BEYOND CLASSES OF SERVICE – A USER GUIDE
Contents
1. O CoS based on the DiffServ model is in theory simple for a service provider to implement inside the network It requires the implementation of local priority rules in the routers As such it doesn t require a lengthy and costly overhaul of the existing network infrastructures 2 2 Limitations On the other hand CoS has some important limitations CoS is focused around infrastructure behavior and does not address end to end user needs CoS is a best effort based approach to QoS and cannot guarantee the performance of individual critical application flows Overall CoS exacerbates the chronic problem of network over sizing The proper implementation of CoS is complex and time consuming for end user organizations CoS doesn t fit easily into heterogeneous architectures CoS leads to rigid implementations CoS can be expensive Focused on infrastructure behavior and does not address end to end user needs A common problem is that there is a difference between a customer expectation s that CoS will improve application performance and the fact that actual performance depends on both infrastructure performance and the way it is actually used It is interesting to note that the majority of poor performance helpdesk calls are due to the customer overloading the network rather than an actual network problem SLAs currently offered only deal with statistical measurement of the infrastructure quality and never with real end to end traffic2. ipanem lechnologieas E mail info ipanematech com http www ipanematech com Business Optimized Networks WHITE PAPER BEYOND CLASSES OF SERVICE A USER GUIDE EXECUTIVE SUMMARY In the past WANs were used for a small number of information and business applications financial accounting supply chain etc now however they are available to every level of the enterprise The flexibility of IP protocols allows the network to be shared by numerous applications Each of these applications impacts the way a company is run and they require very differing performance levels transactional traffic is impacted by average delay large file transfers are bandwidth hungry while voice traffic is very sensitive to jitter This increasing number of different uses leads to greater network requirements along with the need to align network resources with applications requirements Faced with the successful deployment of private networks IP VPNs and the need to differentiate the performance of different applications Service Operators are offering their clients Classes of Service CoS over MPLS networks Even though CoS is a step in the right direction it is often not able to respond to all the issues associated with multiple application deployments on the same network It has some natural inflexibilities and limitations Ipanema has gone beyond the limitations of CoS and has created the Business Network Optimization solut
3. Step 1 Defining application performance objectives Whereas CoS requires modeling of each application with an allocation of bandwidth required per class and per network access the Ipanema System only deals with performance objectives at the user level The client organization will have to determine the business criticality of the application flow and the application performance expected Business criticality is voluntarily limited to four levels and it is straightforward Top High Medium and Low Expected performance is expressed in terms of minimum required bandwidth jitter delay and loss Both expected performance and business criticality can be expressed at the application level e g SAP critical for all users and at the user level e g E mail non critical for the organization except for traders In most cases 20 to 30 parameters suffice to pilot a network with tens of applications hundreds of sites and thousands of users Figure 3 shows how these parameters are set APPLICATION DEFINITION PER USER SERVICE LEVEL DEFINITION Per metric Objective Maximum APPLICATION Criticality TYPE BW kpbs DELAY JITTER LOSS ms ms SAP TOP 50 n a 1 3 Traffic classification is EUREO 100 300 based on layer 3 to layer 7 criteria CITRIX TOP Transac 20 100 300 n a VolP G729 HIGH Real Time 11 50 150 40 80 ORACLE HIGH EUREO 20 100 300 n a Criticality determines LDAP MEDIUM Data 20 200 1000 n a which application
4. traffic should be Web MEDIUM Data 20 200 1000 n a protected when congestion occurs Otner MEDIUM Data 10 200 1000 n a PIR LOW Data Trans 25 n a n a NETBIOS LOW DEIEEICUS 50 n a n a EMAIL LOW Data Trans 25 n a n a The raffic Type determines the most important metrics for optimization Figure 3 Setting the parameters ipanema Business Optimized Networks O Step 2 The System Autonomically optimizes network resources The Ipanema System s optimization is objective driven global and combines multiple mechanisms Objective driven lpanema s optimization is not based on static policies based on a snapshot of network conditions Instead the Ipanema System constantly adapts to network conditions to meet application performance objectives These objectives as mentioned earlier specify the minimum network requirements for each end user to get the right quality of experience In addition the optimization mechanism uses the notion of business criticality to protect critical applications in case of congestion Global Whereas CoS applies priority rules locally Ipanema optimizes performances on a network wide basis In real time it manages all types of congestion locally from the source at the destination due to conflicts between traffic coming from several sources and even network hosted congestion This guarantees that performance objectives are met even in complex networks with a complex traffic matri
