Server Consolidation Recommendations
Topic Last Modified: 2006-08-16
When you identify server consolidation opportunities, it is a good idea to consider best practices that might apply to your specific situation. For example, it is highly recommended to use hardware from the Windows Server Catalog when you design servers running Microsoft® Exchange Server 2003. The Windows Server Catalog lists the hardware that Microsoft has certified for Windows Server™ 2003. The catalog is available at http://go.microsoft.com/fwlink/?linkid=17219. If you want to deploy multiple servers in a Windows cluster, ensure that your hardware is listed in the catalog's "Cluster Solutions" section.
|Microsoft supports hardware for Windows Server 2003 clusters only if it has passed the Windows Hardware Quality Labs (WHQL) test for the Windows Server 2003 family.|
When you plan a server consolidation, keep the following recommendations in mind:
Design server hardware generously Design your server hardware according to current and future requirements to prepare for future growth. You might want to consider additional processors, 2 GB or more of memory, and a reliable storage subsystem that has a capacity of at least two or three times the estimated size of your messaging databases. Note that hardware technology evolves at a rapid pace. Within a relatively short period of time, upgrade options might not be available for your server platform, which can pose a serious problem if future demands require you to increase system performance; for example, in the event that you need additional processors.
Note: To enable the Microsoft Exchange Information Store service to use up to 3 GB of virtual address space on a server that has more than 1 GB of physical memory, modify the Boot.ini file on your server and add the /3GB and /USERVA parameters to the startup line for the operating system. For more information about these parameters, see Microsoft Knowledge Base Article 316739, "How to use the /userva switch with the /3GB switch to tune the User-mode space to a value between 2 GB and 3 GB."
Avoid single points of failure By using fewer servers, you have fewer points of potential failure; however, the impact of failures increases. You can implement redundancy to address potential points of failure. If one component fails, another component can take over. Performance might be degraded temporarily during an outage, but users can continue their work.
You should provide redundancies for the following components:
General server components For processor, memory, motherboard, and other components, consider implementing a Windows Clustering solution if high availability is a critical requirement for your mailbox servers. Exchange 2003 supports clusters running Windows Clustering. If one out of a maximum of eight nodes fails, another node in the cluster assumes the processing until the original node is repaired.
Storage subsystem Windows clusters do not provide redundancies for the storage subsystem. For this reason, it is important to provide redundancies within the storage subsystem itself; for example, you can implement a redundant array of independent disks (RAID). RAID configurations can also lead to an increased server performance when multiple disks perform I/O operations concurrently. The following table lists typical RAID configurations for Exchange 2003.
Typical RAID configurations for Exchange 2003
RAID Technology Description Used For Comments
An array of striped disks without parity.
Possibly client computers' disk drives or servers. Used in conjunction with RAID 1.
RAID 0 achieves very high performance because all disks can read or write data concurrently, but a significant disadvantage of this RAID configuration is lack of fault tolerance and a high risk of disk failure. If only one disk in the array breaks, all data is lost and must be restored from backup. For this reason, RAID 0 without RAID 1 is seldom used in server systems.
An array of mirrored disks.
Operating system, paging file, and transaction logs.
RAID 1 achieves very high performance and very high fault tolerance because all data is mirrored after it is written, and both disks contain the complete data. This configuration is a perfect choice for data that requires the highest reliability. Its most significant disadvantage is its high cost. RAID1 is the recommended technology for the transaction logs on mailbox and public folder servers.
A striped array of mirrored disks without parity.
Exchange databases that require high I/O performance and very high fault tolerance.
This configuration is a combination of the mirrored disks included in a RAID 0 drive. Each disk in the array is mirrored to guarantee a very high level of fault tolerance. This RAID configuration is becoming the configuration of choice for many organizations. For individual disk sizes larger than 18 GB, you should use RAID 0+1 instead of RAID 5. RAID 0+1 offers the highest performance and very high reliability.
An array of striped disks that have parity.
Exchange databases that require mid-range performance and fault tolerance.
RAID 5 works similarly to RAID 0, but includes a mechanism to write a checksum of the data on each stripe to one of the disks. If one disk in the array fails, the system can reconstruct the data from the remaining hard disks. Note that in a RAID 5 array, multiple small disks perform better than fewer large disks. For example, if you have three 27 GB disks, you can use only 54 GB of storage space. If you instead use nine 9 GB disks, you have 72 GB of storage space. However, the more disks you include in the array, the higher the chance that two disks will break at the same time. This requires you to re-create the RAID drive and restore the data from backup. It is safer to use fewer large disks instead of multiple small disks. As previously mentioned, RAID 0+1 is a recommended alternative to RAID 5 in cases where you have large disk capacities.
Implement a powerful backup solution For servers that store a large amount of data, implement a high-performance backup solution that processes multiple backup operations concurrently. Current backup solutions can achieve data transfer rates of more than 100 GB per hour. If your backup solution is not as fast, consider performing concurrent operations to back up Exchange 2003 in a timely way. You can back up multiple storage groups at the same time in separate backup sessions.
When you implement a backup solution for Exchange 2003, do not rely on the backup tool included in Windows Server 2003 because this basic tool is unable to use the Volume Shadow Copy service. Instead, evaluate non-Microsoft backup solutions that include an Exchange 2003 Volume Shadow Copy service requestor. The Volume Shadow Copy service uses requestors to create shadow copies of Exchange 2003 databases. Based on these shadow copies, the Volume Shadow Copy service in Windows Server 2003 can greatly reduce the time that it takes to back up and restore Exchange 2003. Non-Microsoft backup solutions that are based on the Volume Shadow Copy service infrastructure can also use the shadow copies to almost instantaneously restore one or more databases.
The following sequential steps are performed when you back up Exchange 2003 by using the Volume Shadow Copy service:
The backup program starts a manual or scheduled backup process.
The Volume Shadow Copy service requestor in the backup program sends a command to the Volume Shadow Copy service to take a shadow copy of the selected Exchange 2003 storage groups.
The Volume Shadow Copy service communicates with a Volume Shadow Copy writer component in Exchange 2003 to pause new transactions, to finish current transactions, and to flush all cached data to disk.
The Volume Shadow Copy service communicates with the appropriate storage provider to create a shadow copy of storage volumes that contain the Exchange 2003 storage groups.
The Volume Shadow Copy service informs Exchange 2003 to resume normal operations.
The backup program copies the shadow copies of the storage group databases and logs to the tape backup device.
When the tape copy is completed, the Volume Shadow Copy service requestor in the backup program communicates with the Volume Shadow Copy service to delete the shadow copy.