Chapter 9: Implementing Windows Cluster Services and Network Load Balancing


Making Server Clustering Part of Your High-Availability Plan

1.

You have purchased a prepackaged solution that uses an eight-node majority node set (MNS) server cluster. Because you have so many nodes, you have decided to install three nodes in your Atlanta data center, three in your Denver data center, and the last two in your Seattle sales office. You notice fairly soon that the server cluster is experiencing some uptime issues. The nodes in your Atlanta data center seem to fail frequently during times of high WAN utilization. What is likely the problem?

  1. All nodes in an MNS server cluster must be in the same data center.

  2. The high WAN traffic is making the heartbeats take longer than 500 ms to get to all nodes and back.

  3. The nodes in Atlanta are failing, and an MNS server cluster can have two nodes fail before losing quorum and shutting down.

  4. The cluster cannot be in three geographic areas. An MNS server cluster can exist in a maximum of two geographic regions, and high-speed networks must connect the nodes in each region.

 b . no more than 500 ms total round-trip time is allowed between nodes in an mns server cluster. the high wan utilization is probably making the transit time exceed 500 ms. a , c , d . answer a is incorrect because the mns model is specifically designed for geographic distribution. answer c is incorrect because an eight-node mns server cluster can tolerate three failed nodes before the cluster shuts down. answer d is incorrect because there is no limitation on the number of geographic areas used, nor is there any communications speed requirement beyond the 500 ms round-trip issue.

2.

Your data center experiences a power failure, bringing all of your systems down. When power is returned, a single quorum device server cluster you have in use will not start. You examine the event logs and find error messages stating that the quorum drive cannot be found, yet you are able to view the contents of the quorum drive in Windows Explorer. Research reveals that either the disk signature on the quorum drive or the Registry key containing the disk signature for the quorum drive has been corrupted. What steps should you take to recover from this problem?

  1. Evict all other nodes from the server cluster, repartition and reformat the quorum drive, and rejoin the other nodes to the server cluster.

  2. Do a restore of the quorum drive from tape.

  3. Change the location of the quorum resource to another drive, repartition and reformat the quorum drive, and move the quorum resource back to the original quorum drive.

  4. Shut down all nodes except one, perform an ASR restore on that node, and restart all the nodes.

 d . answer d is correct because properly performed asr backups record drive signatures and layouts, including clustered drives. an asr restore will rewrite the signatures and registry settings. a , b , c . answer a is incorrect because this process would destroy the server cluster. answer b is incorrect because a normal restore would not repair corrupted registry keys or disk signatures. the registry does not reside on the quorum drive. answer c is incorrect because the quorum resource is not able to come online in its corrupted state. therefore, you would not be able to move the quorum resource to another drive.

3.

As a consultant, you have been called in to attempt to fix a high-availability configuration that is not performing as designed. Your client wanted to provide high availability for a high-traffic Web site. The client purchased a preconfigured, mid-range, two-node server cluster and implemented IIS on the nodes. Response time for serving Web pages is unacceptable, although there have been no incidents of the application failing over. What is the correct fix for this situation?

  1. More nodes need to be added to the server cluster. Increase the number of nodes until performance reaches an acceptable level.

  2. Add NLB to the server cluster to handle more requests from clients simultaneously.

  3. Convert the server cluster to an NLB cluster.

  4. Move the server cluster to high-end hardware to provide quicker response times.

 c . answer c is correct because the client s stated purpose for the cluster is appropriate for an nlb cluster, not a server cluster. nlb is designed for handling large volumes of traffic. server clusters are designed to provide increased availability for specific applications. a , b , d . answer a is incorrect. adding more nodes will not improve performance because the application is not meant for server cluster use. answer b is incorrect because nlb cannot coexist with a server cluster on the same hardware. answer d is incorrect because the application will not take advantage of the higher-end hardware and will not yield an increase in performance.

4.