5. of up to 400 and finally by affecting end user behavior It is not unusual that due to the Ipanema System clients can sustain increases of 400 of traffic over several years without requiring any bandwidth increase while guaranteeing application performance For more on the financial impact of the Ipanema System refer to the ROI White paper ipanema 12 Business Optimized Networks O SUMMARY The table below summarizes the main differences between traditional router based CoS and optimization of network resources using the Ipanema System Management of application performance objectives Application performance guarantee Handling of individual user application performance SLA Ease of implementation Flexibility Heterogeneous network ADSL multi Telco s etc Selection of class of service Change Management Over provisioning Network costs Impact on core network Business Optimized Networks Policy driven applications are grouped in classes No best effort basis No aggregation of all flows from the same class Network centric Low Low Difficult to impossible to implement At each access site simplified classification Difficult Increased Increased Limited Objective driven per application at user experience level Yes for business critical applications Yes per user session User amp application centric High High Not an issue Centrally network wide detailed classi
6. e underlying network or networks Multiples heterogeneous networks are no barrier to the System s efficiency Allows rightsizing of the network Whereas the local and static traffic model of router based CoS requires much over sizing to manage uncertain traffic situations Ipanema s dynamic and global optimization always makes best use of available bandwidth The need for over provisioning is completely eliminated as high critical real time or transactional flows are individually protected not only from lower criticalities but from bandwidth hungry flows In addition the rightsizing feature provides detailed information about the optimum network required for different levels of application performance taking guesswork out of capacity planning Reduces network budgets Unlike CoS the Ipanema System reduces network related costs The System has both indirect and direct financial impacts It contributes to increasing the overall organization s productivity by eliminating brown outs ipanema 11 Business Optimized Networks to lowering network incident management costs by reducing the number of incidents and the time to repair and to decreasing network change management costs But most importantly the Ipanema System allows the reduction and or control of telecom costs It does so by eliminating the need for bandwidth over provisioning by reducing bandwidth requirements for applications thanks to optimization factors
7. fication Seamless Eliminated Decreased None ipanema 13 O IPANEMA AND CoS Based on the merits and limitations of CoS and Ipanema s Business Optimized Network solution what should clients do Choose just one approach or adapt both Can they co exist on a network The Ipanema System and CoS can co exist on a network The Ipanema System is compatible with the CoS and it can even mark the packets for their class It is our recommendation that clients take limited CoS for instance two classes one for real time traffic VoIP Video and one for all other types of traffic in conjunction with the Ipanema System By doing so they will have the best of both worlds They will protect themselves from eventual shortcomings of their service providers core networks while benefiting from the dynamic objective driven optimization of the Ipanema System Multiplying the number of Classes of Service is on the other hand complex counter productive and expensive The greater the number of Classes of Service the higher the complexity of implementation Dividing bandwidth into several sub bandwidths reduces the amplitude of bandwidth sharing Beyond a certain number of classes this phenomenon offsets the benefits of CoS unless all classes are thoroughly over provisioned The higher the number of classes the higher the overall bandwidth over provisioning and the resulting network costs ipanema 14 Business Optimized Networks ipane
8. ical applications as requested by the end user While CoS is a best effort infrastructure centric approach to QoS the Ipanema System is a user centric guarantee Easy to implement and manage change seamlessly Being objective driven Ipanema s performance optimization is extraordinarily simple to manage The most complex network needs no more than 30 parameters to be managed Unlike router based CoS there is no need to determine configure and implement local traffic policies Thus rolling out a new application or adapting to changes in network configuration are all dealt with just a couple of clicks of a mouse Fully flexible and adaptive to a variety of traffic situations Again the objective driven nature of the Ipanema System and its optimization mechanisms allow it to constantly and instantaneously adapt to changes in traffic situations There is no need to re configure new policies if new critical applications have been added or if the traffic behavior from some specific sites has been permanently or temporarily changed The System will automatically adapt to all these changes Quite a difference from the rigid rules based approach of CoS Can be deployed over heterogeneous networks Unlike CoS the Ipanema System doesn t rely on changes in routers or Service Provider specific engineering rules By using its own devices located at the edge of the network the Ipanema System can optimize the use of network resources whatever th
