Dynamic assignment of traffic classes to a priority queue in a packet forwarding device

US 20040076161A1
(19) United States
(12) Patent Application Publication (10) Pub. No.: US 2004/0076161 A1
Lavian et al. (43) Pub. Date: Apr. 22, 2004
(54) DYNAMIC ASSIGNMENT OF TRAFFIC
CLASSES TO A PRIORITY QUEUE IN A
PACKET FORWARDING DEVICE
(76) Inventors: Tal I. Lavian, Sunnyvale, CA (US);
Stephen Lau, Milpitas, CA (US)
Correspondence Address:
BLAKELY SOKOLOFF TAYLOR & ZAFMAN
12400 WILSHIRE BOULEVARD, SEVENTH
FLOOR
LOS ANGELES, CA 90025 (US)
(21) Appl. No.: 10/685,947
(22) Filed: Oct. 15, 2003
Related US. Application Data
(63) Continuation of application No. 09/227,389, ?led on
Jan. 8, 1999.
Publication Classi?cation
5 1 I nt. C] . 7 ..................................................... .. H04L 1/00
(52) US. Cl. ............. .. 370/395.41; 370/395.21; 370/230
(57) ABSTRACT
An apparatus and method for dynamic assignment of classes
of traf?c to a priority queue. Bandwidth consumption by one
or more types of packet traffic received in the packet
forwarding device is monitored to determine whether the
bandwidth consumption exceeds a threshold. If the band
width consumption exceeds the threshold, assignment of at
least one type of packet traffic of the one or more types of
packet traf?c is changed from a queue having a ?rst priority
to a queue having a second priority.

Patent Application Publication Apr. 22, 2004 Sheet 2 0f 6 US 2004/0076161 A1
QUEVE "I LL
Locum. (>21 men can»
9mm 5 “aux FAQQKL
S
Win‘ LEL; HLAOEQ
Kw Guava EH'ULK
i$SOLS)-RE' Asaouxs,
MLQK
I
I Wok“ (2:04))‘
Ylcl as
- ‘I F -
pmmm pkcmF-T
GM EUG- BEST swam -
. Gwwa V16. '2- B i

I P koaracn
’ ‘ (Jo-'2" iTomE. Enummwc, IP ‘

UZOEuQMDUEZOU

US 2004/0076161 A1
DYNAMIC ASSIGNMENT OF TRAFFIC CLASSES
TO A PRIORITY QUEUE IN A PACKET
FORWARDING DEVICE
FIELD OF THE INVENTION
[0001] The present invention relates to the ?eld of tele
communications, and more particularly to dynamic assign
ment of traf?c classes to queues having different priority
levels.
BACKGROUND OF THE INVENTION
[0002] The How of packets through packet-switched net
Works is controlled by sWitches and routers that forWard
packets based on destination information included in the
packets themselves. A typical sWitch or router includes a
number of input/output (I/ O) modules connected to a sWitch
ing fabric, such as a crossbar or shared memory sWitch. In
some sWitches and routers, the sWitching fabric is operated
at a higher frequency than the transmission frequency of the
I/O modules so that the sWitching fabric may deliver packets
to an I/O module faster than the I/O module can output them
to the netWork transmission medium. In these devices,
packets are usually queued in the I/O module to aWait
transmission.
[0003] One problem that may occur When packets are
queued in the I/O module or elseWhere in a sWitch or router
is that the queuing delay per packet varies depending on the
amount of traf?c being handled by the sWitch. Variable
queuing delays tend to degrade data streams produced by
real-time sampling (e.g., audio and video) because the
original time delays betWeen successive packets in the
stream convey the sampling interval and are therefore
needed to faithfully reproduce the source information.
Another problem that results from queuing packets in a
sWitch or router is that data from a relatively important
source, such as a shared server, may be impeded by data
from less important sources, resulting in bottlenecks.
SUMMARY OF THE INVENTION
[0004] Amethod and apparatus for dynamic assignment of
classes of traffic to a priority queue are disclosed. BandWidth
consumption by one or more types of packet traf?c received
in a packet forWarding device is monitored. The queue
assignment of at least one type of packet traf?c is automati
cally changed from a queue having a ?rst priority to a queue
having a second priority if the bandWidth consumption
eXceeds the threshold.
