ALTQ.CONF(5) NetBSD File Formats Manual ALTQ.CONF(5)
NAME
altq.conf -- ALTQ configuration file
DESCRIPTION
The altq.conf file contains a number of lines specifying the behavior of queueing disciplines. Comments start with a # and extend to the end of the line. The altqd(8) program reads /etc/altq.conf at startup and sets up queueing disciplines. BLUE, CBQ (Class-Based Queueing), FIFOQ (First-In First-Out Queue), HFSC (Hierarchical Fair Service Curve), PRIQ (Priority Queueing), RED (Random Early Detection), RIO (RED with IN/OUT), WFQ (Weighted Fair Queueing), and CDNR (Diffserv Traffic Conditioner) can be configured in this file. Interface Commands interface if_name [bandwidth bps] [tbrsize bytes] [sched_type] [discipline-specific-options] The interface command specifies a network interface to be under control of ALTQ. One interface specification is provided for each network inter- face under control of ALTQ. A system configured as a router may have multiple interface specifications. if_name specifies the name of a network interface (e.g., fxp0). bandwidth specifies the interface bandwidth in bits per second. This is the maximum rate that the queueing discipline will allow on this interface. tbrsize specifies the bucket size of a token bucket regulator in bytes. When tbrsize is omitted, the system automatically sets the bucket size using heuristics. The token rate is set to the interface bandwidth specified by the interface command. sched_type Type of a queueing discipline. It must be either blue, cbq, fifoq, hfsc, priq, red, rio, or wfq. If the inter- face has only traffic conditioners and no queueing disci- pline, sched_type can be omitted. Class Command class sched_type if_name class_name parent_name [red|rio] [ecn] [cleardscp] [discipline-specific-options] The class command specifies a packet scheduling class for CBQ, HFSC, or PRIQ. A class specifier must be provided for each packet scheduling class. sched_type Type of queueing discipline. Must correspond to the dis- cipline name in interface specification. if_name Interface name. Must correspond to name in interface specification. class_name Arbitrary name for this class. Must be unique for this interface. parent_name The name of the parent class for this class (for CBQ or HFSC). Parent class must have been previously defined. PRIQ does not have class hierarchy and parent_name must be NULL for PRIQ classes. red Use RED (Random Early Detection) on this class queue. RED drops packets with the probability proportional to the average queue length. rio Use RIO (RED with In/Out bit) on this class queue. RIO runs triple RED algorithms at the same time. ecn Use RED/ECN (Explicit Congestion Notification) on this class queue (experimental implementation). ECN implies RED. cleardscp Clear diffserv codepoint in the IP header. Filter Commands filter if_name class_name [name fltr_name] [ruleno num] filter_values The filter command specifies a filter to classify packets into a schedul- ing class. A filter specifier determines any statically-defined packet classification rules. if_name Name of a network interface (e.g., fxp0). class_name Name of a class or a conditioner to which matching packets are directed. name Add an arbitrary name to the filter for a future refer- ence. ruleno Specifies explicit order of filter matching. Filter matching is performed from a filter with a larger ruleno. Default is 0. filter_value should be in the following format: filter_values: dst_addr [netmask mask] dport src_addr [netmask mask] sport proto [tos value [tosmask value]] [gpi value] Here dst_addr and src_addr are dotted-decimal addresses of the destination and the source respectively. An address may be followed by netmask keyword. dport and sport are port numbers of the destination and the source respec- tively. proto is a protocol number defined for IP packets (e.g. 6 for TCP). tos keyword can be used to specify the type of service field value. gpi key- word can be used to specify the Security Parameter Index value for IPsec. When filter value 0 is used, it is taken as a wildcard. filter6 if_name class_name [name fltr_name] [ruleno num] filter6_values The filter6 command is for IPv6. filter6_value should be in the following format: filter6_values: dst_addr[/prefix_len] dport src_addr[/prefix_len] sport proto [flowlabel value] [tclass value [tclassmask value]] [gpi value] Here dst_addr and src_addr are IPv6 addresses of the destination and the source respectively. An address may be followed by an optional address prefix length. dport and sport are port numbers of the destination and the source respectively. proto is a protocol number defined for IPv6 packets (e.g. 6 for TCP). flowlabel keyword can be used to specify the flowlabel field value. tclass keyword can be used to specify