9. ion allowing you to align application performance requirements with the enterprise s network resources in a more specific flexible and dynamic way ipanema 1 O h THE NEED TO MAXIMIZE APPLICATION PERFORMANCE Several factors explain why now more than ever maximizing application performance over the network by optimizing network resources is a critical task 1 The network has become the critical infrastructure 2 There are a wide variety of application behaviors 3 The cost of poor quality is an issue 1 The network has become the critical infrastructure There are an ever increasing number of applications running over Wide Area Networks Improved Web based human interfaces require larger data transfers and more turns through the network Performance sensitive applications such as Citrix or VoIP are on the rise It is estimated that two thirds of application response time is now driven by the network 1 1 Lower employee productivity Poor application response time has a negative impact on result in irate customers abandoning baskets full of goods At bank a study showed that there was a direct correlation between banking applications response time and the length of the customer queues at the teller Long queues result in unhappy customers that are open to the competition 1 2 Lower employee productivity Poor application response time has a negative impact on employee s morale attitude and productivity A critical applicat
10. ion response time beyond 2 seconds generates wasted user time that can translate into a 30 productivity loss As the delay exceeds 4 seconds users start complaining to the helpdesk thus generating further costs If the delay exceeds 8 seconds users may stop using the application or work overtime outside of peak hours 2 There are a wide variety of application behaviors The network is carrying an ever increasing number of applications with different behaviors and needs Transactional SAP and thin client Citrix applications often critical for an organization exchange little data but are very sensitive to transfer delays On the other hand real time flows such as VoIP require low jitter Finally file transfers e mail NetBios ftp while usually less critical are not delay sensitive but typically require large amounts of bandwidth TCP the standard protocol on IP networks does not distinguish between applications It allows each flow to take as much available bandwidth as possible A few bandwidth hungry flows e mail with attachments print files can therefore upset a large number of transactional flows that require less ipanema Business Optimized Networks O bandwidth but whose performance is more critical for the organization Critical applications from a business standpoint are not protected 3 The cost of poor quality is an issue Poor network quality and application performance represent a high cost for organ
11. izations Network downtime blackout or even more so non performance brownout leads to losses in revenue and productivity For instance according to a 2004 survey from the Standish Group the cost per minute of downtime for an ERP system for a Fortune 1000 company is 13 000 Cost per Minute Messaging E 1 000 Electronics Funds Transfer D 3 500 Customer Service Centre E 3 700 Internet Banking EEEE 7 000 CRM P 10 000 SCM H 11 000 ERP p 3 000 0 000 2 000 4 000 6 000 8 000 10 000 12 000 14 000 Figure 1 Average per Minute Cost of Downtime Source Standish Group 2004 Study was based upon 250 Fortune 1000 companies with an average of 20 000 users ipanema Business Optimized Networks O eh CLASSES OF SERVICE 1 Definition 2 CoS benefits and limitations 1 Definition Service Providers recognize the need to differentiate the network resources offered to different applications they created Classes of Services or CoS to facilitate this The objective of CoS is to adapt the infrastructures behavior to flow types It is based on the DiffServ model proposed by the IETF Internet Engineering Task Force the Internet standardization body and it uses the following mechanisms Flows are differentiated into a few Classes of Service generally 3 to 6 called Gold Silver Bronze etc The amount of traffic in higher classes is usually limited for instance no more
12. ma Technologies http www ipanematech com Worldwide Headquarters Ipanema Technologies 28 rue de la Redoute 92260 Fontenay aux Roses FRANCE Phone 33 155 52 15 00 Fax 33 155 52 15 01 E mail info fr ipanematech com United States Ipanema Technologies Corp 300 Fifth Avenue 3rd Floor Waltham MA 02451 Phone 1 781 419 6526 Fax 1 781 419 6527 E mail info us ipanematech com Germany Ipanema Technologies GmbH Gustav Stresemann Ring 1 65189 Wiesbaden GERMANY Phone 49 611 97774 285 Fax 49 611 97774 111 E mail info de ipanematech com UK Ipanema Technologies Ltd Abbey House Wellington Way Weybridge Surrey KT13 OTT UNITED KINGDOM Phone 44 0 1932 268 380 Fax 44 0 1932 268 381 E mail info uk ipanematech com Copyright 2007 Ipanema Technologies All rights reserved Ipanema and the Ipanema logo are trademarks of Ipanema Technologies Trademarks and trade names mentioned herein refer to entities claiming those marks and names or to their products Ipanema Technologies disclaims any proprietary interest in any trademarks and trade names other than its own lis z ipanema 15 Business Optimized Networks