You have been asked to design a server cluster. The server cluster will start small, but it may expand as more applications are added and predicted growth is experienced. Your proposal is for two nodes, a shared storage device, Fibre Channel host bus adapters, and switches for connectivity. When you present your proposal to management, you are asked to justify the high cost of the Fibre Channel solution. What justification can you provide for implementing Fibre Channel?

  1. Fibre Channel supports more than two nodes, allowing for the predicted growth.

  2. Fibre Channel is the fastest connectivity solution and will therefore yield the highest performance.

  3. Fibre Channel easily expands to allow more storage to be added to support the future applications.

  4. All of the above.

 d . fibre channel is fast, supports the full eight-node maximum cluster size, and is easily expandable by adding fibre channel storage devices. a , b , c . each of these answers addresses only part of the benefits of a fibre channel implementation.

5.

You are configuring a large, single quorum device server cluster consisting of eight nodes and a dozen shared storage cabinets with 30 logical drives among them. Because of the large number of logical drives, you are using mount points instead of drive letters on most of the drives. After running the Wizard to create your first node, you can see only the drives that have been assigned drive letters. How is this resolved?

  1. Install the second node, which will automatically create mount point resources.

  2. Manually create the disk resources after the first node is created.

  3. Reconfigure the shared storage to reduce the number of logical drives to less than 16.

  4. Temporarily assign drive letters to the mount point drives, and then remove the drive letters after the Wizard finishes installing the first node.

 b . answer b is correct because mounted drives are not automatically detected during node installation. a , c , d . answer a is incorrect because this is not the behavior of the installation of the second node. answer c is incorrect because reducing the number of drives will not affect how mounted drives are configured. answer d is incorrect because you cannot change drive letters on a node and a sufficient number of drive letters are not available to attempt this process.

6.

You are configuring a two-node, single quorum device server cluster with a single public network interface and a single interconnect interface. The network interfaces and storage devices have been configured, and the interconnects on both nodes have been connected with a direct crossover Ethernet cable. The installation of the first node proceeds without incident, but when attempting to create the second node, the installation fails. The Wizard reports problems communicating with the first node over the interconnect. You have verified that the cables are functional and have been properly inserted into the connectors. What is the most likely problem?

  1. The interconnect adapters are configured for auto-negotiation or for different speed and duplex settings.

  2. The direct crossover cable method cannot be used with this cluster configuration.

  3. A second interconnect is required with this cluster configuration.

  4. A switch must be used to handle heartbeat traffic.

 a . the ethernet link is not established. most likely, this is due to conflicting speed and duplex settings. auto-negotiation can have this result as well. b , c , d . answer b is incorrect because the direct crossover cable method does work with this configuration, and is quite common. answer c is incorrect because a second interconnect is not required. the presence of a second interconnect may allow the second node to join the cluster but will not resolve the issues present on the first interconnect. answer d is incorrect because an additional network device such as a switch is required in only a three-or-more node configuration.

7.

You have installed a third-party backup agent on your nodes. The agent is supposed to listen for requests from its control server and send data to it during a backup. Despite this, your backups are failing. The application on the control server reports that it cannot communicate with the agent. You check the node and see that the agent is running properly. What is the most likely problem?

  1. The agent is not server cluster-compatible and cannot be used on a node.

  2. The control server is attempting to communicate with the agent over the interconnect network.

  3. There is a firewall between the control server and the node running the agent.

  4. The agent has configured itself to listen on the interconnect instead of the public network.

 d . answer d is correct because the installation routine for the agent most likely picked the interconnect network for listening for requests. the agent must be reconfigured to listen for requests on the public network interface. a , b , c . answer a is incorrect because the agent may or may not be cluster-compatible. there is not enough information provided to make that determination. answer b is incorrect because the control server will not even be aware of the interconnect network if the interconnect is properly configured. answer c is incorrect because, while this is a technical possibility, it does not fit into the scenario described.

8.

You have created a small, two-node, single quorum device server cluster to act as a print server for several hundred printers. The shared storage is a 4GB drive. Because of the small size of the shared storage and the transient nature of the data, the print spool resource is on the quorum drive. The server cluster operates acceptably for a period of time, and then both nodes are taken down by a sudden power failure. When power is restored, the nodes boot, but the cluster service will not start. How do you fix this problem and prevent it from happening again?

