Network Configuration

This chapter describes the parameters to set up your network interface. These parameters are in the TCPCFGxx member of the PARM data set. Parameters for modifying protocol configuration are also in this file, but are described in TCP, UDP, RAW and IP Protocol Configuration (TCPCFGxx).

This chapter includes these sections:

• TCP Stack Configuration Member (TCPCFGxx)

Describes network interface and protocol-specific information

• Defining Physical Medium

Describes how to use the MEDIA statement

• Defining IP AddressingNETWORK Statement

Describes how to use the MEDIA statement

• Driver Statements

Describes how hardware statements are specified

• CDLC Statement

Describes how to configure using the CDLC protocol

• CETI Driver Configuration

Describes how configure to use CETI devices

• CLAW Driver Configuration

Describes how to use the CLAW driver

• HYPERchannel Configuration

Describes how to interface to a HYPERchannel network

• LCS Configuration Parameters

Describes how to use a 3172 Interconnect Controller and the IBM 8232 LAN Channel Station

• IBM 3172 and 8232 Configuration

Describes how to use the LINK statement for link level adapters such as 3172 or 8232

• ARP Configuration

Describes how to configure for Address Resolution Protocol

TCP Stack Configuration Member (TCPCFGxx)

The TCPCFGxx member in the PARM data set specifies the main configuration parameters for the TCP stack.

Generally, the first statement of TCPCFGxx is the MEDIA statement which defines the physical medium that Cisco IOS for S/390 will be physically attached to. The MEDIA statement is followed by one or more NETWORK statements, zero or more driver statements (CETI, CLAW, HYPER, LINK), and zero or more ARP statements.

Multiple MEDIA statements are allowed for multihomed configurations; however, each must be followed by its own NETWORK, driver and ARP statements.

The NETWORK statement defines the IP addressing of the media. If a subnets-are-local configuration is desired, multiple NETWORK statements may follow the MEDIA statement. Multiple Virtual IP Addresses are defined by specifying a MEDIA VIRTUAL statement, followed by one or more NETWORK statements. IP address 127.0.0.1 is reserved for loopback and cannot be coded.

The driver statements follow the MEDIA statement and may be before or after the NETWORK statements. Driver statements are not permitted if the Media type is Virtual. Multiple driver statements are permitted if multiplexing is desired. The LINK statement requires a prior LCS statement. The LCS statement can appear anywhere in TCPCFGxx prior to the LINK statement.

ARP statements follow the driver statements if static ARP resolution is used on this Media. No ARP statements are required if dynamic ARP is being used (recommended).

ROUTE statements may be placed anywhere after the MEDIA statement, or for simplicity, all ROUTE statements may be grouped together if the MEDIANAME keyword is specified on each ROUTE statement.

MEDIANAME links MEDIA, NETWORK, driver, ARP and ROUTE statements. If MEDIANAME is not used, statements are determined by their placement in the TCPCFGxx member. Read MEDIA Statement Usage Notes for more information about using MEDIANAME.

Defining Physical Medium

The MEDIA statement defines a physical medium that Cisco IOS for S/390 is attached to. The MEDIA statement is followed by one or more NETWORK statements, zero or more driver statements (CETI, CLAW, HYPER, and LINK), and zero or more ARP statements.

MEDIA Statement

Syntax Description

MEDIA Statement Usage Notes

Using the MEDIA Statement in Fault Tolerant Networks

A typical installation requires only one each of a MEDIA statement, a NETWORK statement, and a driver statement. For fault tolerance, at a minimum, you will need multiple driver statements. If you are supporting multiple network numbers on the same physical medium, you will need multiple NETWORK statements to indicate this. A MEDIA statement should correspond to a physical medium such as Ethernet, token ring, or FDDI, and there should be one MEDIA statement for each physical medium this host is attached to.

By default, NETWORK statements following a MEDIA statement are associated with that media. This can be overridden with the MEDIANAME keyword, but it makes the configuration difficult to follow. It is recommended that you keep the NETWORK statements organized under their related MEDIA statement.

Similarly, driver statements are related to a MEDIA statement, and represent the physical related interfaces to the related media. By default, driver statements following a MEDIA statement are associated with that media. As for the NETWORK statements, this can be overridden with the MEDIANAME keyword, but it makes the configuration difficult to follow. It is recommended that you keep your driver statements organized under their related MEDIA statement.

There is no relationship between NETWORK and driver statements. That is, all drivers under a MEDIA statement can respond to all IP addresses for all the NETWORK statements related to that media.

