Topics

You may encounter problems with TCP/IP MTU (Maximum Transmission Unit) sizing depending on the version of the system's TCP/IP stack (end node and routers) and the types of devices and topologies that make up your network. This can happen when end node systems (Domino servers or Notes clients) are located on Token-Ring or FDDI networks directly, located in a mix with Ethernet network segments, or with WAN or SLIP/PPP dial up connections that you are trying to access.

Using the PING TCP/IP tool, you can verify what the limitations of the network are. Make note that not all PING variations offer the same functions. The PING utility must be able to create variable test packets and be able to set the Don't Fragment flag. which prevents the packet from being fragmented by either the Router or the direct end node system. The Windows 95/98/NT version of PING offer these functions.

Use the table below to base your measurements for the value of the test packet, where the test packet returns acknowledgments for each successful packet, plus one more byte to the test packet, should give you an error indicating the packet needed to be fragmented. This break point is the maximum size the pathway supports. In some cases, the larger test packets return one or two errors. This is not an issue with packet sizing but should be investigated with your network administrators as it is an indication of a network health problem.

Depending on your TCP/IP stack and your network devices you may need to set the MTU size manually. Most TCP/IP stacks use Maximum Segment Size (MSS) discovery to learn the local segment (routed segment) TCP data size, which it then translates into the IP packet size. By default, 576 is used when the TCP/IP MSS can not be discovered. Newer TCP/IP stacks use the MTU path discovery method to learn the limits of the entire pathway. In some cases, these mechanisms fail to offer the needed constraint or prevent the effective use of the topologies abilities. Some TCP/IP stacks may need to be manually set or tuned for the local segment topology as their default setting is set the minimum value (576), which is not recommended for most LAN networks (Ethernet, Token-Ring or FDDI). Make sure the stack needs to be set, the size is correct for the LAN topology/ies in use, and any WAN links are set to smaller values. If you need to lock down the MTU manually the use the following guidelines on the needed sizes.

Topology/ Frame type	Frame size	IP packet/ MTU Size	Ping test packet size	Comments
ARPA or SLIP	1024	1006	978	Still used with some old Routers supporting ARPA framing (rare)
Ethernet/ DIX or PPP	1518	1500	1472	Preferred size for Ethernet networks
Ethernet/ 802.2 SNAP	1518	1492	1464	Rarely used in Ethernet only networks
Token-Ring/ 802.2 SNAP	1522	1500	1472	Optimized for TCP/IP crossing Eth w/DIX to TR w/Bridges or Routers
Token-Ring/ 802.2 SNAP	1518	1492	1464	Optimized for TCP/IP crossing Eth w/SNAP to TR w/Bridges or Routers (rarely used)
Token-Ring/ 802.2 SNAP	2048	2022	1994	Required for older 4Mb adapters & networks using them.
Wide-Band/ Frame-Relay	N/A	2048	2020
Token-Ring/ 802.2 SNAP	4096	4070	4042	Optimized for performance
Token-Ring/ 802.2 SNAP	4202	4176	4148	Optimized for TCP/IP with NetWare 3.x/4.x servers
Token-Ring/ 802.2 SNAP	4500	4474	4446	Default size for must Routers
FDDI/802.2 SNAP	1542	1500	1472	Optimized for TCP/IP crossing Eth w/DIX to FDDI w/Bridges or Routers
FDDI/802.2 SNAP	1526	1492	1464	Optimized for TCP/IP crossing Eth w/SNAP to FDDI w/Bridges or Routers (rarely used)
FDDI/802.2 SNAP	4096	4054	4026	Optimized for performance
FDDI/802.2 SNAP	4104	4070	4042	Optimized for TCP/IP crossing TR to FDDI w/Bridges or Routers
FDDI/802.2 SNAP	4202	4160	4132	Optimized for TCP/IP with NetWare 3.x/4.x servers
FDDI/802.2 SNAP	4440	4474	4446	Optimized for TCP/IP crossing TR to FDDI w/Bridges or Routers
FDDI/802.2 SNAP	4394	4352	4324	RFC 1188 IETF standard
FDDI/802.2 SNAP	4500	4458	4430	Optimized for TCP/IP using full FDDI packets (not recommended)

Note: Token-Ring allows larger frame/packet sizes than what is listed here. In most cases, these are the values used in your network. If you require other values use the formulas listed below.

