Traffic Controller Service. UltraDNS Whitepaper

Transcription

1 Traffic Controller Service UltraDNS Whitepaper

2 Table of Contents Global load balancing challenges... 3 Overview... 4 Architecture... 5 Dynamic monitoring system... 6 Traffic Controller techniques... 7 Example... 9 Internal vs external load balancing...10 Management and administration...10 Benefits...11 Interaction case study: F5 Networks...12 Specifications summary...16 Copyright NeuStar, Inc. 2

3 Global load-balancing challenges Maintainability, availability, and disaster recovery planning are critical elements in today s Internet-driven industries. Progressive ebusiness companies must have IT infrastructures and business processes to assure IT systems maintain 100% availability, as downtime even in the order of minutes can mean the loss of tens of thousands in revenue dollars. These concerns are well understood in today s advanced IT operations and successful companies continuously strive to address these needs. This holds true for customer-facing applications and infrastructure as well as internal applications within an enterprise. One of the most important elements of an effective ebusiness infrastructure is global redundant web content distribution, which must include a reliable and scalable DNS infrastructure, to bring users to the web servers, and Global Server Load Balancing (GSLB) solutions to distribute and shape query responses with different geographically distributed data centers. Global Server Load Balancing solutions have been an attractive tool in an ebusiness infrastructure because they solve a number of problems: GSLB can be used as a disaster recovery tool, allowing automatic failover to alternative content servers in other data centers should there be an operational or technical failure, or a catastrophic event. GSLB provides the ability to route users to content servers that are either geographically or (network) topologically close by. GSLB can improve performance of the existing servers resources by preventing busy servers from overloading and increasing traffic to less busy servers, thus maximizing overall utilization of the content network. Although the benefits of robust load balancing systems are easy to justify, the IT infrastructure equipment needed for it is costly to procure, deploy, and maintain. Today s solutions to this problem include a variety of dedicated hardware/software solutions: public and private DNS name servers; local load balancers; http redirection systems, and higher-layer packet switching systems. The complexity of these hardware/software systems in some cases becomes burdensome as tangible and intangible costs are incurred, such as: Copyright NeuStar, Inc. 3

4 Capital expenditures to procure and replace the required equipment over a network s life cycle Costs associated with training new staff on the systems Software upgrade costs Annual licenses and maintenance contract costs Extended downtime due to troubleshooting during outages due to the complexity of the environment and its dependencies Many progressive IT administrators will investigate strategic outsourcing solutions for their infrastructure where it is practical and cost effective to meet business needs. GSLB solutions are provided by some Internet Service Providers (ISPs) and Content Delivery Network (CDN) providers, but are often bundled with other services. In this case the administrator has to make a compromise decision, possibly forcing a non-optimal outsourcing solution, or continuing to keep control of the GSLB capability in-house. NeuStar UltraDNS Traffic Controller Service, a robust Global Server Load Balancing managed service solution, builds on the NeuStar UltraDNS Managed DNS, SiteBacker Monitoring and Failover, and load-balancing intelligence to provide a unique managed service solution to an enterprise s GSBL needs. Overview NeuStar UltraDNS Services provide customers with a set of network tools allowing administrators to define load balancing configurations for content servers residing in one or more geographic locations. Its DNS Traffic Management suite of managed services traffic directs servers/data centers by dynamically changing the responses to DNS requests. This suite of services now includes: UltraDNS SiteBacker Service Integrated Monitoring, Notification, & Failover service. Copyright NeuStar, Inc. 4