Maximum Transmission Unit

The Maximum Transmission Unit (MTU) that can be transmitted between two nodes on a local network generally is fixed for the entire network. For Ethernet, this value is 1500 data bytes excluding the local network header, and for DDN and ARPANET, this value is fixed at 1007. For HYPERchannel, there is no limit imposed by the lower-level protocols, and the remote host determines the MTU by how much it is willing to receive (see ARP statement). FDDI networks have an MTU of 4352 as well as 16 MB Token Ring networks. The 4 MB Token Ring networks have an MTU of 2002.

Maximizing Throughput

Throughput can be maximized by sending data packets that are as large as the MTU for the local network. TCP uses the Maximum Segment Size (MSS) option to regulate the size of TCP segments transmitted on a TCP connection, which in turn determines the maximum size of packets generated by IP. Generally, sending the largest TCP segment possible is more efficient because the total number of segments transmitted is reduced. However, if the communicating hosts are not on the same network (or subnetwork), intervening gateways might need to fragment the packets, which can be less efficient than sending smaller segments in the first place.

The MSSOPT parameter controls when the local TCP uses the MSS TCP option to increase the size of TCP segments it is willing to receive (the default is the MTU). If permitted, TCP increases its receive segment size to a value that lets the remote TCP send segments that are optimal for the local network and will never transmit a segment larger than that advertised by the remote TCP.

Maximum Receive Segment Size

The MSSDEF parameter determines what the maximum receive segment size can be if the MSSOPT parameter does not permit using the optimum value for the local network. This permits using a value larger than the Internet-wide default and smaller than the MTU for the local network. For example, if several local area networks are interconnected by a wide area network whose MTU is smaller than that of the local area networks, MSSDEF should be set to the MTU of the wide area network. When used with the appropriate MSSOPT parameter, TCP can use larger segments while avoiding fragmentation.

The MSSDEF value is specified in terms of the IP packet size and not the actual TCP segment size. In other words, the value specified for MSSDEF should be the largest network packet, excluding the network header but including the normal TCP and IP headers, that can be generated with a maximum size TCP segment.

Defining IP Addressing--NETWORK Statement

The NETWORK statement defines the IP addressing of the media. If a subnets-are-local configuration is desired, multiple NETWORK statements may follow the MEDIA statement. Multiple virtual IP addresses are defined by specifying a MEDIA VIRTUAL statement, followed by one or more NETWORK statements. IP address 127.0.0.1 is reserved for loopback and cannot be coded.

NETWORK Statement

Syntax Description

Network Statement Usage Notes

Consider these usage notes when using the NETWORK statement:

Local Host Name

The NETWORK statement defines the local Internet address by which remote hosts access this host through the local network.

Subnet and Network Masks

If the local network is a subnet of a larger network, a subnet mask must be specified with the SUBNET parameter to define which bits of the Internet address contain the subnet number.

Configuring a Single LNI to Act as Multiple Hosts

Cisco IOS for S/390 supports multiple NETWORK statements bound to a single interface. When using this feature, Cisco IOS for S/390 can be configured such that a single LNI can act as multiple hosts, each with a unique IP address, on the same or different nets or subnets.

Cisco IOS for S/390 will report its associated IP address as the address specified on the first NETWORK statement using that LNI. Cisco IOS for S/390 will respond to ARPs for any NETWORK statement pointing to it, and attempt to use the correct IP address on ARPs it generates.

Multihomed Example

MEDIA ETHERNET MTU(1500) NAME(HOST1)
                     .
                     .
                     .
    NETWORK IPADDRESS(192.130.0.12)
    SUBNET(255.255.255.0)
                     .
                     .
                     .
    NETWORK IPADDRESS(192.130.1.12)
    SUBNET(255.255.255.0)
                     .
                     .
                     .
    CETI CUTYPE(3762)
    DEVADDR(400)
                     .
                     .
                     .

Multihome/Multiplex

A local network is bound to a particular hardware interface by specifying one or more driver statements. All NETWORK statements associated with that media will be associated with all drivers for that media.

Here is an example for a multiplexed environment.