  1. Delete the files under the spool directory, remove the spooler resource, add external storage, and re-create the spool on a different drive.

  2. You cannot resolve this issue. Once the quorum drive is filled, all nodes must be evicted and the server cluster re-created.

  3. Reformat the quorum drive and apply disk quotas to prevent the spooler from filling the drive again.

  4. Perform an ASR restore on the nodes.

 a . answer a is correct because a server cluster cannot start if the quorum drive is full. restructuring the shared storage is the only way to ensure that the problem does not reoccur. b , c , d . answer b is a false statement. removal of files will allow the server cluster to start. answer c is incorrect because reformatting the quorum drive will destroy the server cluster. also, disk quotas would not necessarily resolve the issue. answer d is incorrect because an asr restore would not resolve the out-of-disk-space issue and allow the server cluster to restart.

9.

You are configuring a large, single quorum device server cluster consisting of eight nodes and a dozen shared storage cabinets with 30 logical drives among them. The storage cabinet that contains the quorum drive also contains eight other logical drives and is connected to the last port on your 32-port Fibre Channel switch. While running the Wizard to create your first node, you cannot see any of the drives in the quorum drive’s cabinet, including the quorum drive. Which of the following is a possible cause of the problem?

  1. The maximum number of logical drives recognizable in by a server cluster configuration has been exceeded.

  2. The cabinet containing the quorum drive is not properly connected or powered on.

  3. The cabinet containing the quorum drive must be relocated to a lower numbered Fibre Channel switch port.

  4. The maximum number of storage devices recognizable by a cluster has been exceeded.

 b . given the choices, answer b is the only possible cause of the missing quorum drive and other drives in that cabinet. a , c , d . answer a is incorrect because the number of logical drives specified (30) falls within the number of drives recognizable by windows server 2003 and the cluster service. answer c is incorrect because the ordering of the fibre channel connected devices is unknown and irrelevant to windows server 2003. answer d is incorrect because the number of storage devices is an issue only with the fibre channel configuration, not the configuration of the server cluster or the operating system.

Answers

1.

B. No more than 500 ms total round-trip time is allowed between nodes in an MNS server cluster. The high WAN utilization is probably making the transit time exceed 500 ms.

A, C, D. Answer A is incorrect because the MNS model is specifically designed for geographic distribution. Answer C is incorrect because an eight-node MNS server cluster can tolerate three failed nodes before the cluster shuts down. Answer D is incorrect because there is no limitation on the number of geographic areas used, nor is there any communications speed requirement beyond the 500 ms round-trip issue.

2.

D. Answer D is correct because properly performed ASR backups record drive signatures and layouts, including clustered drives. An ASR restore will rewrite the signatures and Registry settings.

A, B, C. Answer A is incorrect because this process would destroy the server cluster. Answer B is incorrect because a normal restore would not repair corrupted Registry keys or disk signatures. The Registry does not reside on the quorum drive. Answer C is incorrect because the quorum resource is not able to come online in its corrupted state. Therefore, you would not be able to move the quorum resource to another drive.

3.

C. Answer C is correct because the client’s stated purpose for the cluster is appropriate for an NLB cluster, not a server cluster. NLB is designed for handling large volumes of traffic. Server clusters are designed to provide increased availability for specific applications.

A, B, D. Answer A is incorrect. Adding more nodes will not improve performance because the application is not meant for server cluster use. Answer B is incorrect because NLB cannot coexist with a server cluster on the same hardware. Answer D is incorrect because the application will not take advantage of the higher-end hardware and will not yield an increase in performance.

4.

D. Fibre Channel is fast, supports the full eight-node maximum cluster size, and is easily expandable by adding Fibre Channel storage devices.

A, B, C. Each of these answers addresses only part of the benefits of a Fibre Channel implementation.

5.

