NVIDIA UFM-SDN Appliance User Manual v4.12.1
NVIDIA UFM-SDN Appliance User Manual v4.12.1

Appendix – UFM Subnet Manager Default Properties

The following table provides a comprehensive list of UFM SM default properties.

Category

Property

Config File Attribute

Default

Mode/ Field

Description

Generic

Subnet Prefix

subnet_prefix

0xfe80000000000000

RW

Subnet prefix used on the subnet 0xfe80000000000000

LMC

lmc

0

RW

The LMC value used on the subnet: 0-7

Changes to the LMC parameter require a UFM restart.

SM LID

master_sm_lid

0

Force LID for local SM when in MASTER state

Selected LID must match configured LMC

0 disables the feature

Keys

M_Key

m_key

0x0000000000000000

RW

M_Key value sent to all ports -used to qualify the set(PortInfo)

M_Key Lease Period

m_key_lease_period

0

RW

The lease period used for the M_Key on the subnet in [sec]

SM_Key

sm_key

0x0000000000000001

RO

SM_Key value of the SM used for SM authentication

SA_Key

sa_key

0x0000000000000001

RO

SM_Key value to qualify rcv SA queries as 'trusted'

Partition enforcement

part_enforce

  • Out

  • In

  • Both (default- outbound and inbound enforcement enabled)

RO

Partition enforcement type (for switches)

MKEY lookup

m_key_lookup

FALSE

RW

If FALSE, SM will not try to determine the m_key of unknown ports.

M_Key

Per Port

m_key_per_port

FALSE

RW

When m_key_per_port is enabled, OpenSM will generate an M_Key for each port

Limits

Packet Life Time

packet_life_time

0x12

RW

The maximum lifetime of a packet in a switch.

The actual time is 4.096usec * 2^<packet_life_time>

The value 0x14 disables the mechanism

VL Stall Count

vl_stall_count

0x07

RO

The number of sequential packets dropped that cause the port to enter the VL Stalled state. The result of setting the count to zero is undefined.

Leaf VL Stall Count

leaf_vl_stall_count

0x07

RO

The number of sequential packets dropped that causes the port to enter theleaf VL Stalled state. The count is for switch ports driving a CA or gateway port. The result of setting the count to zero is undefined.

Head Of Queue Life time

head_of_queue_lifetime

0x12

RW

The maximum time a packet can wait at the head of the transmission queue. The actual time is 4.096usec * 2^<head_of_queue_lifetime>

The value 0x14 disables the mechanism

Leaf Head Of Queue Life time

leaf_head_of_queue_lifetime

0x10

RW

The maximum time a packet can wait at the head of queue on a switch port connected to a CA or gateway port.

Maximal Operational VL

max_op_vls

3

RW

Limit of the maximum operational VLs

Force Link Speed

force_link_speed

15

(Do NOT change)

RO

Force PortInfo: LinkSpeedEnabled on switch ports.

If 0, do not modify.

Values are:

1: 2.5 Gbps

3: 2.5 or 5.0 Gbps

5: 2.5 or 10.0 Gbps

7: 2.5 or 5.0 or 10.0 Gbps

2,4,6,8-14 Reserved

15: set to PortInfo: LinkSpeedSupported

Limits

Subnet Timeout

subnet_timeout

18 (1second)

RW

The subnet_timeout code that will be set for all the ports.

The actual timeout is 4.096usec * 2^<subnet_timeout>

Local PHY Error Threshold

local_phy_errors_threshold

0x08

RW

Threshold of local phy errors for sending Trap 129

Overrun Errors Threshold

overrun_errors_threshold

0x08

RW

Threshold of credit overrun errors for sending Trap 130

Sweep

Sweep Interval

sweep_interval

10

RW

The time in seconds between subnet sweeps (Disabled if 0)

Reassign Lids

reassign_lids

FALSE (disabled)