5 UltraDNS Directional DNS Service Permits node specific DNS answers based on topology/geography as well as the advanced intelligence of Geo IP targeting. UltraDNS Traffic Controller Service - Performs load balancing based on dynamic metrics obtained by continually monitoring the host servers. The Traffic Controller Service is a robust Global Server Load Balancing managed service solution providing an alternative to the costly hardware-centric solutions commonly deployed in IT infrastructures today. IT administrators are able to define load balancing configurations for content servers residing in one or more geographic locations. Traffic Controller leverages the proven capabilities of existing NeuStar UltraDNS services to provide a unique managed service solution to an enterprise s load balancing needs. Architecture The Traffic Controller Service is integrated into the Managed DNS solution on the UltraDNS Services global production network. Using logic built into our proprietary resolver code and specialized fields within our Oracle database schema DNS answers are able to be balanced across multiple resources. DNS-based load balancing is performed based on either database data or dynamic feedback from probes which are monitoring configured content servers. The dynamic tables used by DNS Traffic Management are created by using the monitoring probes which gather metrics for DNS-based failover and notifications services. The probes are able to measure the responsiveness of designated web/application servers (or the load balancers in front of said servers). Based on the probe results, the Traffic Controller Service then modifies the load factors in DNS in real-time to redirect traffic to the more responsive servers. The Traffic Service can also modify DNS to deactivate sites that have failed, as is performed with the current SiteBacker service. The probes run from multiple agents within the UltraDNS Services Directory Infrastructure Copyright NeuStar, Inc. 5

6 Platform. Users of the UltraDNS Managed Internal DNS solution can also configure agents on the Directory Gateways to create additional monitoring points for both external and internal resources from within the corporate network. Dynamic monitoring system The Traffic Controller Service utilizes the same global system of dynamic monitoring probes that supports the UltraDNS Sitebacker Service, which has provided monitoring and failover service to thousands of services across the Internet for over five years. The probes consist of single or multi-step HTTP scripts with cookie management. This allows probes to effectively test formsbased processes that may fail while the websites themselves respond normally to port or HTTP testing. In addition to HTTP/HTTPS, probes can be configured to utilize other protocols such as SMTP, DNS, FTP, Proxy, Ping, and basic TCP tests to check the state of a resource. The inherent flexibility of the UltraDNS Services combined with the knowledge and skill of our Professional Services Group allows us to also build and deploy customized probe types at the request of our clients. The actions that the dynamic monitoring systems perform are as follows: Agents run probes against resources in the pool and measures the response time of a given resource. The monitoring system uses a true/false test result as well as specific averages from a number of probes with preconfigured dampening factors to determine effective server performance. The average used includes probes from various specific monitoring locations within the UltraDNS Services global production network, or from all locations. (NOTE: If the Traffic Controller Service is being used internal in a customer s network, the probes will run from Directory Gateways configured with internal probes.) Complete unresponsiveness to a probe test or response times exceeding a predefined critical threshold cause the service to stop answering with that resource for a given hostname. Copyright NeuStar, Inc. 6

7 The service continues to probe the failed resource and restores it to the server pool when it becomes available again. The service may also be set to provide notifications when the status of a server changes, whether it s a failure or a return to normal state. Traffic Controller techniques DNS-based round robin is currently provided via NeuStar UltraDNS Managed DNS service. The new Traffic Controller functionality builds on this basic principle as well as utilization of the dynamic monitoring system to provide a more intelligent DNS-based load balancing solution combining, monitored failover and weighted (percentage-based) load balancing. The following is a high level process flow for the DNS-based load balancing functionality of the new Traffic Controller Service: An end user makes a DNS request for which is part of a load balanced pool included in Traffic Controller The DNS request is routed to the topologically closest NeuStar UltraDNS resolver The Traffic Controller resolves the DNS query from a load balancing pool based on logic configured for the record Behind the scenes, Traffic Controller keeps the load balancing pool upto-date with real time information about load balancing resources within the pool. Traffic Controller allows users to set intermediate performance response thresholds Warning and Critical in addition to Failure threshold. The Warning threshold allows the Traffic Controller to send a notification when server responsiveness reaches a certain level but is not Critical or Failed. The Critical threshold also sends a notification but defines when Copyright NeuStar, Inc. 7