B. Answer B is correct because mounted drives are not automatically detected during node installation.

A, C, D. Answer A is incorrect because this is not the behavior of the installation of the second node. Answer C is incorrect because reducing the number of drives will not affect how mounted drives are configured. Answer D is incorrect because you cannot change drive letters on a node and a sufficient number of drive letters are not available to attempt this process.

6.

A. The Ethernet link is not established. Most likely, this is due to conflicting speed and duplex settings. Auto-negotiation can have this result as well.

B, C, D. Answer B is incorrect because the direct crossover cable method does work with this configuration, and is quite common. Answer C is incorrect because a second interconnect is not required. The presence of a second interconnect may allow the second node to join the cluster but will not resolve the issues present on the first interconnect. Answer D is incorrect because an additional network device such as a switch is required in only a three-or-more node configuration.

7.

D. Answer D is correct because the installation routine for the agent most likely picked the interconnect network for listening for requests. The agent must be reconfigured to listen for requests on the public network interface.

A, B, C. Answer A is incorrect because the agent may or may not be cluster-compatible. There is not enough information provided to make that determination. Answer B is incorrect because the control server will not even be aware of the interconnect network if the interconnect is properly configured. Answer C is incorrect because, while this is a technical possibility, it does not fit into the scenario described.

8.

A. Answer A is correct because a server cluster cannot start if the quorum drive is full. Restructuring the shared storage is the only way to ensure that the problem does not reoccur.

B, C, D. Answer B is a false statement. Removal of files will allow the server cluster to start. Answer C is incorrect because reformatting the quorum drive will destroy the server cluster. Also, disk quotas would not necessarily resolve the issue. Answer D is incorrect because an ASR restore would not resolve the out-of-disk-space issue and allow the server cluster to restart.

9.

B. Given the choices, Answer B is the only possible cause of the missing quorum drive and other drives in that cabinet.

A, C, D. Answer A is incorrect because the number of logical drives specified (30) falls within the number of drives recognizable by Windows Server 2003 and the cluster service. Answer C is incorrect because the ordering of the Fibre Channel connected devices is unknown and irrelevant to Windows Server 2003. Answer D is incorrect because the number of storage devices is an issue only with the Fibre Channel configuration, not the configuration of the server cluster or the operating system.

Making Network Load Balancing Part of Your High-Availability Plan

10.

You have installed an NLB cluster onto a 10/100 Mbps switch. Other devices, including some older 10 Mbps-only devices, are also attached to the switch. Your NLB hosts are configured for 100 Mbps and full duplex. Soon, you notice that communications with the 10 Mbps devices have failed. After troubleshooting, you discover that apparently the increased traffic on the switch is preventing the 10 Mbps devices from having sufficient bandwidth for reliable communications. What is the best fix for this problem?

  1. Change the operating mode of the NLB cluster to multicast and enable IGMP support.

  2. Relocate all of the NLB hosts to a different virtual LAN (VLAN).

  3. Relocate all of the 10 Mbps-only hosts to the same VLAN.

  4. Install a firewall between the NLB hosts and the 10 Mbps-only devices and filter all NLB-oriented traffic.

 a . answer a is correct because the failure of the 10 mbps-devices combined with increased traffic on the switch indicates a switch-flooding problem. changing the nlb cluster mode to multicast with igmp support can help resolve this issue by controlling the nlb heartbeat traffic and limiting it to those ports on the switch that are part of the nlb cluster. b , c , d . answers b and c are incorrect because this is a more complicated solution than enabling multicast and igmp. you should not need to encounter such additional router management functions to enable nlb. answer d is incorrect because the problem is resolvable through configuration of the nlb cluster and does not require additional software (the firewall). in addition, any requests from the 10 mbps devices would go unresolved due to the firewall implementation.

11.

You have configured an NLB cluster with 10 hosts. The default port rule has been changed from all possible ports to just port 80. No other port rules have been defined. You have configured each node with IIS and followed the appropriate procedures for installing and securing IIS. After clients begin using the cluster, you notice that clients requesting normal Web pages are being served equally across the cluster, but clients requesting secured Web pages (SSL) and FTP sessions are all going to the host with priority 1. What is the best way to resolve this issue and to balance the SSL and FTP requests?