Note: Refer to your OS or TCP/IP stack vendor for details of altering the MTU or TCP window size as required.

The following formulas are used to derive the values in the above table:

Frame/IP packet sizing

Ethernet w/DIX Frame size - MAC headers (18) = IP packet size
Ethernet w/LLC Frame size - MAC headers (18) - LLC/SNAP headers (8) = IP packet size
Token-Ring Frame size - MAC headers (18) - LLC/SNAP headers (8) = IP packet size
FDDI Frame size - MAC headers {ANSI standard} (34) - LLC/SNAP headers (8) = IP packet size

Ping test data size

IP packet - IP headers (30) - UDP ( TCP) headers (12) = Ping test packet size (TCP data size)

Note: All measurements are in octets. Bytes is quite often used as the term but technically it is not the same.

Within a flat network, make sure the settings of the Notes and Domino server systems are set for the largest Frame/Packet size workable for the LAN topology. If you have a Switched, Bridged or Routed network with either Token-Ring or FDDI, it is recommended you alter the Frame/Packet sizing to 4096 to better match the Switches, Bridges, and Routers memory buffers. With a mixed topology network, you many need to use one of the optimized choices listed in the table above. Here are the guidelines you should follow:

• Transparent Bridge or Straight-Routed networks with Token-Ring or FDDI: Use the table entries as needed.
• With Token-Ring Source Route Bridged networks: An additional 30 bytes must be subtracted to account for the RIF field in the IP packet and MTU settings. This only applies to Token-Ring or FDDI network values and ones that are not optimized for Ethernet.
• Translational Bridge between Token-Ring or FDDI and Ethernet networks: Use the optimized listings that offer the same IP packet size as required (common size).
• Transparent or Source Route Bridges between FDDI and Token-Ring: Use the optimized listings as required.
• ATM with either ClassicIP or LANE should be matched to the same LAN topology the other end node is using.
• WAN links using ATM should use smaller packet sizes to compensate for the BER of the link.
• With Wide area networks using T1/E1, T3/E3 Wide Band links or Frame-Relay links, match them to the LAN topologies MTU or alter the LAN topologies MTU to match Wide Band or Frame-Relay link.

• Between Token-Ring or FDDI to Ethernet networks, or between Token-Ring to FDDI networks: Use the optimized listings that offer the same IP packet size as required (common size).
• Between Token-Ring or FDDI networks accessing across a Leased Line, Fractional T1/E3, T1/E1, Fractional T3/E3, T3/E3 or Frame-Relay network: Use the optimized listings that offer the same IP packet size as required (common size). With these networks we strongly recommend multi-homing so the local LAN traffic accesses a separate NIC than the remote traffic. which requires the tuned interface.
• When using GEO stationary satellite up/down link access you need to multi-home the Domino server with separate NICs at each location so the MTU and TCP window sizes can manually tuned for the satellite path latency.

It is recommended if you are using the NetWare server as an TCP/IP Router with Token-Ring to append "Set Maximum Physical Receive Packet Size = 4096" to the STARTUP.NCF file and follow the needed alterations to the Domino server and Notes client system TCP/IP stacks for a 4096 frame size. This ensures all the end nodes that are accessing across the NetWare server are using the 4096 frame limit.

Note: Your network may require different values if there are network segments set up with lower frame sizes. Consult your network administrator to discuss any possible conflicts, and use the largest value your network can support.</TD></TABLE>