lsm-qnet.so

Transparent Distributed Processing (native QNX network) module

Syntax:

io-pkt ... -p qnet [option[,option]...]

Runs on:

Neutrino

Options:

---

The lsm-qnet.so DLL has many options to alter how Qnet functions (e.g. timeouts, retries, and idle times), but Qnet is optimized to function with its default settings. Don't use these options unless you're trying to overcome an issue related to your environment. Using these options can have a negative effect if used incorrectly. Use commas, not spaces, to separate the options.

---

auto_add=X

When qnet starts up, the builtin io-pkt Ethernet (en) resolver broadcasts its own mappings on all interfaces. This option specifies the time interval, in ticks (see the periodic_ticks option), after which the builtin io-pkt Ethernet (en) resolver rebroadcasts its own mappings on all of its interfaces. This has the effect of automatically populating the /net directory with all of the connected nodes.

The default is 150 ticks, which represents 30 seconds. The value 0 is special because it not only stops the broadcasted transmissions, but it causes unsolicited received broadcasts to be discarded. This has the effect of populating the /net directory only with nodes that applications on the local node specifically open.

bind=enX

Specify the interface (e.g. bind=en0) to use. All /dev/io-net/enX interfaces are used by default. When you specify more than one bind option, Qnet uses all the specified interfaces. This is the fastest packet transport.

---

The combination of bind=en and resolve=dns is invalid.

---

bind=ip

Instead of using raw (DIX blue-book) ethernet packets, encapsulate Qnet packets with an IP header using its registered protocol number. This is useful on larger networks where simple L2 switching isn't possible, and all packets must be routed.

If you use the bind=ip option, you also need to use the resolve=dns option. The resolver is used to map the node name to the IP address; you can't use the default resolver with the bind=ip option.

conn_est_retries=X

The number of times QoS should retry to establish a connection before giving up. The default is 1.

conn_est_timeout=X

The number of periodic ticks before QoS should retransmit a connection-establishment request. The default is 1.

conn_up_idle=X

The number of periodic ticks until QoS should conclude that a connection is idle without any traffic and should be polled with a heartbeat. The default is 50 ticks (10 seconds).

conn_up_retries=X

The number of unanswered connection heartbeats before QoS concludes that a connection is down. The default is 6.

do_crc=X

Enable (1) or disable (0) software-level CRC checking of packets by L4. The default is 0. When you disable CRC checking, it yields the best performance on reliable hardware.

enforce_crc=num

If you use this option in combination with do_crc, only packets that contain a valid CRC are accepted. This option has an effect only when do_crc is also set to 1. Setting enforce_crc to one causes packets that are received without a valid software-level CRC generated by the remote mode (i.e. it's running do_crc=0) to be discarded, because the packet content's integrity is unknown, and could be suspect. The default is zero, which allows received packets without a generated software-level CRC to be processed.

host=hostname

Change the hostname of the machine.

mapany=map_uid

Map any incoming user ID to map_uid and map its group ID to that of map_uid.

maproot=map_uid

If the incoming user ID is 0, map it to map_uid and map its group ID to that of map_uid.

max_num_l4s=num

Specify the number of interfaces. The default is two, and the maximum is four.

max_tx_bufs=num

The number of Tx buffers that Qnet holds in reserve before allocating more; the default is 500. If your application sends large messages, you may want to increase this value for performance. If your application typically sends small messages (most default system traffic is small messages), you may want to decrease this value to save memory.

mount=directory[:[.]domain]]*

Specify a network directory. The default directory is /net. The default domain is either the hostname domain, if it has one, or the directory with the slashes changed to dots and reversed. For example, /net/outside/canada has a domain of canada.outside.net. The first mount is the default directory and domain that the local hostname resolves through.

mtu_en=num

Specify the maximum transmission unit (MTU) of a Qnet packet. The num argument must be greater than 100, and all nodes in network must use the same value. The default is 1500.

no_ack=X

Whether or not to generate and expect ACK packets. These packets are required to guarantee data delivery over networks that may lose packets, e.g. Ethernet. The value of X can be:

0

Generate and expect ACK packets (the default).

1

Don't generate or expect ACK packets. Specify this value only when it isn't possible for a packet to be lost.

---

Configure all hosts on the network to use the same value for the no_ack option.

---

no_slog=X

Specifying a nonzero value to this option instructs Qnet not to log errors or events to slogger. You can use this option to squelch events caused by a noisy network when you're looking for non-network events in the sloginfo output. By default, Qnet logs events to slogger, which corresponds to a zero value for no_slog.

periodic_ticks=X

The number of times per seconds that QoS/L4 should wake up to perform periodic housekeeping tasks. The value must be in the range from 1 to 1000; the default is 5, resulting in a default 200 ms tick.

---

If you change the value of periodic_ticks, you'll affect the timing of all other options that rely on the timer tick.

---

probe_no_l4_time=X

The number of periodic ticks after which QoS should probe a connection on an interface for which there is no mapping for the remote node with a broadcasted packet. The default is the same as the value of conn_up_idle.

qos_per_pri=num
qos_tx_pri=num

The priority to use for the pulse for the QoS periodic transmission thread and the QoS transmission thread.

---

We recommend that you not change these values; we supply the qos_per_pri and qos_tx_pri options for the rare cases where you need to change the priority of the io-pkt subsystems.