13. n sufficient for a 100 SLA for critical applications however 1500 kbps would have been necessary if all applications were to achieve their performance objectives 100 of the time 2 Ipanema System benefits and limitations 2 1 Limitations The main limitation of the Ipanema System is that it does not intervene in the core network It takes the core network conditions as a given Therefore if performance issues are driven by shortcomings in the core the Ipanema System will not be able to improve the situation Having said that it must be noted that in most cases core networks are designed to handle the aggregate traffic from numerous customers without congestion Because of their high bandwidth their contribution to overall traffic delay is generally small POP to POP delays inside a continent will generally be guaranteed to be a few tens of milliseconds with very few losses The main source of congestion loss and delay comes from the big imbalances between 1 LAN and access network bandwidth at the source 2 Core and access network bandwidth at the destination ipanema 10 Business Optimized Networks O 2 2 Benefits Aligns networks resources with business criticality of applications and guarantees their performance from an end user perspective Unlike CoS which allocates traffic flows per group of applications the Ipanema System shapes and protects flows per session As such it can guarantee the performance of crit
14. omplexity of any modifications CoS has very poor flexibility Indeed rolling out a new application or changing the CoS of an existing application will also require changes in many rules a task that is tedious long and error prone As such there is a natural tendency for client organizations and Service Providers alike to delay or ignore the changes necessary for optimal use of the network This is unfortunate as networks are living bodies with constant changes in terms of user behavior new application rollout etc Can be expensive Service providers charge CoS at a premium because it offers the customer a new service with new SLAs it requires more powerful network equipment it increases the Service Provider s management burden it usually increases the network size both for access lines and in the core a Customer organizations should expect an increase of between 10 and 20 of their network budget In addition any changes to CoS rules are charged to the Customer ipanema Business Optimized Networks D IPANEMA S BUSINESS OPTIMIZED NETWORKS The Ipanema System is a network optimization system that in essence provides a real time and flexible Class of Service per user session 1 How does it do that 2 Ipanema System benefits and limitations 1 How does it do that The Ipanema System is made up of a central management software platform physical and virtual devices Depending on network configurati
15. on a combination of physical and virtual devices are deployed at the edges of the network in the relevant branch offices and data centers The network devices and the central management software form a distributed Autonomic Networking System that delivers the visibility optimization and rightsizing features Central software ip boss Fa Branch office Distributed devices _ Regional H Q Figure 2 The Ipanema System The essence of the Ipanema System is straightforward and can be described in three steps Firstly a user ranks applications and user groups subnets per business criticality from high to low then they determine the performance objectives expected for each application and user group Secondly the System automatically and autonomically optimizes network resources to match the performance required It guarantees application performance for critical applications whatever the ipanema Business Optimized Networks network conditions Thirdly the System provides real time information about application performance over the network If non critical applications do not reach their performance objectives 100 of the time it provides precisely the amount of bandwidth required to achieve different levels of performance In other words it allows accurate rightsizing of the network according to application performance objectives or application SLA Let s have a more comprehensive look at each step
16. performance Measurements are on parts of the network only like POP to POP and are generated from router statistics or by test traffic from shadow routers Best effort basis with no guarantee With CoS traffic is classified by group of applications aggregate behavior but no differentiation is made by user session consequently inside a class the standard TCP rules apply and all flows are handled in the same way As such if there are several applications in one class and several users using these applications simultaneously congestion and deterioration of application performance can still be experienced despite CoS This explain why the CoS approach to QoS is best effort based it cannot guarantee the performance of individual application flows Network over sizing As seen previously CoS is similar to splitting the network into several networks To minimize the probability of performance deterioration in higher classes there is a tendency to over provision them As this behavior gets replicated over several classes CoS exacerbates the chronic problem of network over sizing In addition by splitting the network into several networks CoS automatically reduces the infrastructures sharing capabilities Overall CoS will lead to a less efficient use of bandwidth Complex implementation Rules for priority management of classes have to be implemented in all network routers This includes measures such as traffic limiting for higher clas
