C LAUDIO A A RDAGNA, E RNESTO D AMIANI, F ULVIO F RATE, D AVIDE R EBECCANI Universita degli Studi di Milano, Italy M ARCO U GHETTI Telecom Italia, TILab,

Scalability Patterns for Platform-As-A-Service

CLAUDIO A ARDAGNA, ERNESTO DAMIANI,

FULVIO FRATE, DAVIDE REBECCANI

Universita degli Studi di Milano, Italy

MARCO UGHETTITelecom Italia, TILab, Italy

13th November, 2012.

Presented By,Chidambara Nadig.

2012 IEEE 5TH INTERNATIONAL CONFERENCE ON CLOUD COMPUTING

Scalability Patterns for Platform-as-a-Service

2

Abstract Platform as a Service is a cloud based approach where enterprises have little to

do with the underlying cloud infrastructure.

Installing, configuring and managing the underlying middleware, operating

system and hardware is done by the cloud provider.

Thus, Scalability becomes an important factor to decide,

The capabilities of each virtual resource in the cloud.

The number of resources in the cloud.

This paper presents few Scalability Patterns for PAAS infrastructure and a

method to automatically manage scalability.


3

IAAS – Infrastructure as a Service.

PAAS – Platform as a Service.

SAAS – Software as a Service.

IAAS, PAAS and SAAS


4


5

Levels of Abstraction in Cloud Services.


6

Some examples


7

Actions on Resource ScalingVertical Scaling – Scale Up – Additional Resources are added

to a single machine when the load increases. The resources can

be either physical resources added to a server, or virtual

resources dynamically assigned to a virtual machine or its

applications.

Horizontal Scaling – Scale Out – New machines are added to

the system providing more software and hardware resources.

Scale Down – Releasing Resources when they are not necessary.


8

Scalability Patterns

1. Single Platform Pattern (SPP)

2. Shared Platform Pattern (ShPP)

3. Clustered Platform Pattern (CPP)

4. Multiple Shared Platform Patter (MShPP)

5. Multiple Clustered Platform Patter (MCPP)


9

Single Platform Pattern (SPP)Each customer is given a complete virtual machine with

a platform installed on it. SPP is single tenant scenario.

Resource Utilization is scarce.

Scalability is low because the number of virtual

machines and platforms is linear in the number of

customers.


10

Shared Platform Pattern (ShPP)A Multitenant scenario.

One platform is installed on a set of virtual machines and is

shared by multiple tenants.

Each tenant has a right to manage a portion of the platform and

deploy their services on it independently.

Performance of the platform is maintained by up-scaling and

down-scaling the resources assigned to the virtual machine.


11

Whenever the load increases – degrading the performance metrics of

the platform – RAM, CPU, or bandwidth can be increased.

On the other hand, when the load decreases, resources can be freed

and made available to other processes in the architecture.

In ShPP resources are shared and therefore need to be managed to

ensure security and isolation among tenants.

Provides High utilization.

However, ShPP doesn’t provide linear scalability increase due to

increased overheads for resource management.


12

Clustered Platform Pattern (CPP)A single platform is deployed supporting clustering and is shared

by all tenants.

Multiple instances of the platform components can be deployed

in different machines of the cluster.

Similar to ShPP, CPP manages shared resources preserving

security and isolation among tenants.

This pattern also implements load balancing, PAAS monitoring,

and elastic auto-scaling.


13

CPP provides high resource utilization, since the machines

in the cluster are shared among different tenants.

CPP provides some scalability as system resources can be

incrementally extended.

The Clustered Platform Pattern also promises high

reliability and availability due to increased redundancy.


14

Multiple Shared Platform Pattern (MShPP)

MShPP is an extension of ShPP.

Initially a single Shared Platform is deployed.

Upon an increase in the load, additional resources (CPU, RAM or

bandwidth) are assigned to maintain the performance metrics.

In case additional resources are not sufficient, a new platform is deployed

and a part of the existing tenants are migrated to the new platform along

with the resources they own.


15

When the load decreases, the additional platforms can be removed causing the

tenants to migrate back to the available platforms.

MShPP has lower manageability than ShPP owing to the fact that tenants have to

be migrated from one platform to another when a new platform is deployed.

MShPP provides high resource utilization.

Its scalability depends on the specific scenario and number of deployed platforms.

In the worst case, when all tenants experienced a traffic peak, a platform is deployed

for each tenant and therefore scalability of MShPP is equivalent to the one of SPP.

In the average case, MShPP provides high scalability.


16

Multiple Clustered Platform Pattern (MCPP) MCPP is an extension of CPP.