8 S e r v e r S erv er S erv er P BX P BX M inic om put er Mini computer S erv er P BX Server Mi nicom put er PB X M inicom put er T r a f f i c C o n t r o l l e r S e r v i c e Traffic Controller reduces the weighted value of a degraded resource by half, effectively reducing traffic to it while at the same time increasing traffic to the other servers in the pool. The reduction of the assigned weight is done in an attempt to bring it back to a normal performance state without completely failing the server. Should the server return to performance responses below the Critical threshold, the original weighted value will be restored and another notification sent. If the performance continues to degrade and reaches the Failure threshold then notification is sent and the server is completely removed from availability. With the new Traffic Controller service, the weights are applied to the authoritative queries only. (NOTE: Since recursive server caches are not factored into these weights, actual traffic percentages when measured, may vary from the calculated weightings.) Even with recursive server caching as a factor, customers should find this approach preferable as it is currently done in other products, is easy to deploy, and performance metrics validity are acceptable when lower TTLs are used for the DNS zone records. Traffic directed based on probe results of server availability and performance metrics NeuStar UltraDNS Global Network Performance Probe Agent Recursive Server ISP 40% 60% Performance Probes End Users Probes DNS Query DNS Response IP Traffic Local Load Balancer Content Servers Data Center 1 Site A Data Center 2 Site B Copyright NeuStar, Inc. 8

9 Example Weighted DNS distribution takes place in the following manner: Customer configures a load balancing pool for with the following load balancing resources: o Host o Host o Host Each host maintains the load balanced resource ( Customer assigns a weight or percentage to each host (allocating amount of authoritative DNS responses to each host) o Host = 12= 50% o Host = 6 = 25% o Host = 6 = 25% Based on the previous example, Host1 would be served as the DNS resolution twice as often as Host2 or Host3 In addition to weighting, the dynamic monitoring system ensures resources reaching critical thresholds or that become unavailable are pulled from the load balancing pool o If Host1 is pulled from the pool, Host2 and Host3 would receive all traffic equally (essentially in a round robin format) o If Host1 reaches its Critical threshold, its weight would be reduced to 6, and all three servers would be announced equally to about 1/3 of the queries each. NeuStar UltraDNS services also provide the ability to change DNS responses based on the UltraDNS node at which the query hits our network or geographic location of the IP Copyright NeuStar, Inc. 9

10 address making the request (recursive server) through the Directional DNS Service. Internal vs external load balancing NeuStar is committed to providing load balancing services for customers on the public Internet as well as internal to individual customer environments. Load balancing features which are available for external services are also planned for availability internally. Just as customers must use the Managed External DNS service to use Traffic Controller externally, its use internally requires customers to utilize the NeuStar UltraDNS Managed Internal DNS Service to take advantage of Traffic Controller internally. The probes can be forwarded from the external environment to internal resources (assuming customer makes appropriate network modifications to allow the probe to contact internal systems) or run from the Directory Gateway Appliances used to provide local services with Managed Internal DNS. Management and administration The Traffic Controller Service is integrated with the UltraDNS Management Portal, providing the administrator with a simple and fully integrated tool for configuring his or her load balancing schemas. The Management Portal provides: Common, integrated management portal to both Managed DNS (external) services, Internal DNS services, as well traffic management services Role-based access control to support delegated responsibility Provides syntax-based error-checking for input records Powerful search and mass editing features For Traffic Controller, administrators are able to define load balancing techniques via the Management Portal. Administrators can deploy load balancing configurations down to the record level. Default settings can also be configurable to ease change management.. Copyright NeuStar, Inc. 10

11 Benefits The Traffic Controller Service provides many benefits to IT administrators and their organizations seeking a cost effective GSLB solution: Unique Managed Service delivery for Global Server Load Balancing and Traffic Management No equipment or infrastructure requirements specific to the service results in low up-front costs with simple and fast deployments (Note: if Traffic Controller is used by customers internally, Directory Gateway Appliances are already deployed and no additional equipment is necessary.) DNS-based load balancing offers deployment flexibility which is standard industry practice (offered by other competitors) Choice of control parameters (source IPs, dynamic health monitors, etc) Integrated with SiteBacker Monitoring, Failover, and Notification service to provide maximum service uptime Fully Integrated with industry-leading Managed DNS service, Managed Internal DNS Service and Directional DNS Both internally and externally available for use inside or outside customers network perimeters Common, web-based Management Portal for simple administration and reduced Copyright NeuStar, Inc. 11

