The Business Forum

"It is impossible for ideas to compete in the marketplace if no forum for
  their presentation is provided or available."           Thomas Mann, 1896


Getting more from Server Virtualization
Building an Adaptive Computing Infrastructure with Virtualization and ZXTM

Contributed by: Zeus Technology, Inc.

 

 

Introduction

An Adaptive Computing design uses virtualization, monitoring and traffic management together to build a new environment for your applications - an environment that works with the applications to ensure that they meet the needs of your business.

Server Virtualization has dramatically changed the landscape in the datacenter. Organizations are consolidating workloads from underutilized servers and are seeing large reductions in datacenter space, power, cooling and administration.

However, server consolidation is just the beginning of what can be achieved using virtualization technologies. Virtualization unleashes applications and compute workloads, breaking the ties that hold them to physical servers. This new-found freedom makes possible an entirely new datacenter architecture where the hardware serves the applications and the applications serve the business, rather than the other way round.

Server Consolidation

Server Consolidation was enabled by virtualization technologies like VMware’s Virtual Infrastructure 3. Servers running at 10-15% utilization were commonplace and a significant waste of valuable resources.

The most common driver for the adoption of virtualization was a desire to address these inefficiencies. Large numbers of server workloads could be consolidated onto a small number of powerful, reliable servers, running at higher utilizations. The availability of low-cost multi-core processors makes this a very cost effective proposition.

It’s not just server applications that can be virtualized in this way. Many hardware appliances - load balancers, firewalls, security gateways, spam filters, etc - can be replaced by equally capable virtual appliances. VMware’s Virtual Appliance program encourages this with a stringent Virtual Appliance certification program, where high quality virtual appliances like Zeus’ ZXTM product are available through the Virtual Appliance Marketplace1.

Adaptive Computing - the next step

The potential of server virtualization goes far further than simply the consolidation of server workloads.

Virtualized servers are much more flexible, mobile and efficient than their physically-bound predecessors. Perhaps the biggest change is that a server workload no longer needs to be regarded as a long-lived, expensive resource, tied to and limited by the hardware it is running on. Instead, hardware is simply a scalable resource that the compute fabric uses, and servers are completely disposable, created, restarted and destroyed on demand.

No longer does the business need to be enslaved by the capabilities of its applications. When applications can be deployed, resourced and scaled on demand, the workload running in the datacenter can be adapted to the needs of the business, rather than the other way round.

It should not be like this . . .

Imagine a rather old-fashioned business operation, restricted in how it can use its human resources by shortsighted working practices. Each worker could only be employed to do one specific task, even if the task did not merit a full-time position. Sickness and holiday cover was provided for each employee on a full-time basis. There was no reason to employ skilled workers with multiple talents because they could not be used to their full potential. It was so difficult to change working practices that it was easier to employ new staff than retrain existing ones.

This picture may seem implausibly antiquated, but it illustrates the limitations and inefficiencies that current datacenter architectures impose on business applications today:

  • Businesses wish to cut costs, but the one-workload-per-server model is wasteful and inefficient;
     
  • Businesses wish to be agile, but the need to tie server workloads to physical hardware makes change a slow and expensive proposition;
     
  • Businesses wish to invest in hardware wisely, but differing workloads have different requirements. A powerful multi-core server may be more cost efficient but is a poor investment if few of the current workloads merit it;
     
  • Businesses wish to deploy redundant servers to maintain business continuity during planned or unplanned downtime, but this redundancy multiplies the inefficiencies.

Virtualization is the new order

Virtualization decouples the server workload from the physical server and allows physical servers to run multiple workloads when resources allow. In terms of human resources, each worker can perform multiple tasks during their working hours and resources can be quickly redistributed to where they are needed most.

An Adaptive Computing architecture uses this metaphor. It redistributes workloads across the virtualized compute fabric on demand, driven by service level agreements and business requirements to serve the needs of the business.