[0005] Other features and advantages of the invention Will
be apparent from the accompanying draWings and from the
detailed description that folloWs beloW.
BRIEF DESCRIPTION OF THE DRAWINGS
[0006] The present invention is illustrated by Way of
eXample and not limitation in the ?gures of the accompa
nying draWings in Which like references indicate similar
elements and in Which:
[0007] FIG. 1 illustrates a packet forWarding device that
can be used to implement embodiments of the present
invention;
[0008] FIG. 2A illustrates queue ?ll logic implemented by
a queue manager in a quad interface device;
Apr. 22, 2004
[0009] FIG. 2B illustrates queue drain logic according to
one embodiment;
[0010] FIG. 3 illustrates the How of a packet Within the
sWitch of FIG. 1;
[0011] FIG. 4 illustrates storage of an entry in an address
resolution table managed by an address resolution unit;
[0012] FIG. 5 is a diagram of the softWare architecture of
the sWitch of FIG. 1 according to one embodiment; and
[0013] FIG. 6 illustrates an eXample of dynamic assign
ment of traffic classes to a priority queue.
DETAILED DESCRIPTION
[0014] A packet forWarding device in Which selected
classes of netWork traffic may be dynamically assigned for
priority queuing is disclosed. In one embodiment, the packet
forWarding device includes a Java virtual machine for
executing user-coded Java applets received from a netWork
management server (NMS). AJava-to-native interface (JNI)
is provided to alloW the Java applets to obtain error infor
mation and traf?c statistics from the device hardWare and to
alloW the Java applets to Write con?guration information to
the device hardWare, including information that indicates
Which classes of traf?c should be queued in priority queues.
The Java applets implement user-speci?ed traf?c manage
ment policies based on real-time evaluation of the error
information and traf?c statistics to provide dynamic control
of the priority queuing assignments. These and other aspects
and advantages of the present invention are described beloW.
[0015] FIG. 1 illustrates a packet forWarding device 17
that can be used to implement embodiments of the present
invention. For the purposes of the present description, the
packet forWarding device 17 is assumed to be a sWitch that
sWitches packets betWeen ingress and egress ports based on
media access control (MAC) addresses Within the packets.
In an alternate embodiment, the packet forWarding device 17
may be a router that routes packets according to destination
internet protocol (IP) addresses or a routing sWitch that
performs both MAC address sWitching and IP address
routing. The techniques and structures disclosed herein are
applicable generally to a device that forWards packets in a
packet sWitching netWork. Also, the term packet is used
broadly herein to refer to a ?xed-length cell, a variable
length frame or any other information structure that is
self-contained as to its destination address.
[0016] The sWitch 17 includes a sWitching fabric 12
coupled to a plurality of I/O units (only I/O units 1 and 16
are depicted) and to a processing unit 10. The processing
unit includes at least a processor 31 (Which may be a
microprocessor, digital signal processor or microcontroller)
coupled to a memory 32 via a bus 33. In one embodiment,
each I/O unit 1, 16 includes four physical ports P1-P4
coupled to a quad media access controller (QMAC) 14A,
14B via respective transceiver interface units 21A-24A,
21B-24B. Each I/O unit 1, 16 also includes a quad interface
device (QID) 16A, 16B, an address resolution unit (ARU)
15A, 15B and a memory 18A, 18B, interconnected as shoWn
in FIG. 1. Preferably, the sWitch 17 is modular With at least
the I/O units 1, 16 being implemented on port cards (not
shoWn) that can be installed in a backplane (not shoWn) of
the sWitch 17. In one implementation, each port card
includes four I/O units and therefore supports up to 16

