How can I prevent my host headers from occasionally disappearing? Scott Schnoll: Most folks enter host header information in Internet Services Manager; however, when you do that on an Exchange 2000 server, they don't stay. Exchange will change the metabase information back to the way it thinks it should be (based on how it is configured using Exchange System Manager). So, you'll want to enter host headers using Exchange System Manager,...
and not Internet Services Manager. See the Microsoft Knowledge Base article How to configure host header and authentication information in Exchange 2000 Outlook Web Access on a Windows 2000 cluster server for step-by-step instructions for doing this.
I have two back ends, A (Exch1) and B (Exch2). Both have data files in a common storage in separate disks, but not in a cluster. If server A fails I want to assign the disks from S.A to S.B and have S.B working with Exch1 and Exch2. Is this possible? Thank you.
Scott Schnoll: No, that is not possible. If you are not clustering, you won't be able to assign one Exchange 2000 server's storage elements (disks, databases, logs, etc.) to a different, separate Exchange 2000 server.
To clarify, I don't need NNTP on my Exchange 2000 cluster because it is not supported, right?
Scott Schnoll: You do need NNTP on your Exchange 2000 cluster because, even though you cannot use it in a cluster, it is still a hard-coded requirement for installing Exchange 2000. If you don't install it on your nodes, Exchange 2000 will not install.
We already installed the cluster as active/active, DC and GC, and I want to remove it and rebuild it. Can you point me to the proper white papers? Thanks!
Scott Schnoll: I'm not aware of any white papers that discuss the procedure for this. As I see it you have two options:
- Install a standalone Exchange 2000 server and move all of the mailboxes/PFs/roles, etc., to this new system. Then demote the nodes, remove the cluster service and reinstall the cluster service once both nodes are member servers again. Then install Exchange 2000 into the cluster and create at least one EVS (Exchange Virtual Server). Then, move mailboxes/PFs/etc. to the EVS in the cluster. At your discretion, create an additional EVS to return to the active/active model or remain (as is preferred) in the active/passive model with a single EVS.
- Demote the nodes to member services and reconfigure the cluster and Exchange with any permissions that may have broken. See Changing server status on a server cluster node affects security permissions for additional information on this.
Failover is transparent to the client, so doesn't a cluster provide 100% uptime, and not just 99.999%?
Scott Schnoll: Remember that clustering does not protect the shared state between client and server. Any transactions in progress are lost when failover occurs. So, failover isn't really transparent. It may appear that way when using very fast hardware and not-so-busy clients, but there is loss of connection and session information. In fact, per Knowledge Base article 310793, even the latest Outlook client may need to be restarted after a failover has occurred.
I'm trying to use OWA against my Exchange 2000 cluster, but all I get is a page not found error. How can I troubleshoot this?
Scott Schnoll: The most common cause of this is that host headers are missing or have been removed. Most folks enter host header information in Internet Services Manager; however, when you do that on an Exchange 2000 server, they don't stay. Exchange will change the metabase information back to the way it thinks it should be (based on how it is configured using Exchange System Manager). So, you'll want to enter host headers using Exchange System Manager, and not Internet Services Manager. See XCCC: How to configure host header and authentication information in Exchange 2000 Outlook Web Access on a Windows 2000 cluster server, MSKB 287726, for step-by-step instructions for doing this.
What are the drawbacks to also making my cluster nodes Active Directory domain controllers and global catalog servers?
Scott Schnoll: There are three main reasons why you want your Exchange 2000 cluster nodes to be member servers and not domain controllers or global catalog servers. First, Active Directory has its own built-in mechanisms for replication and redundancy, and to a certain extent, failover. Second, both Active Directory and Exchange 2000 are resource hungry applications. If you put them both on the same machine, they will essentially fight it out for resources. Third, adding Active Directory complicates restore and recovery of your nodes. It's much easier to recover from the loss of a node if all you have to do is recover Exchange 2000. If you need to also recover Active Directory, you'll have additional downtime and complexity to deal with. Having said that, if one cluster node in a two-node cluster is a domain controller, the other node must also be a domain controller.
