Toward Fine-Grained,
Unsupervised, Scalable Performance Diagnosis for Production Cloud Computing
Systems
ABSTRACT:
Performance diagnosis is labor intensive in
production cloud computing systems. Such systems typically face many real world
challenges, which the existing diagnosis techniques for such distributed
systems cannot effectively solve. An efficient, unsupervised diagnosis tool for
locating fine-grained performance anomalies is still lacking in production
cloud computing systems. This paper proposes CloudDiag to bridge this gap.
Combining a statistical technique and a fast matrix recovery algorithm,
CloudDiag can efficiently pinpoint fine-grained causes of the performance
problems, which does not require any domain-specific knowledge to the target
system. CloudDiag has been applied in a practical production cloud computing
systems to diagnose performance problems. We demonstrate the effectiveness of
CloudDiag in three real-world case studies
EXISTING SYSTEM:
Typical productions in existing cloud systems are
service-oriented in nature. The response time of user requests directly
reflects the system performance. In this regard, tracing user requests is a
viable means to exposing performance data, so as to help performance diagnosis.
Recent work has shown that it is promising to pinpoint performance anomalies
with end-to-end request tracing data. However, an efficient, unsupervised diagnosis
tool for locating fine-grained performance anomalies is still lacking.
DISADVANTAGES
OF EXISTING SYSTEM:
v It
is very challenging to localize anomalous methods as well as their
corresponding physical replicas. Consequently, huge
human efforts are still required to further pinpoint the subtle primary cause.
v It
is extremely difficult to maintain the behavior models for such evolutional
systems. Hence, a performance diagnosis tool for production cloud systems
should be completely unsupervised, without assuming that any prior knowledge
about the service should be input.
v Coping
with large runtime data generated by a production cloud
system efficiently is a challenging task in performance diagnosis. Many defects
only manifest themselves in an online production cloud that involves a large
number of component replicas.
PROPOSED SYSTEM:
This paper bridges this gap by proposing CloudDiag. CloudDiag
periodically collects the end-to-end tracing data (In particular, execution
time of method invocations) from each physical node in the cloud. It then
employs a customized Map-Reduce algorithm to proactively analyze the tracing
data. Specifically, it assembles the tracing data of each user request, and
classifies the tracing data into different categories according to call trees
of the requests.
ADVANTAGES
OF PROPOSED SYSTEM:
ü CloudDiag
has been successfully launched in diagnosing performance problems for the
production cloud systems in Alibaba Cloud Computing.
ü We
report three case studies in our real-world performance diagnosis experiences
to demonstrate the effectiveness of CloudDiag in helping the operators localize
the primary causes of performance problems.
SYSTEM ARCHITECTURE:
MODULES:
v Cloud
Server Module
v Data
Owner Module
v User
Module
v Data
Collection and Assembling Module
v Diagnosing
Module
MODULES DESCRIPTION:
Cloud
Server Module
In this module, we design a Cloud server in local
host by having the functionalities of a Cloud Storage system. Where the data
owners can upload their files securely and also use it. The Cloud Server allows
the access of authorized users to access the files of the data owners too.
Data
Owner Module
In this module, we designed a data owner module by
having a unique registration for each data owner such that a new data owner
should register in cloud server and then get their access to log in. After that
the data owner has the facility to upload their data in the cloud server.
User
Module
In this module first the user has to be get access
by registering themselves when they are new to the system. Once after
registration they can login the system and can find the details of the files
uploaded by the data owner. The authorized users can make their request to
download the file. RequestID should be unique for every request. It is assigned
when a request arrives the system.
Data
Collection and Assembling Module
In this module we develop the system to have data
collection and assembling them into it. It is named as CloudDiag. It traces
user requests at a given sampling rate to expose performance data. For the
sampled requests, each component replica records the performance data and saves
them in its local storage. An important consideration is what kind of
performance data CloudDiag should collect and how. CloudDiag adopts an
instrumentation-based approach that collects the execution time of each
component method. CloudDiag should first
assemble the performance data distributed in numerous component replicas in a
request-oriented way. In other words, the performance data belonging to the
same requests are correlated together. CloudDiag will then analyze such
request-oriented performance data and infer the call tree of each sampled
request. A customized map-reduce process is utilized to group requests into
different categories based on their call trees.
Diagnosing
Module
From the data collected by the previous module we
design diagnosing module. Where, CloudDiag then identifies the anomalous
categories according to their latency distribution. We can identify the
anomalous method by measuring the execution time deviation of each method one by
one. We design an unsupervised machine learning algorithm to automatically
learn the characteristics of the invoked methods and identify which methods are
anomalous together with on which replicas they are executed
SYSTEM REQUIREMENTS:
HARDWARE REQUIREMENTS:
·
System : Pentium IV 2.4 GHz.
·
Hard Disk : 40 GB.
·
Monitor :
15 inch VGA Colour.
·
Mouse :
Logitech Mouse.
·
Ram : 512 MB
·
Keyboard :
Standard Keyboard
SOFTWARE REQUIREMENTS:
·
Operating System : Windows XP.
·
Coding Language : ASP.NET, C#.Net.
·
Database :
SQL Server 2005
REFERENCE:
Haibo Mi,Student Member, IEEE, Huaimin Wang, Member,
IEEE, Yangfan Zhou, Member, IEEE, Michael Rung-Tsong Lyu,Fellow, IEEE, and Hua
Cai, Member, IEEE “Toward Fine-Grained, Unsupervised, Scalable Performance
Diagnosis for Production Cloud Computing Systems” - IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. 24, NO. 6,
JUNE 2013.