  1. Do nothing. SSL and FTP traffic cannot be load-balanced.

  2. Split the NLB cluster into three clusters and serve the SSL and FTP sessions from different clusters.

  3. Add new port rules for the SSL and FTP traffic.

  4. Change the default port rule back to encompass all possible ports

 c . answer c is correct because it would balance traffic on all relevant ports and provide a more secure configuration. a , b , d . answer a is incorrect because ssl and ftp traffic can be load-balanced if the cluster is configured properly. answer b is incorrect because this would require an enormous amount of administrative effort and, if the same configuration steps were followed, would suffer the same problem. answer d is technically possible, but does not provide the additional security benefit that answer c does.

12.

You are a consultant. You have been called in to troubleshoot a malfunctioning NLB cluster that serves IIS Web pages. The cluster in question consists of six hosts, but only four successfully join the cluster. Two of the hosts never successfully join. When the rest of the hosts are shut down and those two hosts are started up together, they successfully perform convergence and form a cluster. This two-host cluster, however, seems to favor certain types of incoming traffic on each host, rather than equally among the two hosts. What is the most likely reason for this behavior?

  1. The two malfunctioning hosts are configured with different cluster IP addresses and a different host name than the four correctly operating hosts.

  2. The two malfunctioning hosts are underpowered and cannot join the cluster due to poor performance.

  3. The two malfunctioning hosts are configured with different port rules than the four correctly operating hosts.

  4. The two malfunctioning hosts are configured with the same priority.

 c . answer c is correct because nlb will not allow convergence for a host or hosts that have different or different numbers of port rules. since the two hosts do form their own cluster but the traffic pattern differs, it would appear that they are configured for the same port rules, but different rules from the other four hosts. a , b , d . answer a is incorrect because if the two hosts were configured with different ip addresses and a different host name from the other four hosts, you would end up with two nlb clusters. answer b is incorrect because nlb does not perform any sort of performance test before a convergence. answer d is incorrect because if the two hosts had identical host priorities they would not form a cluster of their own.

13.

You are a consultant. You have been called in to troubleshoot a malfunctioning NLB cluster that is supposed to serve Web pages with IIS. The cluster contains four hosts, but only one host at a time will successfully form the cluster. Clients appear to have no problems connecting to any of the single-host cluster configurations. What is the most likely cause of the problem?

  1. The hosts are configured with duplicate priorities.

  2. The hosts are configured with different port rules.

  3. The hosts are configured with different cluster IP addresses.

  4. The hosts are configured with duplicate cluster IP addresses.

 a . since each host forms the cluster individually but not together, it is likely that the hosts are configured with the same host priority. b , c , d . answer b is incorrect because the problem of duplicate port rules does not fit the behavior described. answer c is incorrect because different ip addresses would cause clients to experience connection failures. answer d is incorrect because hosts should be configured with duplicate cluster ip addresses. this would not yield the problem described.

14.

One of your hosts in multiple-host NLB cluster requires maintenance. The cluster is heavily used and central to the profitability of your company. You want to bring the node down for service in the least disruptive way. How should you accomplish this goal?

  1. Use the drainstop option on the host needing maintenance.

  2. Use the drainstop option on all the hosts in the cluster not needing maintenance.

  3. Use the suspend option on the host needing maintenance.

  4. Use the suspend option on all the hosts in the cluster not needing maintenance.

 a . answer a is correct because the drainstop option is used to finish servicing active requests on an active node without accepting any new connections. b , c , d . answer b is incorrect because every host except the one needing service would be shut down. answer c is incorrect because the sessions being serviced by the host would time out and be lost. answer d would cause every session not on the intended node to time out and be lost.

15.

You have been asked to develop a design for an NLB cluster for an IIS-based Web site. The specifications given to you state that the Web application will be using server-side cookies to keep track of a visitor’s session state. Which port-rule filtering mode should you configure to support the application?