MEDIA ETHERNET MTU(1500) NAME(ETHER)
NETWORK IPADDRESS(192.130.0.12) SUBNET(255.255.255.192)
CETI CUTYPE(3762) DEVADDR(400)
LCS NAME(LAN) DEVADDR(500) CUTYPE(3172)
LINK LCSNAME(LAN) ADAPTER(0)

Here is an example for a subnets-are-local and multiplexed environment:

MEDIA ETHERNET MTU(1500) NAME(ETHER)
NETWORK IPADDRESS(192.130.0.12) SUBNET(255.255.255.192)
NETWORK IPADDRESS(192.130.0.14) SUBNET(255.255.255.192)
CETI CUTYPE(3762) DEVADDR(400)
LCS NAME(LAN) DEVADDR(500) CUTYPE(3172)
LINK LCSNAME(LAN) ADAPTER(0)

Here is an example for a multihomed environment:

MEDIA ETHERNET MTU(1500) NAME(ETHER)
NETWORK IPADDR(192.130.0.12) SUBNET(255.255.255.192)
LCS NAME(LAN) DEVADDR(500) CUTYPE(3172)
LINK LCSNAME(LAN) ADAPTER(0)
MEDIA FDDI MTU(4352) NAME(FIBER)
NETWORK IPADDR(192.130.1.12) SUBNET(255.255.255.192)
LINK LSCNAME(LAN) ADAPTER(1)

If multiple drivers are specified for the local network, all of them are capable of responding to the same Internet address. In this case, the host is said to be multiplexed onto the local network. On output, packets are distributed on a round-robin basis. On input, the load sharing mechanism (if any) is determined by the network. Multiplexing can be used to increase throughput or to eliminate single points of failure by providing redundancy.

More than one local network can be defined, each with its own unique Internet address within one Cisco IOS for S/390 address space. Also, the local networks can use different transmission technologies and have different characteristics (for example, Ethernet and FDDI). In this case, the host is said to be multihomed because there are multiple routes to the local host.

NETWORK Example

This example shows the usage of the NETWORK statement:

NETWORK IPADDRESS(26.131.0.2)
        NETMASK(255.0.0.0)
        SUBNET(0.0.0.0)
        METRIC(1) 

Driver Statements

The driver statements follow the NETWORK statements. Driver statements are not permitted if the MEDIA type is VIRTUAL. Multiple driver statements are permitted if multiplexing is desired. The LINK statement requires a previous LCS statement. The LCS statement can appear anywhere in TCPCFGxx prior to the LINK statement.

CDLC Statement

Sending IP directly to the FEP is more efficient, easier to configure, and yields higher performance than encapsulating IP datagrams in SNA Path Information Units (PIUs). IP encapsulation is the approach used in IBM's SNALINK.

You can use a 3745 FEP to attach channel interfaces directly to an IP network. The channel interface in this situation is configured in the Network Control Program (NCP) as a native IP element. The IP channel interface does not fall under control of VTAM.

Cisco IOS for S/390 uses the IBM CDLC protocol to implement support for the IP channel attachment capability. The 3745 FEP appears as a single device, operating in half-duplex mode. The CDLC protocol uses a single subchannel in half-duplex mode.

IP datagrams are passed from an MVS-based IP application over bus-and-tag or ESCON channels to a 3745 FEP. To Cisco IOS for S/390, the channel-attached NCP running native IP support looks like a channel-attached router.

The 3745 FEP cannot share the same bus-and-tag interface between SNA and IP traffic to the mainframe. The native IP attachment requires a dedicated 3745 channel adapter. If you are using ESCON with the 3746-900 frame, IP and SNA traffic can flow over the same channel connection if you create a separate NCP link station for the IP traffic.

CDLC Statement

Use the CDLC Statement to configure your CDLC driver.

Syntax Description

CDLC Statement Usage Notes

Example NCP Coding

This example shows the relevant portions of the NCP used for this configuration.

CETI Driver Configuration

The CETI statement specifies configuration parameters for a channel adapter that implements the CETI.

CETI Statement

Syntax Description

CETI Statement Usage Notes

Control Port Wait Time

The control port wait time (WTIME) controls how much of the channel bandwidth is used by the control port. If there is no activity on the data ports, the control unit pauses for the specified amount of time after every CWRITE command to allow other control units access to the channel. Increasing this value lowers channel utilization but increases response time.

Maximum Message Buffering

The maximum number of full-size network messages that can be buffered is 255. Be aware that these buffers are page-fixed in real storage for as long as the device is active. This will affect system paging activity.

Interrupt Handling

The CETI control unit uses several mechanisms to limit the number of interrupts generated to the CPU. One such mechanism is to delay an attention interrupt after transfer of a message to allow subsequent messages to be transferred as a single burst. Two parameters that control the amount of delay are the buffer threshold and message gap time. The buffer threshold specifies the number of buffers that can be transferred before an attention interrupt is generated. The message gap time specifies the amount of time the control unit waits after a burst of buffers before presenting an attention interrupt, even if the burst contains less than the buffer threshold. Choose the values for these parameters to balance responsiveness with high throughput.