US 2004/0076161 A1
physical ports. The switch backplane includes slots for up to
six port cards, so that the sWitch 17 can be scaled according
to customer needs to support betWeen 16 and 96 physical
ports. In alternate embodiments, each I/O unit 1, 16 may
support more or feWer physical ports, each port card may
support more or feWer I/O units 1, 16 and the sWitch 17 may
support more or feWer port cards. For example, the I/O unit
1 shoWn in FIG. 1 may be used to support four 10baseT
transmission lines (i.e., 10 Mbps (mega-bit per second),
tWisted-pair) or four 100baseF transmission lines (100
Mbps, ?ber optic), While a different I/O unit (not shoWn)
may be used to support a single 1000baseF transmission line
(1000 Mbps, ?ber optic). Nothing disclosed herein should be
construed as limiting embodiments of the present invention
to use With a particular transmission medium, I/O unit, port
card or chassis con?guration.
[0017] Still referring to FIG. 1, When a packet 25 is
received on physical port P1, it is supplied to the corre
sponding physical transceiver 21A Which performs any
necessary signal conditioning (e.g., optical to electrical
signal conversion) and then forWards the packet 25 to the
QMAC 14A. The QMAC 14A buffers packets received from
the four physical transceivers 21A-24A as necessary, for
Warding one packet at a time to the QID 16A. Receive logic
Within the QID 16A noti?es the ARU 15A that the packet 25
has been received. The ARU computes a table index based
on the destination MAC address Within the packet 25 and
uses the index to identify an entry in a forWarding table that
corresponds to the destination MAC address. In packet
forwarding devices that operate on different protocol layers
of the packet (e.g., routers), a forWarding table may be
indexed based on other destination information contained
Within the packet.
[0018] According to one embodiment, the forWarding
table entry identi?ed based on the destination MAC address
indicates the sWitch egress port to Which the packet 25 is
destined and also Whether the packet is part of a MAC
address based virtual local area netWork (VLAN), or a
port-based VLAN. (As an aside, a VLAN is a logical
grouping of MAC addresses (a MAC-address-based VLAN)
or a logical grouping of physical ports (a port-based
VLAN).) The forWarding table entry further indicates
Whether the packet 25 is to be queued in a priority queue in
the I/O unit that contains the destination port. As discussed
beloW, priority queuing may be speci?ed based on a number
of conditions, including, but not limited to, Whether the
packet is part of a particular IP ?oW, or Whether the packet
is destined for a particular port, VLAN or MAC address.
[0019] According to one embodiment, the QID 16A, 16B
segments the packet 25 into a plurality of ?xed-length cells
26 for transmission through the sWitching fabric 12. Each
cell includes a header 28 that identi?es it as a constituent of
the packet 25 and that identi?es the destination port for the
cell (and therefore for the packet 25). The header 28 of each
cell also includes a bit 29 indicating Whether the cell is the
beginning cell of a packet and also a bit 30 indicating
Whether the packet 25 to Which the cell belongs is to be
queued in a priority queue or a best effort queue on the
destined I/O unit.
[0020] The sWitching fabric 12 forWards each cell to the
I/O unit indicated by the cell header 28. In the exemplary
data How shoWn in FIG. 1, the constituent cells 26 of the
Apr. 22, 2004
packet 25 are assumed to be forWarded to I/O unit 16 Where
they are delivered to transmit logic Within the QID 16B. The
transmit logic in the QID 16B includes a queue manager (not
shoWn) that maintains a priority queue and a best effort
queue in the memory 18B. In one embodiment, the memory
18B is resolved into a pool of buffers, each large enough to
hold a complete packet. When the beginning cell of the
packet 25 is delivered to the QID 16B, the queue manager
obtains a buffer from the pool and appends the buffer to
either the priority queue or the best effort queue according
to Whether the priority bit 30 is set in the beginning cell. In
one embodiment, the priority queue and the best effort queue
are each implemented by a linked list, With the queue
manager maintaining respective pointers to the head and tail
of each linked list. Entries are added to the tail of the queue
list by advancing the tail pointer to point to a neWly allocated
buffer that has been appended to the linked list, and entries
are popped off the head of the queue by advancing the head
pointer to point to the next buffer in the linked list and
returning the spent buffer to the pool.
[0021] After a buffer is appended to either the priority
queue or the best effort queue, the beginning cell and
subsequent cells are used to reassemble the packet 25 Within
the buffer. Eventually the packet 25 is popped off the head
of the queue and delivered to an egress port via the QMAC
14B and the physical transceiver (e.g., 23B) in an egress
operation. This is shoWn by Way of example in FIG. 1 by the
egress of packet 25 from physical port P3 of I/O unit 16.
[0022] FIG. 2A illustrates queue ?ll logic implemented by
the queue manager in the QID. Starting at block 51, a cell is
received in the QID from the sWitching fabric. The begin
ning cell bit in the cell header is inspected at decision block
53 to determine if the cell is the beginning cell of a packet.
If so, the priority bit in the cell header is inspected at
decision block 55 to determine Whether to allocate an entry
in the priority queue or the best effort queue for packet
reassembly. If the priority bit is set, an entry in the priority
queue is allocated at block 57 and the priority queue entry
is associated With the portion of the cell header that identi?es
the cell as a constituent of a particular packet at block 59. If
the priority bit in the cell header is not set, then an entry in
the best effort queue is allocated at block 61 and the best
effort queue entry is associated With the portion of the cell
header that identi?es the cell as a constituent of a particular
packet at block 63.
[0023] Returning to decision block 53, if the beginning
cell bit in the cell header is not set, then the queue entry
associated With the cell header is identi?ed at block 65. The
association betWeen the cell header and the queue entry
identi?ed at block 65 Was established earlier in either block
59 or block 63. Also, identi?cation of the queue entry in
block 65 may include inspection of the priority bit in the cell
to narroW the identi?cation effort to either the priority queue
or the best effort queue. In block 67, the cell is combined
With the preceding cell in the queue entry in a packet
reassembly operation. If the reassembly operation in block
67 results in a completed packet (decision block 69), then
the packet is marked as ready for transmission in block 71.
In one embodiment, the packet is marked by setting a ?ag
associated With the queue entry in Which the packet has been
reassembled. Other techniques for indicating that a packet is
ready for transmission may be used in alternate embodi
ments.