the traffic class field value. gpi key- word can be used to specify the Security Parameter Index value for IPsec. When filter value 0 is used, it is taken as a wildcard. CBQ Commands CBQ (Class Based Queueing) achieves both partitioning and sharing of link bandwidth by hierarchically structured classes. Each class has its own queue and is assigned its share of bandwidth. A child class can borrow bandwidth from its parent class as long as excess bandwidth is available. interface if_name [bandwidth bps] [tbrsize bytes] [sched_type] [efficient] if_name specifies the name of a network interface (e.g., fxp0). bandwidth specifies the interface bandwidth in bits per second. tbrsize specifies the bucket size of a token bucket regulator in bytes. sched_type must be either cbq, cbq-wrr (weighted-round robin) or cbq-prr (packet-by-packet round robin). cbq is equivalent to cbq-wrr. efficient Enables CBQ's link efficiency mode. This means that the scheduler will send a packet from the first overlimit class it encounters of all classes of the link-sharing structure when all classes are overlimit. This will also cause the scheduler to use greater than its assigned bandwidth, if the link is capable of more than the assigned bandwidth. By default, this mode is turned off. By adding the keyword efficient to the interface specification line, enables this mode. class sched_type if_name class_name parent_name [admission cntlload|none] [priority pri] [pbandwidth percent] [exactbandwidth bps] [borrow] [default] [control] [maxburst count] [minburst count] [maxdelay msec] [packetsize bytes] [maxpacketsize bytes] [red|rio] [ecn] [flowvalve] [cleardscp] The class command specifies a CBQ class. The classes are organized as a hier- archy, and every class, except for the root class, has a parent. sched_type must be cbq for a CBQ class. if_name Interface name. Must correspond to name in interface specifi- cation. class_name Arbitrary name for this class. Must be unique within the class hierarchy for this interface. The name ctl_class is a reserved class name. parent_name The name of the parent class for this class or NULL if this is the root class. Parent class must have been previously defined. admission The type of admission control and QoS type. cntlload is con- trolled load service for RSVP, otherwise, it should be none. The default is none. priority High numbers are higher priority. Max value is 7 and Min value is 0. Default is 1. pbandwidth The percentage of the interface bandwidth allocated to this class. Generally should add up to 100 percent at each level of the class hierarchy, although other amounts can be speci- fied for purposes of experimentation. exactbandwidth Specify the bandwidth in bits-per-second instead of pbandwidth. Note that the bandwidth allocation of CBQ is not so precise but this is just a way to pass a parameter to CBQ; the user is supposed to know the detailed internals of CBQ. pbandwidth is a preferred way to specify the bandwidth of a class. borrow The class can borrow bandwidth from its parent class when this class is overlimit. If this keyword is not present, then no borrowing is done, and the packet is delayed or dropped when the class is overlimit. default Specify the default class. When this keyword is present, all packets that do not match some classification criteria are assigned to this class. Must be exactly one class on each interface defined as the default class. control Specify the control class. When this keyword is present, the predefined control class packets (RSVP, IGMP, and ICMP) are assigned to this class. Note that when the control class is not specified by the time the default class is created, one is automatically created with default parameters. Thus, if the control class is specified, it must be listed before the default class. Must be exactly one class on each interface defined as the control class. maxburst The maximum burst of back-to-back packets allowed in this class. Default is 16 but the default value is automatically reduced to 4 when the class bandwidth is small (about less than 1Mbps). minburst The minimum burst is used to obtain the steady state burst size. It's the parameter to help compute offtime for the class. Offtime is the amount of time a class is to wait between packets. Default is 2. maxdelay The maxdelay is specified in millisecond and used to obtain the max queue size of the class. If not specified, the default max queue size (30 packets) is used. packetsize The average packet size in bytes to be used in CBQ over-/under-limit computations. Default value is MTU of the interface. maxpacketsize The maximum packet size in bytes for the class. Default value is MTU of the interface. red enables RED on this class queue. rio enables RIO on this class queue. ecn enables