---

qos_verbose=X

The level of verbosity of the output related to connection management by the QoS. The default is 0; the bigger the number, the more verbosity. This option is used for diagnosis.

res_retries=X

The number of retries the Ethernet resolver will perform during an attempted node resolution before giving up. The default is 2.

res_ticks=X

The number of periodic ticks before the Ethernet resolver retransmits a node resolution request. The default is 1.

resolve=resolver[:resolver_parameter]

Add to the resolver list for mountpoints that follow.

The following values for resolver are built into the network manager:

• en_ionet — broadcast requests for name resolution on the LAN (similar to the TCP/IP ARP protocol). This is the default.
• dns — Take the node name, add a dot (.) followed by the node domain, and send the result to the TCP/IP gethostbyname() function.
• file — The resolver_parameter is is the name of the file to use; the default is /etc/qnet_hosts. The format of the file is as follows:

  # This is a comment line
  host.domain      addr1[,addr2]
  ...
     

  The host.domain represents a Qnet fully qualified domain name (FQDN). The addr1 and optional addr2 are the interface addresses for the FQDN. For bind=en, the format of an address is xx:xx:xx:xx:xx:xx (the MAC address).

If you specify something else, Qnet attempts to load nr-resolver.so. The default name resolver is en_ionet. For queries how to create nr-resolver.so, please contact QNX support.

---

The following combinations aren't supported: bind=en and resolve=dns bind=ip and resolve=file

---

ret_drvr_rx_buf=X

How L4 should handle the received (rxd) packets:

• 0 — hold onto multiple rxd packets during reassembly (the default). This results in the best performance.
• 1 — make a copy of the data in the packets and immediately return the packet buffers to the driver. This is useful when there's a limited number of packet buffers provided by the driver.

slow_mode=X

The number of ticks after which to forget about the slow transmit mode (i.e. tightly windowed) for a node. The default is 1200, giving 240 seconds or four minutes. The value 0 is special; it disables slow mode entirely.

tx_retries=X

The number of times Qnet should retry a transmission before giving up. The default is 25.

tx_ticks=X

The number of periodic ticks before L4 retransmits a transmission request. The default is 1.

vtag=tag_number

Cause Qnet to insert a four-byte vlan tag into the packet. The tag_number must be greater than zero. If you use this option, Qnet accepts only packets that have a tag value that matches the given tag_number. If the driver being used doesn't support 1518-byte packets, you must also use the Qnet mtu_en=1496 option.

Description:

The lsm-qnet.so is the lightweight version of the QNX manager that implements native Neutrino networking, Transparent Distributed Processing (TDP).

The QoS software layer implements the node-to-node session protocol and handles the selection of transmission media.

L4 is a software layer beneath the QoS that implements an ISO level-4 style transport to provide guaranteed, non-duplicate delivery of data, using the driver layer below it.

---

You can't umount Qnet, but you can create an io-pkt producer module that supports unmounting.

---

If you specify two or more resolve= options in a series, the resolvers form a list of lookups for the directory specified in the subsequent mount= options.

A list of resolvers is terminated by a mount= option. Any resolve= options placed after a mount= option form a new list — they don't add to the previous list.

For example, the following line:

resolve=a,resolve=b,mount=x,mount=y,resolve=c,mount=z

specifies that:

• mount=x has resolvers a and b
• mount=y also has resolvers a and b
• mount=z has only resolver c

Examples:

You cay start Qnet in two ways: either you start it at the same time as io-pkt, or afterward with the mount utility.

Start a network driver, Qnet, and TCP/IP at once, with Qnet using the default DIX blue book ethernet packet type with no CRC checking, and 1024 descriptors for maximum performance:

io-pkt -d speedo transmit=1024,receive=1024 -p qnet -p tcpip

Start io-pkt, and then use the mount command to start the driver and Qnet in sequence, using the actual file names of the shared libraries:

io-pkt
mount -T io-pkt -o transmit=1024,receive=1024 devn-speedo.so
mount -T io-pkt lsm-qnet.so

These shared libraries (with the .so extension) are actually located in /lib/dll. The mount utility automatically looks for them there. If you wish, you can give the full pathname to the mount utility.

The following example shows how you start everything at once using IP packet encapsulation instead of DIX blue book:

io-pkt -d speedo transmit=1024,receive=1024 -p qnet bind=ip,resolve=dns -p tcpip

See mount and io-pkt for more information.

Files:

/dev/io-net

The directory where, by default, protocol modules and legacy io-net drivers add entries. For more information, the documentation for io-pkt.

/etc/system/config/useqnet

If this file exists, your system is using the default startup files, and io-pkt is running when your system starts up, the system automatically loads lsm-qnet.so. For more information, see the Controlling How Neutrino Starts chapter of the Neutrino User's Guide.

/proc/qnetstats

An entry that Qnet places in the /proc filesystem. If you open this name and read from it, the Qnet resource manager code responds with the current statistics for Qnet.

Caveats:

Qnet doesn't fully support communication between a big-endian machine and a little-endian machine. However, it does work between machines of different processor types (e.g. ARMLE, x86) that are of the same endian-ness. If you require cross-endian networking with Qnet, please contact your QNX sales representative.

See also:

devn-*, devnp-*, io-pkt

Using Qnet for Transparent Distributed Processing in the QNX Neutrino User's Guide

Native Networking (Qnet) in System Architecture

Transparent Distributed Processing Using Qnet in the QNX Neutrino Programmer's Guide

QNX Neutrino Core Networking User's Guide