US 2004/0076161 A1
[0024] FIG. 2B illustrates queue drain logic according to
one embodiment. At decision block 75, the entry at the head
of the priority queue is inspected to determine if it contains
a packet ready for transmission. If so, the packet is trans
mitted at block 77 and the corresponding priority queue
entry is popped off the head of the priority queue and
deallocated at block 79. If a ready packet is not present at the
head of the priority queue, then the entry at the head of the
best effort queue is inspected at decision block 81. If a
packet is ready at the head of the best effort queue, it is
transmitted at block 83 and the corresponding best effort
queue entry is popped off the head of the best effort queue
and deallocated in block 85. Note that, in the embodiment
illustrated in FIG. 2B, packets are drained from the best
effort queue only after the priority queue has been emptied.
In alternate embodiments, a timer, counter or similar logic
element may be used to ensure that the best effort queue 105
is serviced at least every so often or at least after every N
number of packets are transmitted from the priority queue,
thereby ensuring at least a threshold level of service to best
effort queue.
[0025] FIG. 3 illustrates the How of a packet Within the
sWitch 17 of FIG. 1. A packet is received in the sWitch at
block 91 and used to identify an entry in a forWarding table
called the address resolution table at block 93. At
decision block 95, a priority bit in the AR table entry is
inspected to determine Whether the packet belongs to a class
of traffic that has been selected for priority queuing. If the
priority bit is set, the packet is segmented into cells having
respective priority bits set in their headers in block 97. If the
priority bit is not set, the packet is segmented into cells
having respective priority bits cleared their cell headers in
block 99. The constituent cells of each packet are forWarded
to an egress I/O unit by the sWitching fabric. In the egress
I/O unit, the priority bit of each cell is inspected (decision
block 101) and used to direct the cell to an entry in either the
priority queue 103 or the best effort queue 105 Where it is
combined With other cells to reassemble the packet.
[0026] FIG. 4 illustrates storage of an entry in the address
resolution table managed by the ARU. In one embodi
ment, the AR table is maintained in a high speed static
random access memory (SRAM) coupled to the ARU.
Alternatively, the AR table may be included in a memory
Within an application-speci?c integrated circuit (ASIC) that
includes the ARU. Generally, the ARU stores an entry in the
AR table in response to packet forWarding information from
the processing unit. The processing unit supplies packet
forWarding information to be stored in each AR table in the
sWitch Whenever a neW association betWeen a destination
address and a sWitch egress port is learned. In one embodi
ment, an address-to-port association is learned by transmit
ting a packet that has an unknoWn egress port assignment on
each of the egress ports of the sWitch and associating the
destination address of the packet With the egress port at
Which an acknowledgment is received. Upon learning the
association betWeen the egress port and the destination
address, the processing unit issues forWarding information
that includes, for eXample, an identi?er of the neWly asso
ciated egress port, the destination MAC address, an identi
?er of the VLAN associated With the MAC address (if any),
an identi?er of the VLAN associated With the egress port (if
any), the destination IP address, the destination IP port (e.g.,
transmission control protocol (TCP), universal device pro
tocol (UDP) or other IP port) and the IP protocol (e.g.,
Apr. 22, 2004
HTTP, FTP or other IP protocol). The source IP address,
source IP port and source IP protocol may also be supplied
to fully identify an end-to-end IP ?oW.
[0027] Referring to FIG. 4, forWarding information 110 is
received from the processing unit at block 115. At block 117,
the ARU stores the forWarding information in an AR table
entry. At decision block 119, the physical egress port iden
ti?er stored in the AR table entry is compared against
priority con?guration information to determine if packets
destined for the egress port have been selected for priority
egress queuing. If so, the priority bit is set in the AR table
entry in block 127. Thereafter, incoming packets that indeX
the neWly stored table entry Will be queued in the priority
queue to aWait transmission. If packets destined for the
egress port have not been selected for priority queuing, then
at decision block 121 the MAC address stored in the AR
table entry is compared against the priority con?guration
information to determine if packets destined for the MAC
address have been selected for priority egress queuing. If so,
the priority bit is set in the AR table entry in block 127. If
packets destined for the MAC address have not been
selected for priority egress queuing, then at decision block
123 the VLAN identi?er stored in the AR table entry (if
present) is compared against the priority con?guration infor
mation to determine if packets destined for the VLAN have
been selected for priority egress queuing. If so, the priority
bit is set in the AR table entry in block 127. If packets
destined for the VLAN have not been selected for priority
egress queuing, then at block 125 the IP ?oW identi?ed by
the IP address, IP port and IP protocol in the AR table is
compared against the priority con?guration information to
determine if packets that form part of the IP How have been
selected for priority egress queuing. If so, the priority bit is
set in the AR table entry, otherWise the priority bit is not set.
Yet other criteria may be considered in assigning priority
queuing in alternate embodiments. For example, priority
queuing may be speci?ed for a particular IP protocol (e.g.,
FTP, HTTP). Also, the ingress port, source MAC address or
source VLAN of a packet may also be used to determine
Whether to queue the packet in the priority egress packet.
More speci?cally, in one embodiment, priority or best effort
queuing of unicast traf?c is determined based on destination
parameters (e.g., egress port, destination MAC address or
destination IP address), While priority or best effort queuing
of multicast traf?c is determined based on source parameters
(e.g., ingress port, source MAC address or source IP
address).
[0028] FIG. 5 is a diagram of the softWare architecture of
the sWitch 17 of FIG. 1 according to one embodiment. An
operating system 143 and device drivers 145 are provided to
interface With the device hardWare 141. For eXample, device
drivers are provided to Write con?guration information and
AR storage entries to the ARUs in respective I/ O units. Also,
the operating system 143 performs memory management
functions and other system services in response to requests
from higher level softWare. Generally, the device drivers 145
eXtend the services provided by the operating system and are
invoked in response to requests for operating system service
that involve device-speci?c operations.
[0029] The device management code 147 is eXecuted by
the processing unit (e.g., element 10 of FIG. 1) to perform
system level functions, including management of forWard
ing entries in the distributed AR tables and management of