12 administrator expertise/training requirements Record-level load balancing configuration granularity Interaction case study: F5 Networks The power and flexibility of the Traffic Controller Service enables customers to leverage their existing traffic management and load balancing equipment if/when necessary. In the examples below, we demonstrate how Managed DNS services can complement and enhance an existing web infrastructure that uses load balancing solutions from F5 Networks. Similar scenarios can be applied to other local load balancing solutions offered by vendors such as Cisco Systems, Citrix/NetScaler, and others. Scenario 1: Load balancing with F5 3-DNS Controller and BIG-IP load balancers (Figure 1) Copyright NeuStar, Inc. 12

13 1. A user in Boston requests After walking the DNS tree to determine the authoritative DNS server for example.com, the request is sent to the primary name server (San Jose) to look up The F5 3-DNS controller in San Jose is serving as the Primary DNS Server. (NOTE: This type of request typically is sent from the customer to a recursive DNS server and the recursive DNS server handles the DNS request actions on behalf of the customer and subsequently will return the appropriate response to the customer itself.) Figure 1 2. The F5 3-DNS controller (in San Jose) queries the other F5 3-DNS controller(s) to determine which server is best suited to take the next DNS request based on F5 s algorithms configured by the 3-DNS administrators. 3. Based on the information gathered, the 3-DNS controller responds with the best IP Address - San Jose or DC. 4. User makes a connection with the F5 BIG-IP load balancer because of the DNS response given by the F5 3-DNS Controller. 5. The F5 BIG-IP directs the user to the server that is best fitted to receive the request based on BIG-IP load balancing algorithms configured by administrators. Scenario 2: Load balancing with Traffic Controller Service replacing F5 3-DNS controllers (Figure 2) Copyright NeuStar, Inc. 13

14 S e r v e r S erver Ser ver PBX PB X Minicomput er M in icom puter Se rver P BX Se rver Minicomputer PB X Minicomputer T r a f f i c C o n t r o l l e r S e r v i c e UltraDNS / BIG-IP UltraDNS Global DNS Network F5 BIG-IP (4) Customer Boston (1) (2) San Jose Web Server Farm San Jose F5 BIG-IP (3) (4) UltraDNS Dynamic Monitoring System (5) DC Web Server Farm DC Figure 2 1. A user in Boston requests After walking the DNS tree to determine the authoritative DNS server for example.com, the request is sent to a DNS resolver within the closest UltraDNS Node to look up IP Anycast enables the request to be routed to the closest topological node to respond to the DNS request. For example, a Boston user would typically get a response from a resolver located on the East Coast of the United States. 2. Each node in the UltraDNS global DNS network can be set to respond with a different/unique IP address, depending on the node queried. Administrators in organizations that want all East Coast users to be directed to a data center in DC and all the West Coast users to be directed to a data center in San Jose can do this by using Directional DNS. This allows one to get the user geographically closer to the content being served, quickly and efficiently. 3. Alternatively, by using the Traffic Controller Service, an administrator can load balance the requests across both the San Jose and DC BIG-IP controllers. This can be done in a Round Robin format or via some pre-configured percentage. If either BIG-IP controller is determined to be offline by the monitoring probes, all traffic would be sent to the alternate node. 4. Using the response from the UltraDNS node, the user makes a connection to the F5 BIG-IP load balancer because of the DNS response given by the Traffic Controller Service. Copyright NeuStar, Inc. 14

