most probably before) Linux uses "safe"
values of 64 for both TCP and UDP default TTLs. As these values are
"hardcoded", occasional modifications have to be done deep in the
source - change the parameter sk->ip_ttl in file net/inet/af_inet.c to
whatever you think is appropriate for you. Some pointers to find the
position where sk->ip_ttl is set:
Linux V1.12.11: net/inet/af_inet.c line 695
Linux V1.3.45: net/inet/af_inet.c line 678
Linux V2.2.x: net/ipv4/ipconfig.c line 516
MacOS/MacTCP 2.0.x
TCP and UDP values are set to 60.
OS/2 TCP/IP 3.0
TCP and UDP values are set to 64.
OSF/1 V3.2A
TCP TTL is 60 but UDP has a too low TTL of 30.
Modification of TCP/IP stack parameters can be made with netconfig2.3
Solaris 2.x
All versions use a default value of 255 for both TCP and UDP.
SunOS 4.1.3/4
Since version 4.1.3, both TCP and UDP default TTL is set to 60. I
have no info about older releases, but these are obsolete anyhow.
You can find (or modify) the TTL values in
udp_ttl: in /usr/kvm/sys/netinet/in_proto.c
tcp_ttl: in /usr/kvm/sys/netinet/tcp_timer.h by
changing the parameter TCP_TTL
Ultrix
TCP TTL is 60 but UDP has a too low TTL of 30.
Modification of TCP/IP stack parameters can be made with netconfig2.3
VMS/Multinet
All recent versions use a TTL of 64 for TCP and UDP. Should a
modification of these parameters be required, it can be done
as follows:
$ MULTINET SET/KERNEL TCP_TTL newvalue
$ MULTINET SET/KERNEL UDP_TTL newvalue
VMS/TCPware
TCPware for VMS (Alpha or VAX CPUs)
Default TTL = 60 for both TCP & UDP since at least TCPware 4.1-2A
If required for some reason, you can modify the values:
$ RUN TCPWARE:NETCU
NETCU> SET IPDEFAULTTTL default-time-to-live-hops
NETCU> EXIT
VMS/Wollongong Runtime Version 1.1.1.1
TCP TTL uses a safe value of 128, but UDP TTL is set to 32.
There is no way to change the defaults, but a new Runtime Version
2.5 is said to fix the problem (i.e. make the parameters configurable).
VMS/UCX
The latest version uses safe TTL values of 128 for TCP & UDP.
No info on older version available yet.
TTL values can be modified, if needed, by typing the command
$ UCX
UCX> set service/protocol=ip=time_to_live=64
MS Windows for Workgroups
Unsufficient TTL default values for both TCP and UDP. The defaults
can be modified in file SYSTEM.INI, MSTCP section, by adding a line:
DefaultTTL=64
Windows 95
The default TTL is 32 for both TCP and UDP. Modification:
- Double click on program 'regedit.exe' in the Windows95 root directory.
- Go to 'Hkey_Local_Machine\System\CurrentControlSet\Services\VxD\MSTCP.
- Select 'Edit' and in drop-down menu 'new'. Choose a data
format and enter parameter name and value in the appropriate format:
'binary value': DefaultTTL 01 00 00 00
'string format': DefaultTTL 64
Note that regedit adds quotes itself when using string format !
"Regedit-ophobe" people might alternativly have a look at the ttlfix utility.
Windows NT 3.51
The default TTL is 32 for both TCP and UDP. Modification:
- Double click on program 'regedit32.exe' in the '\root\system32' directory.
- Go to 'HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters
- Change the "DefaultTTL" entry to
DefaultTTL REG_DWORD 64
Windows NT 4.0
The default TTL has been raised to 128 for both TCP and UDP according to
http://msdn.microsoft.com/library/backgrnd/html/unicastrouting.htm
- netconfig2.3:
- available from:
ftp://cs.ucsd.edu/pub/csl/Netconfig/netconfig2.3.tar.gz
With netconfig you can display and modify TCP/IP parameters in many
Unix implementations. It does not work for Solaris 2.x however.
Be careful when
installing this tool, by default "make install" sets a "setuid
root" permission, which is probably not exactly what you want.
set_ttl:
An unsupported utility, provided by the HP Support Center, which
lets you change initial TTL values in HP/UX kernels.
HP's set_ttl script is also available
directly from this server.
ttlfix:
A small utility which automagicly changes Windows95 TTL defaults to safe
values. Available from
ftp://ftp.active.ch/pub/internet/windows/ttlfix.zip
ttlfix requires
vbrun300.zip
for startup. Install it in directory '\windows\system' if not yet there.
MAP Data Centre - April '01 - MAP WebMaster