US 2004/0076161 A1
forwarding entries in a master forwarding table maintained
in the memory of the processing unit. The device manage
ment code 147 also includes routines for invoking device
driver services, for example, to query the ARU for traf?c
statistics and error information, or to Write updated con?gu
ration information to the ARUs, including priority queuing
information. Further, the device management code 147
includes routines for Writing updated con?guration informa
tion to the ARUs, as discussed beloW in reference to FIG. 6.
In one implementation, the device management code 147 is
native code, meaning that the device management code 147
is a compiled set of instructions that can be executed directly
by a processor in the processing unit to carry out the device
management functions.
[0030] In one embodiment, the device management code
147 supports the operation of a Java client 160 that includes
a number of Java applets, including a monitor applet 157, a
policy enforcement applet 159 and a con?guration applet
161. A Java applet is an instantiation of a Java class that
includes one or more methods for self initialiZation (e.g., a
constructor method called “Applet( )”), and one or more
methods for communicating With a controlling application.
Typically the controlling application for a Java applet is a
Web broWser executed on a general purpose computer. In the
softWare architecture shoWn in FIG. 5, hoWever, a Java
application called Data Communication Interface (DCI) 153
is the controlling application for the monitor, policy enforce
ment and con?guration applets 157, 159, 161. The DCI
application 153 is executed by a Java virtual machine 149 to
manage the doWnload of Java applets from a netWork
management server (NMS) 170. A library of Java objects
155 is provided for use by the Java applets 157, 159, 161 and
the DCI application 153.
[0031] In one implementation, the NMS 170 supplies Java
applets to the sWitch 17 in a hyper-text transfer protocol
(HTTP) data stream. Other protocols may also be used. The
constituent packets of the HTTP data stream are addressed
to the IP address of the sWitch and are directed to the
processing unit after being received by the I/O unit coupled
to the NMS 170. After authenticating the HTIT data stream,
the DCI application 153 stores the Java applets provided in
the data stream in the memory of the processing unit and
executes a method to invoke each applet. An applet is
invoked by supplying the Java virtual machine 149 With the
address of the constructor method of the applet and causing
the Java virtual machine 149 to begin execution of the applet
code. Program code de?ning the Java virtual machine 149 is
executed to interpret the platform independent byte codes of
the Java applets 157, 159, 161 into native instructions that
can be executed by a processor Within the processing unit.
[0032] According to one embodiment, the monitor applet
157, policy enforcement applet 159 and con?guration applet
161 communicate With the device management code 147
through a Java-native interface (JNI) 151. The JNI 151 is
essentially an application programming interface (API) and
provides a set of methods that can be invoked by the Java
applets 157, 159, 161 to send messages and receive
responses from the device management code 147. In one
implementation, the JNI 151 includes methods by Which the
monitor applet 157 can request the device management code
147 to gather error information and traffic statistics from the
device hardWare 141. The JNI 151 also includes methods by
Which the con?guration applet 161 can request the device
Apr. 22, 2004
management code 147 to Write con?guration information to
the device hardWare 141. More speci?cally, the JNI 151
includes a method by Which the con?guration applet 161 can
indicate that priority queuing should be performed for
speci?ed classes of traf?c, including, but not limited to, the
classes of traf?c discussed above in reference to FIG. 4. In
this Way, a user-coded con?guration applet 161 may be
executed by the Java virtual machine 149 Within the sWitch
17 to invoke a method in the JNI 151 to request the device
management code 147 to Write information that assigns
selected classes of traf?c to be queued in the priority egress
queue. In effect, the con?guration applet 161 assigns virtual
queues de?ned by the selected classes of traf?c to feed into
the priority egress queue.
[0033] Although a Java virtual machine 149 and Java
applets 157, 159, 161 have been described, other virtual
machines, interpreters and scripting languages may be used
in alternate embodiments. Also, as discussed beloW, more or
feWer Java applets may be used to perform the monitoring,
policy enforcement and con?guration functions in alternate
embodiments.
[0034] FIG. 6 illustrates an example of dynamic assign
ment traf?c classes to a priority queue. An exemplary
netWork includes sWitches A and B coupled together at
physical ports 32 and 1, respectively. Suppose that a netWork
administrator or other user determines that an important
server 175 on port 2 of sWitch A requires a relatively high
quality of service (QoS), and that, at least in sWitch B, the
required QoS can be provided by ensuring that at least 20%
of the egress capacity of sWitch B, port 1 is reserved for
traffic destined to the MAC address of the server 175. One
Way to ensure that 20% egress capacity is reserved to traf?c
destined for the server 175 is to assign priority queuing for
packets destined to the MAC address of the server 175, but
not for other traffic. While such an assignment Would ensure
priority egress to the server traf?c, it also may result in
unnecessarily high bandWidth allocation to the server 175,
potentially starving other important traf?c or causing other
important traf?c to become bottlenecked behind less impor
tant traf?c in the best effort queue. For example, suppose that
there are at least tWo other MAC address destinations, MAC
address A and MAC address B, to Which the user desires to
assign priority queuing, so long as the egress capacity
required by the server-destined traf?c is available. In that
case, it Would be desirable to dynamically con?gure the
MAC address A and MAC address B traf?c to be queued in
either the priority queue or the best effort queue according
to existing traf?c conditions. In at least one embodiment, this
is accomplished using monitor, policy enforcement and
con?guration applets that have been doWnloaded to sWitch
B and Which are executed in a Java client in sWitch B as
described above in reference to FIG. 5.
[0035] FIG. 6 includes exemplary pseudocode listings of
monitor, policy enforcement and con?guration applets 178,
179, 180 that can be used to ensure that at least 20% of the
egress capacity of sWitch B, port 1 is reserved for traf?c
destined to the server 175, but Without unnecessarily deny
ing priority queuing assignment to traffic destined for MAC
addresses A and B. After initialiZation, the monitor applet
178 repeatedly measures of the port 1 line utiliZation from
the device hardWare. In one embodiment, the ARU in the I/O
unit that manages port 1 keeps a count of the number of
packets destined for particular egress ports, packets destined