CETI Usage of ARP

CETI uses Address Resolution Protocol (ARP) to map higher-level protocol addresses into lower-level hardware addresses. A dynamic cache of address mappings is maintained for this purpose.

CETI Example

This example shows the usage of the CETI statement:

CETI CUTYPE(3722)
     DEVADDR(0240)
     IBUF(20)
     IPARM(4 16)
     OBUF(20)
     OPARM(18 65535)
     RESTART(30)
     WTIME(128)

CLAW Driver Configuration

Use the CLAW statement to specify configuration parameters for an interface to an RS/6000,
Cisco 7000, or Cisco 7500 series running the Common Link Access to Workstation (CLAW) protocol.

CLAW Protocol

CLAW is a protocol for communicating between a mainframe and a channel-attached workstation. The protocol solves two major problems:

CLAW Statement Syntax

Syntax Description

CLAW Statement Usage Notes

Tracing the Channel Program

Because the channel program is being dynamically updated, the Generalized Trace Facility (GTF) has difficulty tracing it. Use GTF to trace I/O error problems and use TCPEEP or Component Trace to look at data.

Subunit Considerations

The CLAW interface creates a separate subnet link between TCP/IP running on the host and TCP/IP on the workstation. This is considered to be a separate subnet. The host and the workstation are normally the only hosts on the subnet. Multiple TCP/IP address spaces may share a single channel as long as each TCP/IP address space has its own subchannel pair and if the address is defined in the workstation/router. In this case, all TCP/IP address spaces and the workstation share the subnet.

Accessing Networks Attached to the Workstation

To access any network attached to the workstation (in other words any Ethernet, Token Ring, or FDDI ring), the workstation must be defined to Cisco IOS for S/390 as an IP router. Use the ROUTE configuration statement to define the workstation as the default router:

ROUTE DEST(0.0.0.0) ROUTE(a.b.c.d)

where a.b.c.d is the Internet address of the CLAW interface on the workstation.

In addition, routing must be permitted on the workstation. Use the no -a command to view TCP/IP parameters on an RS/6000. Use the no -o ipforwarding=1 command to forward IP datagrams on the RS/6000.

Defining a Path to the Host

To make Cisco IOS for S/390 accessible from other networks or subnets, define either a static or dynamic path to the host via the workstation.

This is an example of a static route command on a UNIX system:

    route add 138.42.136.1 

Where 138.42.128.234 1, 138.42.136.1 is the Internet address of the host, 138.42.128.234 is the Internet address of the workstation, 1 is the hop count.

Dynamic Routing with CLAW

In order to use dynamic routing, the CLAW workstation must be running some type of routing protocol such as RIP and HELLO. GateD (which is preferable) or routed can be run on the workstation to let the workstation announce the presence of the host.

When configuring GateD on the workstation, the CLAW interface must be defined as passive unless GateD is running on the host. In other words, if there is no gateway protocol running on the host, then the GateD on the workstation must not delete the route due to lack of routing responses from the host.

This is what a sample gated.conf file (the configuration file for GateD on the workstation) looks like:

    interface ca0 passive;
    rip supplier {
         interface ca0 noripout noripin;
    };
    static {
        138.42.136.1 gateway 138.42.136.1 preference 0;
    };

LAN Message Headers

IP packets sent and received over the CLAW interface do not have a LAN message header prepended to them as Ethernet, token ring, and FDDI have. This allows the TCP/IP MTU size to be specified up to the size defined by the BUFSIZE parameter on the CLAW statement.

Maximizing Buffer Space

An RS/6000 supports a buffer pool of up to 212,992 bytes, which is fifty-two 4096-byte buffers. If the majority of IP traffic going over the channel is destined for hosts other than the RS/6000, and those hosts are on an Ethernet, consider reducing the block size being transferred across the channel from 4096 to 2048. This reduces the amount of wasted space in each buffer from 63 percent to
26 percent and at the same time doubles the number of buffers.

Set the block size to 2048 on both the host (BUFSIZE(2048)) and on the workstation (Transmit Buffer Size and Receive Buffer Size). This change also allows the maximum number of buffers on the workstation to be increased from 52 to 104 (52 input/52 output).

Specifying the Number of Buffers

The number of input and output buffers defined on the host need not match the number of buffers defined on the workstation. The number of buffers required on the host vary by site. Consider these factors when determining the number of buffers:

• The I/O buffers used for CLAW are permanently page-fixed. Defining a buffer pool that is too large reduces the total amount of real storage available in your system and may increase the total paging rate of your system.

• The TCP/IP receive window size limits the amount of data that can be sent or received in a burst. For instance, a window size of 32,768 requires 9 buffers if the remote host is the CLAW workstation (32,768/4056) or 23 buffers if the remote host is attached on the Ethernet (32,768/1460).