RED/ECN on this class queue. flowvalve enables RED/flow-valve (a.k.a. red-penalty-box) on this class queue. cleardscp clears diffserv codepoint in the IP header. HFSC Commands HFSC (Hierarchical Fair Service Curve) supports both link-sharing and guaran- teed real-time services. H-FSC employs a service curve based QoS model, and its unique feature is an ability to decouple delay and bandwidth allocation. HFSC has 2 independent scheduling mechanisms. Real-time scheduling is used to guarantee the delay and the bandwidth allocation at the same time. Hierarchi- cal link-sharing is used to distribute the excess bandwidth. When dequeueing a packet, HFSC always tries real-time scheduling first. If no packet is eli- gible for real-time scheduling, link-sharing scheduling is performed. HFSC does not use class hierarchy for real-time scheduling. interface if_name [bandwidth bps] [tbrsize bytes] [sched_type] if_name specifies the name of a network interface (e.g., fxp0). bandwidth specifies the interface bandwidth in bits per second. tbrsize specifies the bucket size of a token bucket regulator in bytes. sched_type must be hfsc for HFSC. class sched_type if_name class_name parent_name [admission cntlload|none] [[sc m1 d m2]] [[rt m1 d m2]] [[ls m1 d m2]] [pshare percent] [grate bps] [default] [qlimit count] [red|rio] [ecn] [cleardscp] The class command specifies a HFSC class. The classes are organized as a hierarchy, and every class, except for the root class, has a parent. Each HFSC class has 2 service curves, the real-time service curve and the link-sharing service curve. Service curves are specified by [type m1 d m2]. type should be either sc, rt or ls. sc (service curve) is used to set the same values to both service curves. rt (real-time) is used to specify the real-time service curve. ls (link-sharing) is used to specify the link-shar- ing service curve. m1 is the slope of the first segment specified in bits- per-second. d is the x-projection of the intersection point of the 2 segments specified in milliseconds. m2 is the slope of the second segment specified in bits-per-second. sched_type must be hfsc for a HFSC class. if_name Interface name. Must correspond to name in interface specifi- cation. class_name Arbitrary name for this class. Must be unique within the class hierarchy for this interface. The name root is a reserved class name for the root class. The root class for the interface is automatically created by the interface com- mand. parent_name The name of the parent class for this class. Keyword root is used when the parent is the root class. Parent class must have been previously defined. admission The type of admission control and QoS type. cntlload is con- trolled load service for RSVP, otherwise, it should be none. The default is none. pshare Percent of the link share. This specifies a linear link-shar- ing service curve as a fraction of the link bandwidth. It is a short hand of [ls 0 0 (link-bandwidth * percent / 100)]. grate Guaranteed rate. This specifies a linear real-time service curve. It is a short hand of [rt 0 0 bps]. default Specify the default class. When this keyword is present, all packets that do not match some classification criteria are assigned to this class. Must be exactly one class on each interface defined as the default class. qlimit The maximum queue size in number of packets. Default value is 50. red enables RED on this class queue. rio enables RIO on this class queue. ecn enables RED/ECN on this class queue. cleardscp clears diffserv codepoint in the IP header. PRIQ Commands PRIQ (Priority Queueing) implements a simple priority-based queueing. A higher priority class is always served first. Up to 16 priorities can be used with PRIQ. interface if_name [bandwidth bps] [tbrsize bytes] [sched_type] if_name specifies the name of a network interface (e.g., fxp0). bandwidth specifies the interface bandwidth in bits per second. tbrsize specifies the bucket size of a token bucket regulator in bytes. sched_type must be priq for PRIQ. class sched_type if_name class_name parent_name [priority pri] [default] [qlimit count] [red|rio] [ecn] [cleardscp] sched_type must be priq for a PRIQ class. if_name Interface name. Must correspond to name in interface specifi- cation. class_name Arbitrary name for this class. Must be unique for this inter- face. parent_name Parent class must be NULL for PRIQ. priority High numbers are higher priority. Max value is 15 and Min value is 0. Default is 0. A higher priority class is always served first in PRIQ. Priority must be unique for the inter- face. default Specify the default class. When this keyword is present, all packets that do not match some classification criteria are assigned to this class. Must be exactly one class on each interface defined as