RW

If TRUE (enabled), all LIDs are reassigned

Force Heavy Sweep

force_heavy_sweep_window

-1

RW

Forces heavy sweep after number of light sweeps

(-1 disables this option and 0 will cause every sweep to be heavy)

Sweep On trap

sweep_on_trap

TRUE (enabled)

RW

If TRUE every trap 128 and 144 will cause a heavy sweep

Alternative Route Calculation

max_alt_dr_path_retries

4

RW

Maximum number of attempts to find an alternative direct route towards unresponsive ports

Fabric Rediscovery

max_seq_redisc

2

RW

Max Failed Sequential Discovery Loops

Offsweep Rebalancing Enable

offsweep_balancing_enabled

FALSE

RW

Enable/Disable idle time routing rebalancing

Offsweep Rebalancing Window

offsweep_balancing_window

180

RW

Set the time window in seconds after sweep to start rebalancing

Handover

SM Priority

sm_priority

15

RO

SM (enabled). The priority used for deciding which is the master. Range is 0 (lowest priority) to 15 (highest)

Ignore Other SMs

ignore_other_sm

FALSE (disabled)

RO

If TRUE other SMs on the subnet should be ignored

Polling Timeout

sminfo_polling_timeout

10

RO

Timeout in seconds between two active master SM polls

Polling Retries

polling_retry_number

4

RO

Number of failing remote SM polls that declares it non-operational

Honor GUID-to-LID File

honor_guid2lid_file

FALSE

(disabled)

RO

If TRUE, honor the guid2lid file when coming out of standby state, if the guid2lid file exists and is valid

Allowed SM GUID list

allowed_sm_guids

(null)

(disabled)

List of Host GUIDs where SM is allowed to run when specified. OpenSM ignores SM running on port that is not in this list.

If 0, does not allow any other SM.

If null, the feature is disabled.

Threading

Max Wire SMPs

max_wire_smps

8

RW

Maximum number of SMPs sent in parallel

Transaction Timeout

transaction_timeout

200

RO

The maximum time in [msec] allowed for a transaction to complete

Max Message FIFO Timeout

max_msg_fifo_timeout

10000

RO

Maximum time in [msec] a message can stay in the incoming message queue

Routing Threads

routing_threads_num

0

RW

Number of threads to be used for parallel minhop/updn calculations.

If 0, number of threads will be equal to number of processors.

Routing Threads Per Core

max_threads_per_core

0

RW

Max number of threads that are allowed to run on the same processor during parallel computing.

If 0, threads assignment per processor is up to operating system initial assignment.

Logging

Log File

log_file

/opt/ufm/files/log/opensm.log

RO

Path of Log file to be used

Log Flags

log_flags

Error and Info

0x03

RW

The log flags, or debug level being used.

Force Log Flush

force_log_flush

FALSE

(disabled)

RO

Force flush of the log file after each log message

Log Max Size

log_max_size

4096

RW

Limit the size of the log file in MB. If overrun, log is restarted

Accumulate Log File

accum_log_file

TRUE

(enabled)

RO

If TRUE, will accumulate the log over multiple OpenSM sessions

Dump Files Directory

dump_files_dir

/opt/ufm/files/log

RO

The directory to hold the file SM dumps (for multicast forwarding tables for example). The file is used collects information.

Syslog log

syslog_log

0x0

RW

Sets a verbosity of messages to be printed in syslog

Misc

Node Names Map File

node_name_map_name

Null

RW

Node name map for mapping node's to more descriptive node descriptions

SA database File

sa_db_file

Null

RO

SA database file name

No Clients Reregistration

no_clients_rereg

FALSE

(disabled)

RO

If TRUE, disables client reregistration

Exit On Fatal Event

exit_on_fatal

TRUE

(enabled)

RO

If TRUE (enabled), the SM exits for fatal initialization issues

Switch Isolation From Routing

held_back_sw_file

Null

RW

File that contains GUIDs of switches isolated from routing

Enable NVIDIA SHARP support

sharp_enabled

Enabled

RW

Defines whether to enable/disable NVIDIA SHARP on supporting ports.