Adaptive Computing

An Adaptive Computing architecture can be used with server applications that are made up of multiple components. This includes a wide range of enterprise applications; multi-tiered services may be comprised of web server, application server and database tiers, and SOA (Service Oriented Architecture) applications may be made up of tens or hundreds of discrete SOA components.

In a traditional, server-based environment, each service tier or group of SOA components is likely to be deployed on a dedicated physical server. Each tier or component communicates with the next using the network, most commonly using HTTP or an HTTP-based protocol such as SOAP.

Adaptive Computing breaks the ties with the physical servers making the application mobile, responsive and much easier to manage. There are four cornerstones to an Adaptive Computing architecture:

  • Virtualization of the workloads
     

  • Management of the virtual machines
     

  • Monitoring of correct operation and SLA conformance
     

  • Traffic Management between scalable and reliable components

Virtualization

The first cornerstone of Adaptive Computing architectures is the virtualization layer.  Virtualization makes it possible to deploy and resource server workloads flexibly and with minimal delay.

Server Virtualization, such as VMware’s Virtual Infrastructure 3, has made adaptive computing possible, but other virtualization technologies can be used to equal effect. For example, applications can be broken into discrete SOA components, and these components deployed across a cluster of Java application servers.

Tools like VMware’s VirtualCenterTM provide a central point to manage virtual machines.  A SOAP-based API (or equivalent) makes it very easy to drive the virtual infrastructure from other applications.

Management

The second cornerstone is the management layer. This lightweight layer contains the business and application management logic that controls how applications are deployed, being aware of the application limits and interdependencies. Each application will have its own unique requirements. As there are no commonly accepted standards as to how these are specified and deployed, the management layer is often built using custom code that interfaces to the standard APIs of the other parts of the architecture.

Monitoring

The monitoring capability detects when applications are performing outside their desired service level (for example, poor response times) or if an application component has failed completely. It informs the management layer, which can then chose to take remedial action such as deploying a new service instance or creating a new virtual machine.

Traffic Management

The traffic management capability allocates tasks to the individual workers in the virtualized compute fabric. It makes application scaling possible by load-balancing traffic across virtual machines, applying session persistence and migrating users from one generation to another.

The Monitoring and Routing functions can be provided by a sophisticated application traffic manager like Zeus’ ZXTM product.  

Introducing ZXTM

ZXTM is a non-intrusive, software traffic manager that load balances network services across clusters of servers, both physical and virtual. It is the only traffic management product available as a virtual appliance, and can be run directly on virtualization platforms from VMware, Microsoft, Xen and VirtualIron.

Managing Traffic

ZXTM provides a wide range of functions to manage traffic, from content-based routing to bandwidth and rate control. It contains a uniquely powerful traffic inspection engine and a configuration language called TrafficScriptTM that allows the end user to construct very sophisticated traffic management policies to meet their specific requirements.

Monitoring

ZXTM’s Service Level Monitoring capability monitors the response time of server nodes and services, raising alerts and modifying traffic management logic if services begin to fall outside desired Service Level Agreement (SLA) standards. Custom health monitors can perform complex tests against server nodes, detecting application and network-level errors.

Integration

The standards-based SOAP Control API makes it easy to integrate ZXTM with other management systems, reconfiguring ZXTM when a new server node is added. The SNMP interface allows another system to monitor ZXTM’s operating parameters, including service health and performance statistics. The pluggable alerting framework makes it possible for ZXTM to initiate a custom action when a critical event such as a server failure occurs.

The monitoring, traffic management and integration capabilities of ZXTM can be used with the reconfigurable virtualized compute resources to build an architecture that reacts and adapts to the needs of the applications it hosts.

Building an Adaptive Computing architecture

Setting the scene  Let us consider how we could take an existing application and construct an Adaptive Computing architecture around it.

Our application is comprised of:

  • A web-based front end, running Apache or IIS. The front end hosts static pages and content and proxies application requests back to one of the application servers.
     