the default class. qlimit The maximum queue size in number of packets. Default value is 50. red enables RED on this class queue. rio enables RIO on this class queue. ecn enables RED/ECN on this class queue. cleardscp clears diffserv codepoint in the IP header. WFQ Commands WFQ (Weighted Fair Queueing) implements a weighted-round robin scheduler for a set of queue. A weight can be assigned to each queue to give a different pro- portion of the link capacity. A hash function is used to map a flow to one of a set of queues, and thus, it is possible for two different flows to be mapped into the same queue. interface if_name [bandwidth bps] [tbrsize bytes] [sched_type] [nqueues count] [qsize bytes] [hash policy] if_name specifies the name of a network interface (e.g., fxp0). bandwidth specifies the interface bandwidth in bits per second. tbrsize specifies the bucket size of a token bucket regulator in bytes. sched_type must be wfq for WFQ. nqueues The number of queues in WFQ. Default value is 256. qsize The size of each queue in number of bytes. Default value is 64K bytes. hash Type of hash policy to select a queue. dstaddr specifies a hashing policy by IP destination address. full specifies a hashing policy by IP addresses and ports. srcport specifies a hashing policy by IP source port number. Default is dstaddr FIFOQ Commands FIFOQ (First-In First-Out Queueing) is a simple tail-drop FIFO queue. FIFOQ is the simplest possible implementation of a queueing discipline in ALTQ, and can be used to compare with other queueing disciplines. FIFOQ can be also used as a template for those who want to write their own queueing disciplines. interface if_name [bandwidth bps] [tbrsize bytes] [sched_type] [qlimit count] if_name specifies the name of a network interface (e.g., fxp0). bandwidth specifies the interface bandwidth in bits per second. tbrsize specifies the bucket size of a token bucket regulator in bytes. sched_type must be fifoq for FIFOQ. qlimit The maximum queue size in number of packets. Default value is 50. RED Commands RED (Random Early Detection) is an implicit congestion notification mechanism that exercises packet dropping or packet marking stochastically according to the average queue length. RED can be viewed as a buffer management mechanism and can be integrated into other packet scheduling schemes. red min_th max_th inv_pmax The red command sets the default RED parameters. min_th and max_th are the minimum and the maximum threshold values. inv_pmax is the inverse (recipro- cal) of the maximum drop probability. For example, 10 means the maximum drop probability of 1/10. interface if_name [bandwidth bps] [tbrsize bytes] [sched_type] [qlimit count] [packetsize bytes] [weight n] [thmin n] [thmax n] [invpmax n] [ecn] [flowvalve] if_name specifies the name of a network interface (e.g., fxp0). bandwidth specifies the interface bandwidth in bits per second. tbrsize specifies the bucket size of a token bucket regulator in bytes. sched_type must be red for RED. qlimit The maximum queue size in number of packets. Default value is 60. packetsize The average packet size in number of bytes. This parameter is used to calibrate the idle period. Default value is 1000. weight The inverse of the weight of EWMA (exponentially weighted mov- ing average). thmin The minimum threshold. thmax The maximum threshold. invpmax The inverse of the maximum drop probability. ecn enables ECN. flowvalve enables flowvalve. RIO Commands ALTQ/RIO has 3 drop precedence levels defined for the Assured Forwarding of DiffServ (RFC2597). Since adaptive flows are likely to stay under the medium drop precedence level under congestion, the medium drop precedence would pro- tect adaptive flows from unadaptive flows. The original RIO has 2 sets of RED parameters; one for in-profile packets and the other for out-of-profile packets. At the ingress of the network, profile meters tag packets as IN or OUT based on contracted profiles for customers. Inside the network, IN packets receive preferential treatment by the RIO drop- per. It is possible to provision the network not to drop IN packets at all by providing enough capacity for the total volume of IN packets. Thus, RIO can be used to provide a service that statistically assures capacity allocated for users. This mechanism can be extended to support an arbitrary number of drop precedence levels. ALTQ supports 3 drop precedence levels. rio low_min_th low_max_th low_inv_pmax medium_min_th medium_max_th medium_inv_pmax high_min_th high_max_th high_inv_pmax The rio command sets the default RIO parameters. The parameters are RED parameters for 3 (low, medium, high) drop precedence. interface if_name [bandwidth bps] [tbrsize bytes] [sched_type] [qlimit count] [packetsize bytes] [weight n] [lo_thmin n] [lo_thmax