Multicast

Disable Multicast

disable_multicast

FALSE

(disabled)

RO

If TRUE, OpenSM should disable multicast support and no multicast routing is performed

Multicast Group Parameters

default_mcg_mtu

0

RW

Default MC group MTU for dynamic group creation. 0 disables this feature, otherwise, the value is a valid IB encoded MTU

Multicast

Multicast Group Parameters

default_mcg_rate

0

RW

Default MC group rate for dynamic group creation. 0 disables this feature, otherwise, the value is a valid IB encoded rate

Multicast

Enable incremental multicast routing

enable_inc_mc_routing

FALSE

RW

Enable incremental multcast routing

Multicast

MC root file

mc_roots_file

null

RW

Specify predefined MC groups root guids

QoS

Settings

qos

FALSE

(disabled)

*From UFM v3.7 and on

RW

If FALSE (disabled), SM will not apply QoS settings

Unhealthy Ports

Enabling Unhealthy Ports

hm_unhealthy_ports_checks

TRUE

RW

Enables Unhealthy Ports configuration

Configuration file

hm_ports_health_policy_file

null

RW

Specifies configuration file for health policy

Unhealthy actions

hm_sw_manual_action

no_discover

RW

Specifies what to do with switch ports which were manually added to health policy file

MADs validation

validate_smp

TRUE

RW

If set to TRUE, opensm will ignore nodes sending non-spec compliant MADs. When set to FALSE, opensm will log the warning in the opensm log file about non-compliant node

Routing

Unicatst Routingengine

routing_engine

(null)

RW

By default, ar_updn routing engine is used by the SM.
Supported routing engines are minhop, updn, dnup, ftree, dor, torus-2QoS, kdor-hc, kdor-ghc , dfp, dfp2, ar_updn, ar_ftree and ar_dor.

Randomization

scatter_ports

8

RW

Assigns ports in a random order instead of round-robin. If 0, the feature is disabled, otherwise use the value as a random seed.
Applicable to the MINHOP/UPDN routing algorithms

Randomization

guid_routing_order_no_scatter

TRUE

RO

Do not use scatter for ports defined in guid_routing_order file

Unicast Routing Caching

use_ucast_cache

TRUE

RW

Use unicast routing cache for routing computation time improvement

GUID Ordering During Routing

guid_routing_order_file

NULL

RW

The file holding guid routing order of particular guids (for MinHop, Up/Down)

Torus Routing

torus_config

/opt/ufm/files/conf/opensm/torus-2QoS.con

RW

Torus-2QoS configuration file name

Routing Chains

pgrp_policy_file

NULL

RW

The file holding the port groups policy

topo_policy_file

NULL

RW

The file holding the topology policy

rch_policy_file

NULL

RW

The file holding the routing chains policy

max_topologies_per_sw

1

RO

Defines maximal number of topologies to which a single switch may be assigned during routing engine chain configuration.

Incremental Multicast Routing (IMR)

enable_inc_mc_routing

TRUE

RW

If TRUE, MC nodes will be added to the MC tree incrementally. When set to FALSE, the tree will be recalculated per eachg change.

MC Global root

mc_primary_root_guid/mc_secondary_root_guid

0x0000000000000000 (for both)

RW

Primary and Secondary global mc root guid

Scatter ports

use_scatter_for_switch_lid

FALSE

RW

Use scatter when routing to the switch’s LIDs

updn lid tracking mode

updn_lid_tracking_mode

FALSE

RW

Controls whether SM will use LID tracking or not when updn or ar_updn routing engine is used

Events

Event Subscription Handling

drop_subscr_on_report_fail

FALSE

RW

Drop subscription on report failure (o13-17.2.1)

Event Subscription Handling

drop_event_subscriptions

TRUE

RW

Drop event subscriptions (InformInfo and ServiceRecords) on port removal and SM coming out of STANDBY

Virtualization

Virtualization enabled

virt_enabled

Enabled

RW

Enables/disables virtualization support

Maximum ports in virtualization process

virt_max_ports_in_process

64

RW

Sets a number of ports to be handled on each virtualization process cycle

Router

Router aguid enable

rtr_aguid_enable

0 (Disabled)

RW

Defines whether the SM should create alias GUIDs required for router support for each HCA port

Router path record flow label

rtr_pr_flow_label

0

RW

Defines flow label value to use in multi-subnet path query responses

Router path record tclass

rtr_pr_tclass

0

RW

Defines tclass value to use in multi-subnet path query responses.