US 2004/0076161 A1
for particular MAC addresses, packets destined for particu
lar VLANS, packets that form part of a particular IP ?oW,
packets having a particular IP protocol, and so forth. The
ARU also tracks the number of errors associated With these
different classes of traf?c, the number of packets from each
class of traf?c that are dropped, and other statistics. By
determining the change in these different statistics per unit
time, a utiliZation factor may be generated that represents the
percent utiliZation of the capacity of an egress port, an I/O
unit or the overall sWitch. Error rates and packet drop rates
may also be generated.
[0036] In one embodiment, the monitor applet 178 mea
sures line utiliZation by invoking methods in the JNI to read
the port 1 line utiliZation resulting from traf?c destined for
MAC address A and for MAC address B every 10 millisec
onds.
[0037] The policy enforcement applet 179 includes vari
ables to hold the line utiliZation percentage of traf?c destined
for MAC address A (A%), the line utiliZation percentage of
traf?c destined for MAC address B (B%), the queue assign
ment (i.e., priority or best effort) of traf?c destined for the
server MAC address (QA_S), the queue assignment of traf?c
destined for MAC address A (QA_A) and the queue assign
ment of traf?c destined for MAC address B. Also, a constant,
DELTA, is de?ned to be 5% and the queue assignments for
the MAC address A, MAC address B and server MAC
address traf?c are initially set to the priority queue.
[0038] The policy enforcement applet 179 also includes a
forever loop in Which the line utiliZation percentages A%
and B% are obtained from the monitor applet 178 and used
to determine Whether to change the queue assignments
QA_A and QA_B. If the MAC address A traf?c and the
MAC address B traf?c are both assigned to the priority
queue (the initial con?guration) and the sum of the line
utiliZation percentages A% and B% eXceeds 80%, then less
than 20% line utiliZation remains for the server-destined
traf?c. In that event, the MAC address A traffic is reassigned
from the priority queue to the best effort queue (code
statement 181). If the MAC address A traffic is assigned to
the best effort queue and the MAC address B traf?c is
assigned to the priority queue, then the MAC address A
traf?c is reassigned to the priority queue if the sum of the
line utiliZation percentages A% and B% drops beloW 80%
less DELTA (code statement 183). The DELTA parameter
provides a deadband to prevent rapid changing of priority
queue assignment.
[0039] If the MAC address Atraf?c is assigned to the best
effort queue and the MAC address B traffic is assigned to the
priority queue and the line utiliZation percentage B%
eXceeds 80%, then less than 20% line utiliZation remains for
the server-destined traffic. Consequently, the MAC address
B traffic is reassigned from the priority queue to the best
effort queue (code statement 185). If the MAC address B
traf?c is assigned to the best effort queue and the line
utiliZation percentage B% drops beloW 80% less DELTA,
then the MAC address B traf?c is reassigned to the priority
queue (code statement 187). Although not speci?cally pro
vided for in the exemplary pseudocode listing of FIG. 6, the
policy enforcement applet 179 may treat the traf?c destined
for the MAC A and MAC B addresses more symmetrically
by including additional statements to conditionally assign
traf?c destined for MAC address A to the priority queue, but
Apr. 22, 2004
not traf?c destined for MAC address B. In the eXemplary
pseudocode listing of FIG. 6, the policy enforcement applet
179 delays for 5 milliseconds at the end of each pass through
the forever loop before repeating.
[0040] The con?guration applet 180 includes variables,
QA_A and QA_B, to hold the queue assignments of the
traffic destined for the MAC addressesA and B, respectively.
Variables LAST_QA_A and LAST_QA_B are also provided
to record the history (i.e., most recent values) of the QA_A
and QA_B values. The LAST_QA_A and LAST_QA_B
variables are initialiZed to indicate that traffic destined for
the MAC addressesAand B is assigned to the priority queue.
[0041] Like the monitor and policy enforcement applets
178,179, the con?guration applet 180 includes a forever
loop in Which a code sequence is eXecuted folloWed by a
delay. In the eXemplary listing of FIG. 6, the ?rst operation
performed by the con?guration applet 180 Within the forever
loop is to obtain the queue assignments QA_A and QA_B
from the policy enforcement applet 179. If the queue assign
ment indicated by QAA is different from the queue assign
ment indicated by LAST_QA_A, then a JNI method is
invoked to request the device code to recon?gure the queue
assignment of the traf?c destined for MAC address A
according to the neW QA_A value. The neW QA_A value is
then copied into the LAST_QA_A variable so that subse
quent queue assignment changes are detected. If the queue
assignment indicated by QA_B is different from the queue
assignment indicated by LAST_QA_B, then a J NI method is
invoked to request the device code to recon?gure the queue
assignment of the traf?c destined for MAC address B
according to the neW QA_B value. The neW QA_B value is
then copied into the LAST_QA_B variable so that subse
quent queue assignment changes are detected. By this opera
tion, and the operation of the monitor and policy enforce
ment applets 178, 179, traffic destined for the MAC
addresses A and B is dynamically assigned to the priority
queue according to real-time evaluations of the traffic con
ditions in the sWitch.
[0042] Although a three-applet implementation is illus
trated in FIG. 6, more or feWer applets may be used in an
alternate embodiment. For eXample, the functions of the
monitor, policy enforcement and con?guration applets 178,
179, 180 may be implemented in a single applet. Alterna
tively, multiple applets may be provided to perform policy
enforcement or other functions using different queue assign
ment criteria. For eXample, one policy enforcement applet
may make priority queue assignments based on destination
MAC addresses, While another policy enforcement applet
makes priority queue assignments based on error rates or
line utiliZation of higher level protocols. Multiple monitor
applets or con?guration applets may similarly be provided.
[0043] Although queue assignment policy based on des
tination MAC address is illustrated in FIG. 6, myriad
different queue assignment criteria may be used in other
embodiments. For eXample, instead of monitoring and
updating queue assignment based on traf?c to destination
MAC addresses, queue assignments may be updated on
other traf?c patterns, including traf?c to speci?ed destination
ports, traf?c from speci?ed source ports, traf?c from speci
?ed source MAC addresses, traf?c that forms part of a
speci?ed IP ?oW, traf?c that is transmitted using a speci?ed
protocol (e.g., HTTP, FTP or other protocols) and so forth.