n] [lo_invpmax n] [med_thmin n] [med_thmax n] [med_invpmax n] [hi_thmin n] [hi_thmax n] [hi_invpmax n] [ecn] if_name specifies the name of a network interface (e.g., fxp0). bandwidth specifies the interface bandwidth in bits per second. tbrsize specifies the bucket size of a token bucket regulator in bytes. sched_type must be rio for RIO. qlimit The maximum queue size in number of packets. Default value is 60. packetsize The average packet size in number of bytes. This parameter is used to calibrate the idle period. Default value is 1000. weight The inverse of the weight of EWMA (exponentially weighted mov- ing average). lo_thmin The minimum threshold for low drop precedence. lo_thmax The maximum threshold for low drop precedence. lo_invpmax The inverse of the maximum drop probability for low drop precedence. med_thmin The minimum threshold for medium drop precedence. med_thmax The maximum threshold for medium drop precedence. med_invpmax The inverse of the maximum drop probability for medium drop precedence. hi_thmin The minimum threshold for high drop precedence. hi_thmax The maximum threshold for high drop precedence. hi_invpmax The inverse of the maximum drop probability for high drop precedence. ecn enables ECN. BLUE Commands interface if_name [bandwidth bps] [tbrsize bytes] [sched_type] [qlimit count] [packetsize bytes] [maxpmark n] [holdtime usec] [ecn] if_name specifies the name of a network interface (e.g., fxp0). bandwidth specifies the interface bandwidth in bits per second. tbrsize specifies the bucket size of a token bucket regulator in bytes. sched_type must be blue for BLUE. qlimit The maximum queue size in number of packets. Default value is 60. packetsize The average packet size in number of bytes. Default value is 1000. maxpmark specifies the precision of marking probability. holdtime specifies the hold time in usec. ecn enables ECN. CDNR Commands The conditioner command specifies a diffserv traffic conditioner. A traffic conditioner is not a queueing discipline but a component to meter, mark or drop incoming packets according to some rules. As opposed to a queueing discipline, a traffic conditioner handles incoming packets at an input interface. If no queueing discipline (e.g., CBQ) is used for the interface, a null interface command should be used to specify an input network interface. interface if_name [bandwidth bps] [tbrsize bytes] The conditioner command has the following syntax. conditioner if_name cdnr_name <action> if_name Interface name. Must correspond to name in interface speci- fication. cdnr_name Arbitrary name for this conditioner. Must be unique for this interface. action Action of the conditioner. An action can be a recursively defined action. The following actions are defined. pass pass allows the packet to go through without any modifica- tion to the packet. drop drop rejects the packet. The packet is immediately dis- carded. mark value mark sets the specified value to the ds field in the IP header. Then, the packet is allowed to go through. tbmeter rate depth <in_action> <out_action> tbmeter is a token bucket meter configured with rate and depth parameters. Rate is token rate in bits-per-second. Depth is bucket depth in KB. When an incoming packet is in profile (available token is more than the packet size), tbmeter takes in_action. Otherwise, tbmeter takes out_action. trtcm cmtd_rate cmtd_depth peak_rate peak_depth <green_action> <yellow_action> <red_action> [coloraware|colorblind] trtcm is a 2-rate 3 color marker for Assured Forwarding. A trtcm consists of 2 token buckets, one for a committed rate and the other for a peak rate. When an incoming packet is in the committed profile, trtcm takes green_action. When the packet is out of the committed profile but in the peak profile, trtcm takes yellow_action. Otherwise, tbtcm takes red_action. A trtcm is either color-aware or color-blind. A color-aware trtcm do not raise the color (ds field value), that is, a yellow packet can be yellow or red but can not be blue. Default is color-blind. tswtcm cmtd_rate peak_rate avg_interval <green_action> <yellow_action> <red_action> tswtcm is a time sliding window 3 color marker for Assured Forwarding. A tswtcm differs from trtcm in that a tswtcm probabilistically marks packets. A tswtcm consists of 2 rates, one for a committed rate and the other for a peak rate. When an incoming packet is in the committed pro- file, tswtcm takes green_action. When the packet is out of the committed profile but in the peak profile, tswtcm takes yellow_action. Otherwise, tswtcm takes red_action. cmtd_rate and peak_rate are specified in bits per second. avg_interval provides the size of time window for averag- ing incoming rate, and is specified in milliseconds. 500 msec is ok for normal settings.