17. ses to protect the quality of delivered service against customer overloads As bandwidth allocated to each class may vary from customer to customer and over time for a given customer entering and maintaining these rules is a heavy management burden for the service providers ipanema Business Optimized Networks O However the biggest burden is on the customer organization that needs to predict the traffic for each class This is much more complex to do than for access as a whole as the customer needs to define and maintain the following per access point gt the allocation of traffic flows to the various classes the corresponding sizing of the classes Doesn t fit easily into heterogeneous architectures The implementation of CoS depends on each telcos unique engineering rules As such heterogeneous architectures requiring multiple peering points are a challenge for Cos implementations Indeed each peering point will require adaptation and mapping of CoS rules number of classes limits and re coloring rules etc In addition sizing at peering interfaces is hard to define For these reasons low cost access networks relying on ADSL accesses may not be CoS compatible Accesses may rely on the incumbent service provider who cannot or will not implement other Service Providers CoS rules Leads to rigid implementation The complexity of implementation required by CoS introduces a certain level of rigidity Due to the risk and c
18. than 20 of packets can be Gold A class overflow may result in packets being discarded or relegated to a lower class E The class selection is marked in a field in each packet DSCP or TOS field The marking is often done by a router at the customer site or close to it within the network All or some routers in the network will then implement local priority rules based on the value of this field basically a packet from a higher class will be routed prior to a packet in a lower class It is equivalent to partitioning the initial single class network into 3 to 6 individual networks with different precedence and performance A higher class may sometimes borrow bandwidth from others dependant on the specific telcos implementation 2 CoS benefits and limitations 2 1 Benefits CoS addresses the issue of undifferentiated treatment of application flows By assigning some flows to a higher class the traffic will spend less time in congested router queues and will have a minimum amount of dedicated bandwidth As a result the higher class traffic will have lower transfer delays and lower packet losses Real time traffic like VoIP along with delay and loss sensitive protocols like SNA over IP will benefit from having the lowest possible delay within network routers Critical transactional applications may also benefit from a higher class than that of non transactional bandwidth hungry applications ipanema Business Optimized Networks
19. width objectives while other flows will be allocated a share of the remaining bandwidth according to their relative business criticality Smart Packet Forwarding While the DBA allocates bandwidth according to business criticality and network conditions the Smart Packet Forwarding SPF feature decides in real time within allocated bandwidth and dependant on the nature of the flow i e real time transactional data transfer which packet to forward first Thus the constraints of application delay and jitter sensitivity are protected Advanced Compression Ipanema s Advanced Compression is based on intra packet and inter packet redundancy elimination By reducing the bandwidth used by the application flows it creates additional bandwidth and increases a link speed Configuration of the compression tunnels is fully automated even in meshed and redundant environments Furthermore this feature is fully compatible with MPLS mechanisms While compression is very useful and powerful on specific links it is rarely required across the whole network Step 3 Rightsizing according to application performance objectives Unlike CoS which leads to over provisioning with the Ipanema System links can be rightsized Indeed the System provides information that directly links bandwidth to application performance Clients can therefore make informed trade offs between cost and performance In the following example 768Kbps would have bee
20. x This is a mandatory requirement when several data centers are running simultaneously or with meshed traffic VoIP Non critical packets are delayed at the source to avoid congestion at destination Critical flow 2 Mbits s Non critical flow Figure 4 Meshed WAN traffic requires global optimization Combining multiple mechanisms The Ipanema System uses three optimization mechanisms that work together Dynamic Bandwidth Allocation DBA Smart Packet Forwarding SPF and Advanced Compression By doing so it assures that the network always delivers the best possible application performance according to the business objectives I panema s Advanced Compression creates additional bandwidth and increases the link s speed The Dynamic Bandwidth Allocation DBA mechanism insures that the additional resources are used by business critical applications instead of recreational ones And the Smart Packet Forwarding SPF features protect application delays and jitter ipanema Business Optimized Networks O Dynamic Bandwidth Allocation Whereas CoS allocates static bandwidth per access to each of the classes Ipanema allocates bandwidth dynamically to each active user session according to the objectives defined above the real time measurement of available bandwidth between all access points the real time tracking of the traffic demand During congestion Ipanema ensures that each critical flow gets its band
Download Pdf Manuals
Related Search