US 2004/0076161 A1
Also, queue assignments may be updated based on environ
mental conditions such as time of day, changes in network
con?guration (e.g., due to failure or congestion at other
netWork nodes), error rates, packet drop rates and so forth.
Monitoring, policy enforcement and con?guration applets
that combine many or all of the above-described criteria may
be implemented to provide sophisticated traf?c handling
capability in a packet forWarding device.
[0044] Although dynamic assignment of traf?c classes to
a priority egress queue has been emphasiZed, the methods
and apparatuses described herein may alternatively be used
to assign traf?c classes to a hierarchical set of queues
anyWhere in a packet forWarding device including, but not
limited to, ingress queues and queues associated With deliv
ering and receiving packets from the sWitching fabric.
Further, although the queue assignment of traf?c classes has
been described in terms of a pair of queues (priority and best
effort), additional queues in a prioritiZation hierarchy may be
used Without departing from the spirit and scope of the
present invention.
[0045] In the foregoing speci?cation, the invention has
been described With reference to speci?c exemplary embodi
ments thereof. It Will, hoWever, be evident that various
modi?cations and changes may be made to the speci?c
exemplary embodiments Without departing from the broader
spirit and scope of the invention as set forth in the appended
claims. Accordingly, the speci?cation and draWings are to be
regarded in an illustrative rather than a restrictive sense.
What is claimed is:
1. In a packet forWarding device, a method comprising:
monitoring bandWidth consumption by one or more types
of packet traf?c received in the packet forWarding
device;
determining Whether the bandWidth consumption by the
one or more types of packet traf?c exceeds a threshold;
and
automatically changing assignment of at least one type of
packet traf?c of the one or more types of packet traf?c
from a queue having a ?rst priority to a queue having
a second priority if the bandWidth consumption exceeds
the threshold.
2. The method of claim 1 further comprising receiving
program code in the packet forWarding device after instal
lation of the packet forWarding device in a packet commu
nications netWork and Wherein said monitoring, determining
and automatically changing is implemented by the executing
program code.
3. The method of claim 2 Wherein receiving the program
code comprises receiving a sequence of virtual machine
instructions and Wherein executing the program code com
prises executing the sequence of virtual machine instructions
using a virtual machine included in the packet forWarding
device.
4. The method of claim 3 Wherein receiving the sequence
of virtual machine instructions comprises receiving a
sequence of Java byte codes and Wherein executing the
sequence of virtual machine instructions using a virtual
machine comprises executing the sequence of Java byte
codes in a Java virtual machine included in the packet
forWarding device.
Apr. 22, 2004
5. The method of claim 1 Wherein monitoring bandWidth
in the packet forWarding device comprises determining a
measure of bandWidth consumption in the packet forWarding
device due to traf?c associated With a physical port on the
forWarding device.
device due to traf?c associated With a particular netWork
address.
7. The method of claim 6 Wherein determining a measure
of bandWidth consumption in the packet forWarding device
due to traf?c associated With the particular netWork address
comprises determining a measure of bandWidth consump
tion due to traf?c associated With a particular media access
control (MAC) address.
device due to traffic associated With a particular communi
cations protocol.
9. The method of claim 8 Wherein determining a measure
of bandWidth consumption in the packet forWarding device
due to traf?c associated With the particular communications
protocol comprises determining a measure of bandWidth
consumption in the packet forWarding device due to traf?c
associated With at least one of the folloWing protocols: ?le
transfer protocol (FTP), hyper-text transfer protocol
(HTTP), transmission control protocol/internet protocol
(TCP/IP).
10. A packet forWarding apparatus comprising:
a plurality of input/output (I/O) ports to transmit and
receive packets of information;
?rst and second queues to buffer the packets prior to
transmission via one or more of the I/O ports, packets
buffered in the ?rst queue having higher transmission
priority than packets buffered in the second queue;
queue assignment logic to assign the packets to be buff
ered in either the ?rst queue or the second queue
according to a packet type associated With each packet,
each of the packets being associated With at least one of
a plurality of packet types; and
one or more agents to monitor bandWidth consumption by
packets associated With a ?rst packet type of the
plurality of packet types and to automatically change
assignment of packets associated With the ?rst packet
type from the ?rst queue to the second queue if band
Width consumption of packets associated With the ?rst
packet type exceeds a threshold.
11. The apparatus of claim 10 further comprising:
a processing unit coupled to the plurality of I/O ports, the
processing unit including a memory and a processor;
and
a data communications interface to receive program code
in the memory of processing unit after installation of
the packet forWarding apparatus in a packet communi
cations netWork and Wherein the one or more agents are

