Bandwidth and Application Management

Service Provider Evolution

In this era of e-commerce and e-businesses, many companies rely on

distributed network services as a significant part of their entire business

model. As a result, Internet-related initiatives require recruiting, staffing

and funding to operate. Typically, these expenses represent a significant cost

and not one that decreases with time. Service providers address these concerns

by creating economies of scale. By serving numerous clients, they reduce costs

for everyone. In fact, if they manage bandwidth and service usage properly,

service providers can provide better customer service than internal providers.

However, service providers need to manage the following issues:

• Bandwidth hogging that decreases
  
  access to network resources

• Maximizing potential business
  
  applications

• Ensuring high performance across
  
  applications

• Customizing bandwidth to expand
  
  services

In order to manage these issues, service providers must have an accurate map

of their network. They must monitor traffic to determine normal and abnormal

usage and establish baseline metrics to be able to efficiently monitor

productivity.

Factors to consider

• Burst traffic: Burst traffic is
  
  abnormal usage that typically lasts only for a short time during peak
  
  periods, for example, when a popular, downloadable file goes online.
  
  Typically, burstable traffic consists of lower priority requests that may
  
  receive decreased bandwidth.

• Interactive traffic: Interactive
  
  traffic typically refers to normal use of a web site, such as users
  
  following links. If it is high-priority traffic, it must be fast. Other
  
  application-level clients require a minimal amount of bandwidth to provide
  
  mission-critical services.

• Non-mission critical traffic:
  
  Employees behind a router may be using network resources for personal
  
  reasons. Such traffic may usually be assigned lower bandwidth and priority.

• Mission-critical traffic: With
  
  mission-critical traffic, incoming orders for example, must be given a high
  
  priority, if not the highest priority, and a substantial amount of available
  
  bandwidth.

• SLAs: In some cases, service
  
  providers may have Service Level Agreements (SLAs) in place with clients,
  
  and are contractually bound to provide a set amount of dedicated bandwidth
  
  and server resources. Service providers must monitor this traffic to meet
  
  contractual obligations.

Bandwidth Management

This process includes classifying traffic, managing bandwidth allocation and

mapping traffic classes.

Classifying traffic

There are a number of ways to classify traffic—server names, network

subnets, destination IP and port, source IP and port, requested file type—the

list goes on. However, propagating a client request to a fulfillment server

should be automatic and transparent to the requestor. Classifying traffic

enables network managers to better track requests and allocate resources based

on priorities. Is an FTP request from accounting more important than an HTTP

request from human resources? This functionality also enables service providers

to establish priorities based on application usage and the nature of request as

well as any other metric they want to use.

Managing bandwidth allocation

Once traffic is classified, specific configurations can be defined to control

how bandwidth is distributed. The two most widely used methods are partitions

and policies.

• Partitions: Partitioning
  
  creates a separate, exclusive channel for traffic that manages the total
  
  network usage by traffic type, for example, FTP traffic might be assigned 10
  
  percent of network resources, e-mail might receive 10 percent and incoming
  
  HTTP requests might receive 50 percent. Unused bandwidth in that traffic
  
  type can be placed in a pool that is available to other applications to
  
  speed up the overall network.

• Policies: Rate-based
  
  policies set a minimum amount of guaranteed bandwidth for burst traffic.
  
  Priority policies set aside bandwidth for traffic that must compete with
  
  burst traffic. These policies provide network stability and efficient
  
  fulfillment of all requests and responses.

Application Management

As network infrastructure evolve, administrators have started implementing

bandwidth management solutions. Now that service providers are providing still

more mission-critical services, they need a new set of tools to help them make

the best use of their existing infrastructure to provide exceptional service to

their customers.

There are several tools that can help.

• SLAs vs. ASLAs: SLAs
  
  commit to specified uptimes or times between failures. However, these
  
  traditional measurements are no longer adequate when service providers enter
  
  the application marketplace. Application Service Level Agreements (ASLAs)
  
  are precise, per application, measurable agreements, specifying the nature
  
  and quality of application deliverables. They cover per application response
  
  times, availability and other metrics, as well as the types of reports
  
  describing application performance.