What would you recommend for providing high-availability or failure for the application and/or database?
Scott Schnoll: Microsoft Cluster Services will provide high-available for Exchange 2000. If your primary cause of outages is hardware failure, then look to cluster Exchange 2000. If your outages are caused by other reasons (e.g., corrupt databases, viruses, network outages, etc.), then clustering won't do anything for you. In that case, you'll want to look at the causes of downtime and find solutions that prevent the downtime in the first place or mitigate the damage done as a result of the outage.
How do you address wide area disaster recovery for Exchange 2000?
Scott Schnoll: I'm not entirely sure what you mean by "wide area disaster recovery" for Exchange 2000. For Exchange 2000 DR (disaster recovery), I follow Microsoft's guidelines and best practices. They have some great documentation, including Disaster Recovery for Microsoft Exchange 2000 Server, which discusses the various tasks and planning required for DR. I would also have a look at Exchange 2000 Server Database Recovery, and the other white papers.
Is Load Balancing supported in active/passive?
Scott Schnoll: No, you cannot use Windows Load Balancing and Cluster Services on the same system. When clustering, the best you can achieve is something referred to sometimes as "static load balancing." This means you basically carve up the users and distribute them evenly across your EVSes (e.g., 50% on one EVS and 50% on another EVS).
What are the security advantages to clustering Exchange 2000?
Scott Schnoll: There are no security advantages to clustering.
You mentioned two facts specific to active/active clusters. MS recommends not more than 40% CPU utilized. Is that the percentage you said? Second question: how many MAPI clients supported in an active/active cluster? I didn't catch that.
Scott Schnoll: Yes, 40% is correct (and what I said ). In the active/active model, you must limit the number of concurrent MAPI users to 1,900 per node (with SP2 or greater; with SP1, the limit is 1,500, but we both know you really want to be running SP3 anyway).
What am I doing wrong? I am unable to write to Active Directory on a clustered server with two nodes? Is there a document explaining how to setup two nodes on a clustered server? Thanks.
Scott Schnoll: It could be a permissions issue. It's hard to say because I don't know what error message you are getting, when you are getting it or what you are doing when you get it. But if you are getting the error during setup of Exchange 2000, then I would research the error message in TechNet, or the Microsoft Knowledge Base.
Could you elaborate on the requirement of the disks to be basic instead of dynamic? Is there a specific reason for this that you know of other than the fact that it is unsupported?
Scott Schnoll: Microsoft's Windows Server products don't support dynamic disks in a server cluster environment. I'm told that the Volume Manager for Windows 2000 add-on product from Veritas can be used to add the dynamic disk features to a server cluster. When the Veritas Volume Manager product is installed on a cluster, Veritas should be the first point of support for cluster issues and not Microsoft. Unfortunately, all Microsoft says is that you cannot configure external disks as dynamic disks or spanned volumes (volume sets) if they will be used as cluster resources. They do not elaborate on why they are not supported. Also keep in mind that you cannot use the Encrypting File System, Remote Storage, mounted volumes or parse points on the cluster storage, and you cannot use software RAID either.
Is there anything wrong with configuring an Exchange and a file/print server as active/active?
Scott Schnoll: I'm not sure what you mean. Do you mean multiple EVSes and multiple file/printer shares? If so, I would not do that. Multiple EVSes are going to demand all of the resources they can get, and file and print sharing will reduce Exchange's performance. Also, keep in mind that when you optimize the server service, you can choose from "optimize for network applications," which you would pick for Exchange, and "optimize for file sharing," which you would pick for file/print sharing. In other words you tune the server service for what is being served, and it gets tuned differently for these two resources. Therefore, I would not mix them in this way.