EXAMPLES
CBQ Example # # cbq configuration for vx0 (10Mbps ether) # give at least 40% to TCP # limit HTTP from network 133.138.1.0 up to 10%, use RED. # other traffic goes into default class # interface vx0 bandwidth 10M cbq # class cbq vx0 root_class NULL priority 0 pbandwidth 100 class cbq vx0 def_class root_class borrow pbandwidth 95 default class cbq vx0 tcp_class def_class borrow pbandwidth 40 filter vx0 tcp_class 0 0 0 0 6 class cbq vx0 csl_class tcp_class pbandwidth 10 red filter vx0 csl_class 0 0 133.138.1.0 netmask 0xffffff00 80 6 filter vx0 csl_class 133.138.1.0 netmask 0xffffff00 0 0 80 6 # # sample filter6 command # filter6 vx0 csl_class ::0 0 d000:a:0:123::/64 80 6 HFSC Example # # hfsc configuration for hierarchical sharing # interface pvc0 bandwidth 45M hfsc # # (10% of the bandwidth share goes to the default class) class hfsc pvc0 def_class root pshare 10 default # # bandwidth share guaranteed rate # CMU: 45% 15Mbps # PITT: 45% 15Mbps # class hfsc pvc0 cmu root pshare 45 grate 15M class hfsc pvc0 pitt root pshare 45 grate 15M # # CMU bandwidth share guaranteed rate # CS: 20% 10Mbps # other: 20% 5Mbps # class hfsc pvc0 cmu_other cmu pshare 20 grate 10M filter pvc0 cmu_other 0 0 128.2.0.0 netmask 0xffff0000 0 0 class hfsc pvc0 cmu_cs cmu pshare 20 grate 5M filter pvc0 cmu_cs 0 0 128.2.242.0 netmask 0xffffff00 0 0 # # PITT bandwidth share guaranteed rate # CS: 20% 10Mbps # other: 20% 5Mbps # class hfsc pvc0 pitt_other pitt pshare 20 grate 10M filter pvc0 pitt_other 0 0 136.142.0.0 netmask 0xffff0000 0 0 class hfsc pvc0 pitt_cs pitt pshare 20 grate 5M filter pvc0 pitt_cs 0 0 136.142.79.0 netmask 0xffffff00 0 0 PRIQ Example # # priq configuration for fxp0 (100Mbps ether) # icmp: high priority # tcp: medium priority # others: low priority # interface fxp0 bandwidth 100M priq # class priq fxp0 high_class NULL priority 2 filter fxp0 high_class 0 0 0 0 1 class priq fxp0 med_class NULL priority 1 filter fxp0 high_class 0 0 0 0 6 class priq fxp0 low_class NULL priority 0 default WFQ Example interface pvc0 bandwidth 134000000 wfq FIFOQ Example interface rl0 bandwidth 10M fifoq Conditioner Example # interface fxp0 # # a simple dropper # discard all packets from 192.168.0.83 # conditioner fxp0 dropper <drop> filter fxp0 dropper 0 0 192.168.0.83 0 0 # # EF conditioner # mark EF to all packets from 192.168.0.117 # conditioner pvc1 ef_cdnr <tbmeter 6M 64K <mark 0xb8><drop>> filter fxp0 ef_cdnr 0 0 192.168.0.117 0 0 # # AF1x conditioner # mark AF1x to packets from 192.168.0.178 # AF11 (low drop precedence): less than 3Mbps # AF12 (medium drop precedence): more than 3Mbps and less than 10Mbps # AF13 (high drop precedence): more than 10Mbps # conditioner fxp0 af1x_cdnr <trtcm 3M 32K 10M 64K <mark 0x28><mark 0x30><mark 0x38>> filter fxp0 af1x_cdnr 0 0 192.168.0.178 0 0
SEE ALSO
altqd(8)
BUGS
This man page is incomplete. For more information read the source. NetBSD 3.1.1 September 28, 1999 NetBSD 3.1.1
Powered by man-cgi (2024-08-26). Maintained for NetBSD by Kimmo Suominen. Based on man-cgi by Panagiotis Christias.