  • Several application servers, authentication servers and custom gateways that perform different functions for the application.
     
  • Several data sources - local databases and remote services.
     

Each component is running as a virtual machine on a virtualization platform like VMware VI3:

Web Tier-Application Servers-Database Tier

This configuration is fragile and not scalable. If any one virtual machine were to fail or be overloaded with traffic, the entire application would fail. It is difficult to manage. Application upgrades cannot be easily tested and deployed without disrupting the live system.

Adding reliability and scalability

The first step in building the adaptive infrastructure is to use a traffic management device like ZXTM to load-balance and scale traffic within the infrastructure.

Where possible, each virtual machine should be cloned and a pair (or more) of machines run in its place. ZXTM can load-balance traffic across the cloned machines:

In this configuration, the ZXTM traffic manager sits ‘in line’, observing all transactions within the application, verifying the performance and correct operation of each tier and load-balancing transactions to the most appropriate virtual servers.

Web and Application servers in a compound application can always be scaled in this way. ZXTM’s session persistence capabilities can be used if a series of transactions must be pinned to a particular web or application server. Active-passive clustering with replication can be used for components such as databases that are less easily scaled because of state sharing requirements.

 

ZXTM monitors and manages all application transactions, routing each network request to the most appropriate, fastest responding virtual machine.

ZXTM can detect when a service is performing poorly, or if a virtual machine is producing errors or has locked up.

When services need re-provisioning, ZXTM informs the appropriate agent which restarts or redeploys servers in the virtualized environment.

When a new server is started, the provisioning agent informs ZXTM so that it can load-balance traffic to it.

 

Example 1: Fault Tolerance

A single virtual machine can be easily duplicated to a pair of identical machines, with little additional resources required. In general, these virtual machines should be run on separate hardware in the virtualized cluster, using appropriate affinity rules. In this case, even if one of the hardware servers hosting the virtual machines were to fail, the application will be unaffected.

A self-healing application

Virtual machines may fail in a variety of ways, most of which are undetectable by hardware watchdogs and heartbeats. For example, an unattended server may run out of local disk space (perhaps exacerbated by a denial of service attack) and the server software may lock up or return error messages rather than valid responses. When faced with such a problem, the quickest resolution is generally to reboot the server or (preferably) restart it from a trusted snapshot.

The customizable health monitors in ZXTM are able to detect a wide range of application errors, whether the application has hung or is returning invalid responses. When an error is detected, ZXTM will stop routing transactions to the failed machine until the monitors detect that it is operating correctly again. ZXTM raises an alert, and you can plug into the alerting framework to invoke custom actions when a server fails.

For example, if an administrator desired that a node is restarted from a snapshot when it failed, he could provide a script or application that communicated with VMware’s VirtualCenterTM management software using the management API to invoke this action.

In another case, it may be sufficient to provide a script that logs on to the failed server and restarts the software.

Example 2: Resource Scheduling

CPU resource scheduling is ineffective

Virtual machines run on shared servers and compete for resources - CPU, Memory and Network bandwidth. To prevent a runaway virtual machine from impacting the performance of other virtual machines, resource limits can be applied, but these are often ineffective.

CPU resource allocations can be split between virtual machines, but unless these are over-allocated, you’ll never achieve good hardware utilization. On the other hand, over-allocation carries the real risk that a small number of busy virtual machines will use all the available resources.

Furthermore, static CPU restrictions have very little bearing on the performance of the application. End users measure application performance in terms of response time, and organizations will wish to impose Service Level Agreements (SLAs) to define acceptable response times for each application or component.

ZXTM’s Service Level Monitoring Capability

ZXTM’s Service Level Monitoring capability monitors the response times for each service, and measures the service’s performance against a predefined SLA target.