If the majority of IP datagram traffic is comprised of small messages (in other words, not equal to the MTU), an increase in buffers allows for quicker transfer to or from the host. Consider reducing the size of the buffers.

CLAW Example

This example shows the usage of the CLAW statement:

CLAW DEVADDR(0FA0)
     BUFSIZE(4096)
     HOSTNAME(IBMHOST)
     IBUF(26)
     OBUF(26)
     RESTART(120)
     WSNAME(CLAWWS)

HYPERchannel Configuration

Use the HYPER statement to specify configuration parameters for an interface to a 50 MB HYPERchannel network using N220 or EN642 adapters manufactured by Network Systems Corporation.

HYPER Statement

This is the syntax for the HYPER statement.

Syntax Description

HYPER Statement Usage Notes

ARP Usage

The HYPERchannel LNI does not use Address Resolution Protocol (ARP) to map higher-level protocol addresses into lower-level hardware addresses. A static table of address mappings is maintained for this purpose.

The address resolution cache is initialized from a static address resolution table during LNI startup. This table is generated with the ARPTABLE and ARP statements. Static entries in the ARP cache are never removed and never updated. If ARPTABLE has been specified, no action is taken and only hosts defined with ARP statements can be accessed.

HYPER Example

This example shows the usage of the HYPER statement:

HYPER CUTYPE(N220)
      DEVADDR(0400)
      IBUF(20)
      OBUF(12)
      RESTART(60)

LCS Configuration Parameters

The LCS statement in the TCPCFGxx member of the PARM data set specifies LNI configuration parameters for channel attached devices that are compatible with the 3172 Interconnect Controller and the IBM 8232 LAN Channel Station.

LCS Statement

Syntax Description

IBM 3172 and 8232 Configuration

Use the LINK statement to define the link level network adapter as IBM 3172 or 8232 and to associate the link level with the LCS device level specification.

LINK Statement Syntax

Syntax Description

LINK Statement Usage Notes

Statement Order

The LINK statement references a previous LCS statement. The LCS statement must precede the LINK statement.

Adapter Types

The 3172 supports up to 4 adapters, which may be of different types. The following types are supported:

• ETHERNET 10MBps Ethernet--MTU(1500)

• TOKEN4  4MBps Token Ring--MTU(2002)

• TOKEN16  16MBps Token Ring--MTU(4352)

• FDDI  100MBps Token Ring--MTU(4352)

ARP Considerations

When Cisco IOS for S/390 needs to broadcast an addressing resolution packet on a token ring network, there are two methods for determining how the packet traverses token ring bridges. When SINGLERT is specified, the ARP is transmitted using a Single Route Broadcast. When ALLRT is specified, the ARP is transmitted using an All Route Broadcast. SINGLERT should be specified only if the token ring bridges are configured to forward packets with the Single Route indicator set. Setting SINGLERT otherwise may cause portions of the token ring network to be unreachable.

LINK Example

This example shows the usage of the LINK statement and its associated LCS statement:

LCS  NAME(ETHER) DEVADDR(500)
LINK LCSNAME(ELINK)
     ADAPTER(0) MEDIANAME(ETHER)

ARP Configuration

Address resolution protocol is used to dynamically discover hardware addresses corresponding to IP addresses. This information is stored in an ARP table and aged and refreshed as needed. This protocol is normally used over Ethernet, token ring, and FDDI. It is therefore unusual for these media to need static ARP information. If desired, static ARP entries may be created by coding ARP statements associated with a MEDIA statement.

ARP Statement Syntax

Syntax Description

ARP Statement Usage Notes

Specifying Addresses

Standard dot notation is used to specify protocol and hardware addresses where each byte of the address is coded as a decimal number (0:255) and separated from each other by periods.

The specification of the physical hardware address must be consistent with the type of hardware specification on the ARPTABLE statement that begins the table. The hardware address for Ethernet is 48 bits (6 bytes) long, and the hardware address for HYPERchannel is 32 bits (4 bytes) long. The 16-bit HYPERchannel addresses should be coded as 32-bit addresses by putting two bytes of zero in front of the address.

Subnet Numbers

The MEDIANAME will reference a MEDIA statement and its associated statements. The network and subnetwork numbers contained in the protocol address in the ARP statement must be consistent with the local Internet address defined with those NETWORK statements.

ARP Examples

This example shows the usage of the ARP statement:

ARP  PA(27.132.0.1) HA(0.0.17.0)
ARP  PA(27.132.0.2) HA(0.0.34.0)