Method for metric ranking in invariant networks of distributed systems
View Patent ↗A method for metric ranking in invariant networks includes, given an invariant network and a set of broken invariants, two ranking processes are used to determine and rank the anomaly scores of each monitoring metrics in large-scale systems. Operators can follow the rank to investigate the root-cause in problem investigation. In a first ranking process, given a node/metric, the method determines multiple scores by integrating information from immediate neighbors to decide the anomaly score for metric ranking. In a second ranking process, given a node/metric, an iteration process is used to recursively integrate the information from immediate neighbors at each round to determine its anomaly score for metric ranking.
1. A method for metric ranking in invariant networks, the method comprising:
considering an invariant network and a set of broken invariants in said invariant network;
under first process, given a node/metric of said invariant network, determining multiple scores by integrating information from immediate neighboring nodes of said node to decide an anomaly score of said node for metric ranking; and
under a second process, given a node/metric of said invariant network, using an iteration to recursively integrate information from immediate neighboring nodes of said at each round to determine an anomaly score of said node for metric ranking,
wherein said first and second process facilitate a determination and ranking of said anomaly scores of each monitoring metrics in large scale invariant network systems for enabling following said ranking to investigate a problem in said invariant network, and
wherein said first process comprises determining an iScore for every node in the invariant network, said iScore of a node, V i (1<i<N), within an invariant network I being based on
iScore
V
i
=
number
of
broken
links
of
V
i
number
of
all
links
of
V
i
.
2. The method of claim 1 ,
wherein said first process comprises determining an ixScore for every node in the invariant network,
said ixScore being used to measure anomaly degree of each node in said invariant network, and
said ixScore combining multiple evidences from a node itself and its neighboring nodes to infer its anomaly degree.
3. The method of claim 1 ,
wherein said first process comprises determining an rScore is based on
rScore
(
V
i
)
=
∑
k
=
1
K
r
ia
k
K
,
where for a node Vi with K related broken links, there are K ratios r iak (1_k_K),
where ak is the node index which means that the node Vak is connected to node Vi via a broken link.
4. The method of claim 1 ,
wherein said second process comprises determing wiScore, which is a weighted iScore,
said wiScore of a node, V i (1≧i≧N), within an invariant network I being based on
wiScore
V
i
=
∑
V
k
∈
BINNs
of
V
i
(
1
-
iScore
V
k
)
*
1
number
of
all
links
of
V
i
,
where V k denotes an individual node of BINNs of node V i and BINNs represents broken-invariant-neighboring-nodes of a node, each of which connects to said node with a broken link.
5. The method of claim 4 ,
wherein said second process comprises wiScore being updated for every node in said invariant network based on
wiScore
V
i
r
+
1
=
∑
V
k
∈
BINNs
of
V
i
(
1
-
wiScore
V
k
r
)
*
1
number
of
all
links
of
V
i
,
where wiScore V i r+1 denotes the wiScore of node V i at the end of round r+1 iteration.
6. A method for metric ranking in invariant networks, the method comprising:
considering an invariant network and a set of broken invariants in said invariant network;
under first process, given a node/metric of said invariant network, determining multiple scores by integrating information from immediate neighboring nodes of said node to decide an anomaly score of said node for metric ranking; and
under a second process, given a node/metric of said invariant network, using an iteration to recursively integrate information from immediate neighboring nodes of said at each round to determine an anomaly score of said node for metric ranking,
wherein said first and second process facilitate a determination and ranking of said anomaly scores of each monitoring metrics in large scale invariant network systems for enabling following said ranking to investigate a problem in said invariant network, and
wherein said first process comprises determining an xScore for every node in the invariant network, said xScore of a node, V i (1≧i≧N), within an invariant network I being based on
xS
core
V
i
=
1
-
number
of
broken
links
related
to
BINNs
number
of
all
links
related
to
BINNs
,
where BINNs represents broken-invariant-neighboring-nodes of a node, each of which connects to said node with a broken link.
7. The method of claim 6 ,
wherein said first process comprises determining an ixScore for every node in the invariant network,
said ixScore being used to measure anomaly degree of each node in said invariant network, and
said ixScore combining multiple evidences from a node itself and its neighboring nodes to infer its anomaly degree.
8. The method of claim 6 ,
wherein said first process comprises determining an rScore is based on
rScore
(
V
i
)
=
∑
k
=
1
K
r
ia
k
K
,
where for a node Vi with K related broken links, there are K ratios r iak (1_k_K),
where ak is the node index which means that the node Vak is connected to node Vi via a broken link.
9. The method of claim 6 ,
wherein said second process comprises determing wiScore, which is a weighted iScore, said wiScore of a node, V i (1≧i≧N), within an invariant network I being based on
wiScore
V
i
=
∑
V
k
∈
BINNs
of
V
i
(
1
-
iScore
V
k
)
*
1
number
of
all
links
of
V
i
,
where V k denotes an individual node of BINNs of node V i and BINNs represents broken-invariant-neighboring-nodes of a node, each of which connects to said node with a broken link.
10. The method of claim 9 ,
wherein said second process comprises wiScore being updated for every node in said invariant network based on
wiScore
V
i
r
+
1
=
∑
V
k
∈
BINNs
of
V
i
(
1
-
wiScore
V
k
r
)
*
1
number
of
all
links
of
V
i
,
where wiScore V i r+1 denotes the wiScore of node V i at the end of round r+1 iteration.