At initialization time, a single, shared, multi-tenant platform supporting

clustering is deployed.

Upon an increase in the load, additional resources (i.e., machines in the cluster)

are added to maintain the performance level.

In case the extended cluster is not sufficient to manage the new load, a new

platform supporting clustering is deployed, and a part of the existing tenants

are migrated to the new clustered platform together with the services they own.


17

When the load decreases, the additional platforms can be removed causing

the tenants to migrate back to the available platforms.

MCPP has the lowest manageability among patterns.

MCPP usually provides high utilization of resources, although utilization

may decrease in case of multiple platform deployments.

Promises

High Scalability.

High Availability.

High Reliability.


18

Overview


19

Performance Measurement1. Performance Metrics at the Platform Level

Total Count (TC) – Number of messages forwarded to a given end

point. If the metrics exceeds a known threshold, the performance

could be affected and an alarm is raised.

Fault Count (FC) – Number of messages that resulted in a fault while

being forwarded to the end point.

Minimum Time (MinT)

Maximum Time (MaxT)

Average Time (AveT)

Time Taken to send a request to an end point and receive a response.


20

2. Performance Metrics at the Host Level

CPU Load (CL) – CPU Utilization on host and guest systems. High values of CL in

a Virtual Machine signifies a problem in the fulfillment of request messages

backlog.

Memory Occupancy (MO) – Memory Utilization on host and guest systems.

Services that require a huge amount of data may require substantial portions of

memory at the detriment of other services.

Network Utilization (NU) – Utilization of the network bandwidth. High values of

NU may suggest re-allocation of external resources to manage a peak of requests.

Host Availability (HA) – Number of virtual machines available and accessible

through the network. The falling of the HA under a pre-defined threshold

indicates the new for new machines.


21

Performance MonitoringBased on the certain measurements of the Performance

metrics, certain alarms are raised.

Two Categories of Alarms:

1. Message Alarm – A message alarm is raised when:

System is not able to manage the message queue efficiently.

Average message delivery time is above a preset threshold.

The difference between the maximum and minimum message time is

above a preset threshold.

2. Processing Alarm – A process alarm is raised when service

execution may involve high execution time or a lot of resources.


22

Alarm Rules

HIGH and LOW thresholds in the above table can be defined on

the basis of previous experimental tests and/or expert knowledge.


23

Alarm-driven selection of scalability patterns

The initial node ∗ represents the basic installation scenario in which different tenants share the same platform with default configurations.


24

The two-fold Monitoring Approach Upon an increase in the load that raises a message alarm, the algorithm moves

to node ShPP and applies a ShPP pattern

If a processing alarm is raised, the algorithm moves to node CPP and applies a

CPP pattern.

When the ShPP pattern is not sufficient to solve further alarms, it moves to node

MShPP in case of message alarms or to node CPP in case of processing alarms.

The algorithm moves from CPP to MCPP for both types of alarms, while it moves

from MShPP to MCCP in case of processing alarms.


25

Experimental Setting WSO2 Platform is a cloud-deployable, Java-based service-oriented platform.

A WSO2 platform with default configurations is used as the experimental environment.

A realistic scenario is simulated where concurrent requests come from different clients.

Each test case starts with 20 active clients sending SOAP (Simple Object Access Protocol) requests, which ramp up to a maximum of 100 clients.

All test cases have a duration of 60 seconds.

Load Varying is done by increasing the requests per second (rps) from 10rps to 500 rps.


26

Baseline Measurement without security

RT – Response Time (in a logarithmic scale)

TPS – Transactions per Second

rps – Requests per second


27

Baseline Measurement with security

RT – Response Time (in a logarithmic scale)

TPS – Transactions per Second

rps – Requests per second


28

Performance of ShPP without security


29

Performance of ShPP with security


30

Performance of CPP without security


31

Performance of CPP with security


32

Comment..Result 1 – Security causes a substantial decrease in

the performance of a SOA deployed on the cloud.

Result 2 – ShPP results in a performance gain both on

TPS and RT with respect to the baseline.

Result 3 – CPP provides a further improvement with

respect to ShPP.


33

THANK YOU!

C LAUDIO A A RDAGNA, E RNESTO D AMIANI, F ULVIO F RATE, D AVIDE R EBECCANI Universita degli Studi di Milano, Italy M ARCO U GHETTI Telecom Italia, TILab,

Documents

C LAUDIO A A RDAGNA, E RNESTO D AMIANI, F ULVIO F RATE, D AVIDE R EBECCANI Universita degli Studi di Milano, Italy M ARCO U GHETTI Telecom Italia, TILab,