Method and apparatus for local adaptive provisioning at a node
A method ( 300 ) and apparatus ( 200 ) for local adaptive provisioning at a node is disclosed. The method may include determining ( 320 ) a per packet latency for a class of packet network traffic in a queue of a plurality of queues for a plurality of classes of packet network traffic at a node, establishing ( 330 ) a reward function for the class of packet network traffic based on a packet latency limit, based on the per packet latency, and based on a source rate for the class of packet network traffic, and adjusting ( 340 ) provisioning of a queue at the node based on the reward function.
1. A method comprising:
determining a per packet latency for a class of packet network traffic in a queue of a plurality of queues for a plurality of classes of packet network traffic at a node;
establishing, using a local adaptive provisioning module at the node, a reward function for the class of packet network traffic based on a packet latency limit, based on the per packet latency, and based on a source rate for the class of packet network traffic; and
adjusting provisioning of a queue servicing module for the queue at the node based on the reward function, where the provisioning comprises the priority of accessing each of the plurality of queues by the queue servicing module,
wherein the reward function penalizes end to end packet failures as measured by end to end latency exceeding a threshold based on relative weighting assigned to failures of each class, and
wherein the reward function is further based on an average hop count and based on at least one other class of packet network traffic.
2. The method according to claim 1 , wherein determining a per packet latency includes determining a number of packets for a class of packet network traffic in a queue of a plurality of queues for a plurality of classes of packet network traffic at the node.
3. The method according to claim 1 , wherein the reward function reflects an anticipated end-to-end success of the class of packet network traffic.
4. The method according to claim 1 , further comprising determining changes in the reward function with respect to changes in the source rate for the class of packet network traffic with respect to at least one other class of packet network traffic,
wherein adjusting comprises increasing provisioning for the class of packet network traffic to maximize the change in the reward function.
5. The method according to claim 4 , wherein adjusting comprises increasing provisioning for the class of packet network traffic if the change in the reward function is maximized and above a threshold.
6. The method according to claim 1 , wherein the per packet latency comprises a mean per packet latency.
7. The method according to claim 1 , wherein the reward function is further based on additional measured information from a network management system.
8. The method according to claim 1 , wherein adjusting the provisioning of a queue at the node is based on a change in provision Δs for a class i according to:
Δ
s
i
=
α
(
ⅆ
R
ⅆ
s
i
+
c
)
,
where α represents a scaling factor, R represents the reward function, dR/ds i represents a rate of change in the reward function, R, with respect to a throughput, s, for a given class, i, and c represents a normalizing constant.
9. An apparatus comprising:
an input configured to receive a packet;
a plurality of queues coupled to the input, the plurality of queues configured to queue a plurality of classes of packet network traffic at a node, where the plurality of queues comprise a plurality of hardware queues;
a queue servicing module coupled to the plurality of queues, the queue servicing module configured to access the plurality of queues to give priority to higher packet priority level queues; and
a local adaptive provisioning module coupled to the plurality of queues and coupled to the queue servicing module, the local adaptive provisioning module configured to determine a per packet latency for a class of packet network traffic in a queue of the plurality of queues, establish a reward function for the class of packet network traffic based on a packet latency limit, based on the per packet latency, and based on a source rate for the class of packet network traffic, and adjust provisioning of the queue servicing module for the queue at the node based on the reward function
wherein the reward function penalizes end to end packet failures as measured by end to end latency exceeding a threshold based on relative weighting assigned to failures of each class, and
wherein the reward function is further based on an average hop count and based on at least one other class of packet network traffic.
10. The apparatus according to claim 9 , wherein the local adaptive provisioning module determines a per packet latency by determining a number of packets for a class of packet network traffic in the queue of the plurality of queues for a plurality of classes of packet network traffic at the node.
11. The apparatus according to claim 9 , wherein the reward function reflects an anticipated end-to-end success of the class of packet network traffic.
12. The apparatus according to claim 9 , wherein the local adaptive provisioning module is configured to determine changes in the reward function with respect to changes in the source rate for the class of packet network traffic with respect to at least one other class of packet network traffic and adjust the provisioning by increasing provisioning for the class of packet network traffic to maximize the change in the reward function.
13. The apparatus according to claim 12 , wherein the local adaptive provisioning module increases provisioning for the class of packet network traffic if the change in the reward function is maximized and above a threshold.
14. The apparatus according to claim 9 , wherein the per packet latency comprises a mean per packet latency.
15. The apparatus according to claim 9 , wherein the local adaptive provisioning module adjusts the provisioning of a queue at the node based on a change in provision Δs for a class i according to:
Δ
s
i
=
α
(
ⅆ
R
ⅆ
s
i
+
c
)
,
where α represents a scaling factor, R represents the reward function, dR/ds i represents a rate of change in the reward function R with respect to a throughput s for a given class i, and c represents a normalizing constant.
16. A method comprising:
determining an average per packet latency for a class of packet network traffic based on a number of packets for the class of packet network traffic in a queue of a plurality of queues for a plurality of classes of packet network traffic at a node;
establishing, using a local adaptive provisioning module at the node, a reward function for the class of packet network traffic based on a packet latency limit, based on the per packet latency, and based on a source rate for the class of packet network traffic; and
adjusting provisioning of the queue at the node based on the reward function based on a change in provision Δs for a class i according to:
Δ
s
i
=
α
(
ⅆ
R
ⅆ
s
i
+
c
)
,
where α represents a scaling factor, R represents the reward function, dR/ds i represents a rate of change in the reward function R with respect to a throughput s for a given class i, and c represents a normalizing constant.