US 2004/0076161 A1
implemented by execution of the program code in the
processor of the processing unit.
12. The apparatus of claim 11 Wherein the packet for
Warding apparatus further comprises program code that,
When executed by the processing unit, implements a virtual
machine, and Wherein the program code received via the
data communications interface comprises a sequence of
instructions that is executed by the virtual machine to
implement one or more agents.
13. The apparatus of claim 12 Wherein the program code
received via the data communications interface includes a
sequence of Java byte codes and Wherein the virtual machine
is a Java virtual machine.
14. The apparatus of claim 10 Wherein the ?rst packet type
comprises packets associated With a particular one of the I/O
ports.
comprises packets comprises packets associated With a
particular netWork address.
16. The apparatus of claim 15 Wherein the particular
netWork address is a particular media access control (MAC)
address.
comprises packets comprises packets associated With a
particular communications protocol.
communications protocol is a hyper-text transfer protocol
(HTTP).
communications protocol is a ?le transfer protocol (FTP).
20. A communications netWork comprising a packet for
Warding device, the packet forWarding device including:
a plurality of input/output (I/O) ports to transmit and
receive packets of information from one or more other
devices in the communications netWork
?rst and second queues to buffer the packets prior to
transmission via one or more of the I/O ports, packets
buffered in the ?rst queue having higher transmission
priority than packets buffered in the second queue;
queue assignment logic to assign the packets to be buff
ered in either the ?rst queue or the second queue
Apr. 22, 2004
according to a packet type associated With each packet,
each of the packets being associated With at least one of
a plurality of packet types; and
one or more agents to monitor bandWidth consumption by
packets associated With a ?rst packet type of the
plurality of packet types and to automatically change
assignment of packets associated With the ?rst packet
type from the ?rst queue to the second queue if band
Width consumption of packets associated With the ?rst
packet type exceeds a threshold.
21. The communications netWork of claim 20 Wherein the
packet forWarding device further includes:
a processing unit coupled to the plurality of I/O ports, the
processing unit including a memory and a processor;
and
a data communications interface to receive program code
in the memory of processing unit after installation of
the packet forWarding device in the communications
netWork and Wherein the one or more agents are
implemented by execution of the program code in the
processor of the processing unit.
22. The communications netWork of claim 21 Wherein the
packet forWarding device further includes program code
that, When executed by the processing unit, implements a
virtual machine, and Wherein the program code received via
the data communications interface includes a sequence of
instructions that is executed by the virtual machine to
implement one or more agents.
23. In a packet forWarding device, a method comprising:
monitoring an error rate associated With one or more types
of packet traf?c received in the packet forWarding
device;
determining Whether the error rate associated With the one
or more types of packet traf?c exceeds a threshold; and
automatically changing assignment of at least one type of
packet traffic of the one or more types of packet traf?c
from a queue having a ?rst priority to a queue having
a second priority if the error rate exceeds the threshold.
* * * * *

Dynamic assignment of traffic classes to a priority queue in a packet forwarding device

More Related Content

What's hot (20)

Similar to Dynamic assignment of traffic classes to a priority queue in a packet forwarding device (20)

More from Tal Lavian Ph.D. (20)

Recently uploaded (20)

Dynamic assignment of traffic classes to a priority queue in a packet forwarding device