system event log - school of computingmind/papers/deeplog_short.pdf · 2017-12-04 · 29 deeplog a...

Post on 10-Mar-2020

0 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

TRANSCRIPT

2

3

System Event LogExist practically on

every computer system!

Automatic Analysis?

4

SystemEventLog

Started service A on port 80Executor updated: app-1 is now LOADING……

5

SystemEventLog

Structured Data

Log key

printf(“Started service %s on port %d”, x, y);

LOG PARSING

Started service A on port 80Executor updated: app-1 is now LOADING……

Started service * on port *Executor updated: * is now LOADING……

6

SystemEventLog

Structured Data Anomaly Detection

LOG ANALYSIS

LOG PARSING

Started service A on port 80Executor updated: app-1 is now LOADING……

Started service * on port *Executor updated: * is now LOADING……

Log key

printf(“Started service %s on port %d”, x, y);

7

8

SPELLA streaming log

parser published in ICDM’16

Deletion of file1 complete. Deletion of file1 complete.

log keylog message

Deletion of file2 complete. Deletion of * complete.

parameters

[ ]

[file2]

9

DeepLog

Anomaly Detection Diagnosis

10

TrainingStage

DetectionStage

MODELS

11

DetectionStage

MODELS

12

DetectionStage

MODELS

13

DetectionStage

MODELS

14

DetectionStage

MODELS

15

TrainingStage

MODELS

16

TrainingStage

MODELS

17

TrainingStage

MODELS

18

TrainingStage

MODELS

19

TrainingStage

MODELS

20

TrainingStage

MODELS

21

Use long short-term memory (LSTM) architecture

In detection stage, DeepLog checks if the actual next log key is among its top g probable predictions.

22

Method 1: Using LSTM prediction probabilities

Method 2: Using co-occurrence matrix

23

24

Multi-variate time series data anomaly detection problem!--- Leverage LSTM to check reconstruction error.

25

Evaluation results on HDFS log data. (over a million log entries with labeled anomalies)

PCA (SOSP’09), IM (UsenixATC’10), N-gram (baseline language model)

26

Evaluation results on Blue Gene/L log, with and without online model update.

27

Evaluation results on OpenStack cloud log with different confidence intervals (CIs)

28

Diagnosis using constructed workflow.

Injected anomaly: during VM creation, network speed from controller to compute node is throttled.

29

DeepLog

➢ A realtime system log anomaly detection framework.

➢ LSTM is used to model system execution paths and log parameter values.

➢ Workflow models are built to help anomaly diagnosis.

➢ It supports online model update.

Min Dumind@cs.utah.edu

top related