Router path record sl

rtr_pr_sl

0

RW

Defines sl value to use in multi-subnet path query responses

Router path record MTU

rtr_pr_mtu

4 (IB_MTU_LEN_2048)

RW

Define MTU value to use in multi-subnet path query responses

Router path record rate

rtr_pr_rate

16 (IB_PATH_RECORD_RATE_100_GBS)

RW

Defines rate value to use in multi-subnet path query responses

SA Security

SA Tnhanced Trust Model (SAETM)

sa_enhanced_trust_model

FALSE

RW

Controls whether SAETM is enabled.

Untrusted GuidInfo records

sa_etm_allow_untrusted_guidinfo_rec

FALSE

RW

Controls whether to allow Untrusted Guidinfo record requests in SAETM.

Guidinfo record requests by VF

sa_etm_allow_guidinfo_rec_by_vf

FALSE

RW

Controls whether to allow

Guidinfo record requests by vf in SAETM.

Untrusted proxy requests

sa_etm_allow_untrusted_proxy_requests

FALSE

RW

Controls whether to allow

Untrusted proxy requests in SAETM.

Max number of multicast groups

sa_etm_max_num_mcgs

128

RW

Max number of multicast groups per port/vport that can be registered.

Max number of service records

sa_etm_max_num_srvcs

32

RW

Max number of service records per port/vport that can be registered.

Max number of event subscriptions

sa_etm_max_num_event_subs

32

RW

Max number of event subscriptions (InformInfo) per port/vport that can be registered.

SGID spoofing

sa_check_sgid_spoofing

TRUE

RW

If enabled, the SA checks for SGID spoofing in every request with GRH included, unless the SLID is from a router port at that request.

Single-root I/O virtualization (SR-IOV) enables a PCI Express (PCIe) device to appear to be multiple separate physical PCIe devices.

UFM is ready to work with SR-IOV devices by default. You can fine-tune the configuration using the SM configuration.

The following arguments are available for ConnectX-5 and later devices:

Argument

Value

Description

virt_enabled

  • 0 – no virtualization support

  • 1 – disable virtualization on all virtualization supporting ports

  • 2 – enable virtualization on all virtualization supporting ports (default)

Virtualization support

virt_max_ports_in_process

Possible values: 0-65535; where 0 processes all pending ports
Default: 64

Maximum number of ports to be processed simultaneously by the virtualization manager

virt_default_hop_limit

Possible values: 0-255
Default: 2

Default value for hop limit to be returned in path records where either the source or destination are virtual ports

UFM can isolate particular switches from routing in order to perform maintenance of the switches with minimal interruption to the existing traffic in the fabric.

Isolating a switch from routing is done via UFM Subnet Manager as follows:

  1. Create a file that includes either the node GUIDs or the system GUID of the switches under maintenance. For example:

    Copy
    Copied!
                

    0x1234566 0x1234567

  2. Import the file into the UFM Appliance. Run:

    Copy
    Copied!
                

    ib sm configuration import held-back-sw

Once SM completes rerouting, the traffic does not go through the ports of isolated switches.

To attach the switch to the routing:

  1. Remove the GUID of the switch from the list of isolated switches defined in Step 1 of the isolation process.

  2. Import the file into the UFM Appliance. Run:

    Copy
    Copied!
                

    ib sm configuration import held-back-sw

Once SM completes rerouting, traffic will go through the switch.

© Copyright 2023, NVIDIA. Last updated on Sep 5, 2023.