Having said that, if you really meant a single EVS on one node, and file/print services on another node, then that should be OK for you. During failover, performance will likely suffer a little, but provided you have enough hardware resources on your nodes, you should be OK.
A recent article I read said that DNS is not cluster-aware. Is this true?
Scott Schnoll: That is correct. Unlike WINS and DHCP, which can run in a cluster, DNS cannot. However, DNS has its own fault-tolerance capabilities in the form of primary and secondary DNS servers. Therefore, there's really no reason to need to cluster it.
Can you install IIS and Exchange 2000 on cluster?
Scott Schnoll: Yes, and in fact if you are clustering Exchange 2000, you *must* install IIS and Exchange 2000 in the cluster.
Can you make the cluster a PDC?
Scott Schnoll: Yes, but best practices state that you do not. It's best to install the node as a member server.
What is the 2000 Advance Server?
Scott Schnoll: Clustering Exchange 2000 requires Windows 2000 Advanced Server or Windows 2000 Datacenter Server, and Exchange 2000 Enterprise Edition.
For how many recipients would you recommend a cluster configuration versus single server?
Scott Schnoll: That depends on what you mean by "recipient." In some cases (e.g., when recipients are external), it doesn't matter so much how many there are. If by "recipient," however, you mean mailboxes, then sizing an Exchange 2000 server or cluster is easy. A standalone server's capabilities directly relate to the available resources. Microsoft and Compaq/HP both have some great Exchange Server sizing guides you can use to properly size your server. Also, LOADSIM and ESP are both great tools for simulating client loads and validating your hardware configuration. An active/passive Exchange 2000 cluster will get sized the same way a standalone, unclustered server would. You're only limited in this case by available resources. However, in active/active, you must limit the number of concurrent MAPI users to 1,900 per node, and make sure the CPU usage on each node stays under 40%.
What is the easiest way to programatically detect and enumerate clustered Exchange server names on a node?
Scott Schnoll: That depends on what language you want to use. The cluster APIs are fairly straightforward, and Microsoft has some good sample code on how to do this in VB and VC++. If you're using VB, see Enumerate cluster resources by resource type in Visual Basic; if you are using VC++, see Enumerate cluster resources by resource type in Visual C++. Also, have a look at the Windows clustering section of Microsoft's Platform SDK, which discusses the various enumeration methods available for cluster components and resources. You may also be interested in the article in that section called "Enumerating Objects with Cluster Automation Server."
Can I use two public address for the same EVS?
Scott Schnoll: No, you'll want to use a single address for each EVS.
I'm using three NICs for redundancy. How can I make sure that I will have service even if one of the network switches on one network is down?
Scott Schnoll: Have a look at Network failure detection and recovery in a two-node server cluster and Impact of network adapter failure in a cluster for details on how Windows 2000 Advanced Server's detection algorithm works to detect network failures on a cluster.
Concerning geographically displaced clusters, how far can the nodes be apart and what are the drawbacks?
Scott Schnoll: First, let's discuss the requirements for geographically dispersed clusters. The nodes in a cluster may be on different physical networks; however, the private and public network connections between cluster nodes must appear as a single, non-routed LAN using technology such as a virtual LAN (VLAN). Round-trip communication latency between the cluster nodes must be no more than 500 milliseconds. Each VLAN must fail independently of all other cluster networks.
To test roundtrip latency issues with your network infrastructure, Microsoft has a tool called the Time Latency Tool that can be used to verify the latency of heartbeat messages between your cluster nodes. The tool monitors the round-trip latency of UDP (user datagram protocol) packets between nodes in your cluster. If it discovers that the latency between nodes is above the pre-defined value, it writes a warning message to a log file. Click here to learn more about it.
Generally speaking, due to the complexity of geographically dispersed clusters, you're going to need help from the hardware vendor whose geo-clustering equipment you are purchasing. Often, there are third-party drivers that are needed from these vendors, as well.