Response times and SLA conformance statistics are published via SNMP, so an external agent could monitor these figures and resource services on the fly. Alternatively, ZXTM will raise an alert if a service falls outside of an SLA target. This alert could initiate a custom provisioning action, such as re-balancing resource allocations within the virtual machine manager, or deploying a new virtual machine and informing ZXTM (via the SOAP-based Control API) to load-balance traffic to that new machine.

So, just as a traditional application may fork a new worker thread or process when it becomes overloaded, with ZXTM you can fork a new virtual machine on demand.

Example 3: Rate Shaping

SLA-based resource balancing goes a long way towards ensuring that all applications receive the resources that they need, but an organization must be ready for exceptional situations where there aren’t sufficient resources to go round.

Consider the case where a virtual infrastructure is managing several applications. One of these applications comes under a denial of service attack, where users unintentionally or maliciously flood it with large numbers of requests. Attempts to drag it back to within its SLA by stealing resources from other applications causes all applications to suffer as there are not enough resources for all the applications.

Rate Shaping

In this case, it’s necessary to use a ‘brake’ that limits the number of transactions that the application is asked to perform. ZXTM’s rate shaping functionality can be used to prioritize and limit the number of requests to the application.

This has the effect that the application operates at peak efficiency within its resource allocation and does not become overwhelmed.

Rate shaping limits can be combined with Service Level Monitoring to build dynamic, adaptable traffic management policies that are only applied when service performance falls outside the desired SLA.

Example 4: Preparing for VMotion

A virtual machine may become unavailable during a short period of disruption. For example, if it is to be migrated from one VMware ESX host to another, network connections may stall and time out. Additionally, the VMotion process is hampered when the source VM is busy with large quantities of memory writes, which is common on a busy web or application server.

Connection Draining with ZXTM

ZXTM’s Connection Draining capability is used to take a machine out of service in a managed way. When a machine’s connections are drained, ZXTM does not route any new requests to the machine; only requests in already established sessions are routed to it. Once these sessions complete, the machine can be safely removed from service without any disruption.

A virtual machine may be prepared for VMotion by draining connections from it for several seconds beforehand. This will minimize the change of dropped transactions and speed up the VMotion process.


Conclusion

Virtualization is a long-term investment. Consolidation is a short-term gain, and Adaptive Computing is big prize to follow.

You can realize the benefits of Adaptive Computing when you understand the needs of your applications, the needs of your business and match the two together. A traffic management device like Zeus’ ZXTM Virtual Appliance is a key component of the Adaptive Computing architecture, managing application traffic and monitoring the health and performance of the applications.

Ultimately, this architecture can deliver an SLA-driven computing fabric that meets the needs of your business by managing the needs of your applications.


For Further Information

The ZXTM KnowledgeHub is a key resource for developers and system administrators wishing to learn about ZXTM and Zeus’ Traffic Management solutions. It is located at http://knowledgehub.zeus.com/


Visit the Authors Web Site

Website URL:

 http://www.zeus.com/

Your Name:
Company Name:
E-mail:

Inquiry Only - No Cost Or Obligation


3D Animation : red star  Click Here for The Business Forum Library of White Papers   3D Animation : red star


Search Our Site

Search the ENTIRE Business Forum site. Search includes the Business
Forum Library, The Business Forum Journal and the Calendar Pages.


Disclaimer

The Business Forum, its Officers, partners, and all other
parties with which it deals, or is associated with, accept
absolutely no responsibility whatsoever, nor any liability,
for what is published on this web site.    Please refer to:

legal description


Home    Calendar    The Business Forum Journal     Features    Concept    History
  Library    Formats    Guest Testimonials    Client Testimonials    Experts    Search  
News Wire
      Join Why Sponsor     Tell-A-Friend     Contact The Business Forum



The Business Forum

Beverly Hills, California United States of America

Email:  [email protected]

Graphics by DawsonDesign

Webmaster:  bruceclay.com
 


© Copyright The Business Forum Institute 1982 - 2009  All rights reserved.