15 5. The F5 BIG-IP directs the user to the server that is best fitted to receive the request based on BIG-IP load balancing algorithms configured by administrators. 6. Using the monitoring system, UltraDNS monitors a given BIG-IP Virtual IP Address and will fail over the DNS responses given in the event of an outage. (the monitoring system will stop responding with the failed IP Address and respond with an active IP Address that has been configured in the Management Portal.) At pre-configured intervals (configured by administrators), various monitoring probes located across the UltraDNS global DNS network (or within a customer s network if being used internally) will check the sites (or F5 BIG-IPs) and determine if/when availability is returned. Given the two data centers in this scenario, if DC became unavailable, all traffic would be sent to the San Jose location. An notification is sent to notify the administrator of an outage (configurable via the Management Portal). Typical failover time is between 5-10 minutes and is based mainly on TTLs configured for the service/record. The monitoring nodes will continue to monitor the down location; when it is determined that the location/service is up again the resource is marked as being available again. Summary The example provided is just one case of how the UltraDNS Traffic Controller Service can be applied in a global load balancing solution. Essentially, Traffic Controller can take the place of any DNS-based redirection product to include F5 3-DNS, Cisco Distributed Director and others. To provide further competitive advantage, external and internal Traffic Controller is maintained in the same user interface (the UltraDNS Services Management Portal) as core services such as Managed DNS and Managed Internal DNS. This allows organizations to conduct business using a single platform, significantly reducing the need for diverse and highly technical administrators requiring several different forms of training. Couple this with UltraDNS Customer Support as part of its managed services and the cost of doing business in a highly technical load balanced world becomes justifiable to most organizations with multiple failover resources. Copyright NeuStar, Inc. 15

16 Specifications summary Services - Global server Load balancing (GSLB) - Server Monitoring, Failover, and Notification Load Balancing Techniques - DNS-Based Load Balancing Algorithms - DNS-based Round Robin Pools - DNS-based Failover Pools - DNS-based Weighted Round Robin (Percentage-based) - SuperPools consisting of multiple options above (i.e. Round Robin Pools in different locations controlled by a Failover Pool) Monitoring probe types - PING - SMTP - FTP - HTTP or HTTPS with content matching - Multi-step transaction probes (HTTP) - DNS Probes - TCP-connects to any custom port - Proxy Monitoring notifications - Notifications sent for failed server status - Notifications sent for servers past Warning or Critical thresholds (no configuration changes) Dynamic Load Balancing Metrics - Response time Management - Web-based Management Portal - XML-API Copyright NeuStar, Inc. 16

17 Support - Backed by UltraDNS Services Customer Support - 24x7x365 availability About NeuStar and NeuStar UltraDNS Services NeuStar, Inc. (NYSE: NSR) is a provider of clearinghouse and directory services to the global communications and Internet industry. Through its UltraDNS Services, NeuStar provides solutions to organizations that rely on the Internet Protocol (IP) for their critical business processes, applications, and communication services. NeuStar offers these solutions via one of the world s largest proprietary non-bind Directory Services Platforms. Providing integral Authoritative DNS services at the root level for top-level domain (TLD) registries and second-level domain (SLD) registrants, we have thousands of enterprise, service provider, and core infrastructure customers. NeuStar UltraDNS Services power the resolution of over 20 million domains around the world, and NeuStar is the only company of its kind that is actively involved at every level of the DNS tree. NeuStar Corporate Headquarters NeuStar, Inc Center Oak Plaza Sterling, VA Press Contact Marc Abshire Director of Global Media Relations Phone: (571) marc.abshire@neustar.biz UltraDNS East Coast Office Ridgetop Circle 2nd floor Sterling, VA West Coast Office 1000 Marina Blvd. (Suite 400) Brisbane, CA Copyright NeuStar, Inc. 17

18 Midwest Office 150 N. Michigan Avenue (28th Floor) Chicago, IL European Office London, England 4 Lombard Street London EC3V 9HD +44 (0) Customer Support Support@UltraDNS.com Phone: (888) Sales & Information Call (888) or click here to send a sales or information request. Copyright NeuStar, Inc. 18