  1. Single host

  2. Multiple host/Affinity: None

  3. Multiple host/Affinity: Single

  4. Multiple host/Affinity: Class C

 c . answer c is correct because multiple host/single affinity supports server-side cookies, meeting the given specifications. a , b , d . answer a is incorrect because single-host filtering directs all traffic for a specific port to a specific host, depending on the port rule configuration. answer b is incorrect because the multiple host/affinity: none mode would lead to lost state information as clients are redirected to other hosts in the cluster. answer d is incorrect because the multiple host/affinity: class c mode would appear fully functional to most clients, but could fail if clients changed ip addresses while their session state was being tracked (as happens with some isps).

Answers

10.

A. Answer A is correct because the failure of the 10 Mbps-devices combined with increased traffic on the switch indicates a switch-flooding problem. Changing the NLB cluster mode to multicast with IGMP support can help resolve this issue by controlling the NLB heartbeat traffic and limiting it to those ports on the switch that are part of the NLB cluster.

B, C, D. Answers B and C are incorrect because this is a more complicated solution than enabling multicast and IGMP. You should not need to encounter such additional router management functions to enable NLB. Answer D is incorrect because the problem is resolvable through configuration of the NLB cluster and does not require additional software (the firewall). In addition, any requests from the 10 Mbps devices would go unresolved due to the firewall implementation.

11.

C. Answer C is correct because it would balance traffic on all relevant ports and provide a more secure configuration.

A, B, D. Answer A is incorrect because SSL and FTP traffic can be load-balanced if the cluster is configured properly. Answer B is incorrect because this would require an enormous amount of administrative effort and, if the same configuration steps were followed, would suffer the same problem. Answer D is technically possible, but does not provide the additional security benefit that Answer C does.

12.

C. Answer C is correct because NLB will not allow convergence for a host or hosts that have different or different numbers of port rules. Since the two hosts do form their own cluster but the traffic pattern differs, it would appear that they are configured for the same port rules, but different rules from the other four hosts.

A, B, D. Answer A is incorrect because if the two hosts were configured with different IP addresses and a different host name from the other four hosts, you would end up with two NLB clusters. Answer B is incorrect because NLB does not perform any sort of performance test before a convergence. Answer D is incorrect because if the two hosts had identical host priorities they would not form a cluster of their own.

13.

A. Since each host forms the cluster individually but not together, it is likely that the hosts are configured with the same host priority.

B, C, D. Answer B is incorrect because the problem of duplicate port rules does not fit the behavior described. Answer C is incorrect because different IP addresses would cause clients to experience connection failures. Answer D is incorrect because hosts should be configured with duplicate cluster IP addresses. This would not yield the problem described.

14.

A. Answer A is correct because the drainstop option is used to finish servicing active requests on an active node without accepting any new connections.

B, C, D. Answer B is incorrect because every host except the one needing service would be shut down. Answer C is incorrect because the sessions being serviced by the host would time out and be lost. Answer D would cause every session not on the intended node to time out and be lost.

15.

C. Answer C is correct because multiple host/single affinity supports server-side cookies, meeting the given specifications.

A, B, D. Answer A is incorrect because single-host filtering directs all traffic for a specific port to a specific host, depending on the port rule configuration. Answer B is incorrect because the Multiple host/Affinity: None mode would lead to lost state information as clients are redirected to other hosts in the cluster. Answer D is incorrect because the Multiple host/Affinity: Class C mode would appear fully functional to most clients, but could fail if clients changed IP addresses while their session state was being tracked (as happens with some ISPs).




MCSE Planning and Maintaining a Windows Server 2003 Network Infrastructure. Exam 70-293 Study Guide and DVD Training System
MCSE Planning and Maintaining a Windows Server 2003 Network Infrastructure: Exam 70-293 Study Guide and DVD Training System
ISBN: 1931836930
EAN: 2147483647
Year: 2003
Pages: 173

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net