• Delivery models: ASLA
  
  content and management strategy depends on a service provider’s delivery
  
  model. This can be extranet or intranet-based and is specified in the ASLA.

• Extranet model: For many
  
  service providers, the fastest and most effective way to deliver application
  
  services is by utilizing the client’s existing WAN that connects end-users
  
  to a centralized data center. The service provider provides an extranet
  
  connection to this data center to relay hosted applications and content.
  
  Clients control performance within their own networks.

• Intranet model: Some
  
  service providers combine managed network and application services to
  
  deliver a comprehensive solution. In addition to delivering hosted
  
  applications and data to the client’s data center, the service provider
  
  takes responsibility for the end-to-end quality of the user experience.
  
  Since applications flow across multiple networks, including delivery-chain
  
  providers and the client before reaching the end-user, it is often difficult
  
  to determine where and when problems arise. Support costs can spiral out of
  
  control if service providers spend too much time troubleshooting problems
  
  that fall outside their responsibility, so they must carefully define
  
  service boundaries which lie beyond their control.

Typically, service providers should set a boundary where

their applications enter the data center. Additional boundaries may be necessary

for other network services.

Performance Analysis

Both the service provider and the client should validate

deliverables for different reasons. The client wants to confirm that they got

what they paid for. The service provider needs a comprehensive analysis

including detailed per-application metrics. The optimum solution should offer

the flexibility to assist multiple parties and include:

• High-level graphics and reports
  
  for clients and service providers to compare committed versus actual
  
  performance

• Remote access that enables remote
  
  validation such as browser-based user interface

• Detailed graphs and reports for
  
  diagnostic and planning purposes

• Metrics to import into
  
  third-party reporting applications

• FHSS systems are primarily used
  
  for low-power, low-range applications, such as 2.4GHz cordless phones, and
  
  do not inter-operate with DSSS products.

Deployment and Integration

The solution should not change users’ desktops, servers,

routers or applications. Installation should be straightforward, quick,

compatible and integrated with existing infrastructures. A third-party solution

should not become an obstacle to fulfillment.

Enforcement Examples

Service providers should make sure they meet their

performance obligations. The key to enforcing performance is to differentiate

each application needing special treatment, and then appropriately and precisely

assign resources.

Example #1

A service provider hosts Microsoft Exchange and a Great

Plains accounting application running over Citrix MetaFrame. The benefits of

e-mail, scheduling, information sharing and public folders can consume

additional bandwidth and affect the other applications. The accounting

application is business critical and the service provider needs to deliver

consistent, prompt performance. The control requirements differ for the two

delivery models:

• The Extranet Model: This
  
  model requires the service provider to balance only Microsoft Exchange and
  
  Great Plains performance and deliver it to the client’s data center.

• The Intranet Model: This
  
  model requires the same, plus control of performance for all subscriber
  
  applications running over the enterprise WAN, all the way to the end user.

Example #2

A service provider adds a streaming media application, such

as Voice over IP (VoIP), to the hosted Repertoire. The service provider defines:

• A per-session bandwidth minimum
  
  of 18 Kbps for each session to secure good reception for all. If any
  
  minimums are not used, they are available as excess rate.

• A polite block to new sessions if
  
  VoIP is overextended. Late users receive a polite message if they seek
  
  service when earlier conversations occupy all bandwidth reserved for VoIP.

Example #3

A service provider expands to incorporate managed network

services, stepping up to the Intranet Model to manage application performance

across the subscriber’s WAN. In this case, the service provider defines:

• A per-session cap on applications
  
  that use an inappropriate amount of bandwidth. Web browsing, news and large
  
  file transfers are typical candidates for caps.

• Per-application bandwidth
  
  minimums and caps for the client’s own, non-outsourced applications.

• Efficiency methods such as
  
  slowing the sender to suit a slow recipient.

GB Kumar, general manager, business programs, Intel India