From disk subsystems to file formats and permissions, here are some best practices to follow when implementing Windows NT.

Windows NT Best Practices

From disk subsystems to file formats and permissions, here are some best practices to follow when implementing Windows NT.

As you read this, another year is already underway. The installed base of Windows NT has grown dramatically the past couple of years. There’s a chance that this’ll be the year of the long-awaited Windows 2000 (Windows NT 5.0). Given those facts, I thought I’d use this column to present several “best practices” that you should implement with any individual Windows NT machine or, in some cases, the Windows NT network services.

Hardware Check

The place to start with Windows NT is always the hardware. Make sure that all of the components of your hardware platform are on the Hardware Compatibility List (HCL). As obvious as this is, it’s often overlooked and the results are frustrating. This is particularly important if you’re building or buying custom-built machines. In these cases, make sure to check that components such as the motherboard and controllers are on the list. As a second protection, write into your purchase order that the hardware as configured will reliably support Windows NT. The HCL changes over time, so be sure to review it frequently. (Microsoft posts it online at Finally, remember that this list will change dramatically for Windows 2000.

The next place to check is memory. The Windows NT marketing literature states that the minimum RAM configuration should be 16M (and beta 2 of NT 5.0 requires a minimum of 500M of free disk space). Although this number will help you get certified, we all know it’s ridiculous. Windows NT isn’t the OS of choice if your main goal is to use as few hardware resources as possible in the process. Windows NT loves memory. The more you add, the more it’ll allocate to processes and cache. The minimum kernel and executive-services memory footprint of NT Workstation is 24M—and this is before any applications are loaded. If you have anything less, you’re going to page code that’s providing OS services. Today, realistic minimum RAM on an NT workstation is 32M, and I could build a strong case to make it 64M. Don’t be surprised when Windows 2000 comes out and you a hear recommendation of 128M of memory for a workstation (and beta 2 of NT 5.0 requires a minimum of 500M of free disk space).

A related issue is the paging file used for virtual memory management. You want to make sure that the page file isn’t on the same drive as the system partition. Microsoft often recommends that the file not be on the same partition, but you really want it off the physical drive entirely. One of the keys to disk access is that you’re dealing with spindles, not just partitions. A partition is just a logical construct on the drive. When data resides on different partitions, the controller and head must deal with another layer of abstraction. I also recommend that you avoid multiple partitions on the same disk and use the entire disk for each partition. Furthermore, you should spread the page file over several spindles to create a stripe environment for the file. My best practice recommendation for page files, however, is not to page. Friends don’t let friends page memory to disk. If you follow the best practices in terms of adding enough memory, you’ll avoid most paging in the first place.

Another aspect of disk consideration is fault tolerance. NT has some software-based disk management, including extended partitions, simple striping, disk mirroring, and striping with parity. First, don’t use extended partitions. They’re useful in a pinch if, for some reason, you haven’t been monitoring disk consumption and you need some more space on the fly. But don’t use them on an ongoing basis. If any of the underlying drives in the partition fail, you’ll lose all your data and will have to rely on a backup.

The best place to create disk fault-tolerant subsystems is in hardware. For example, if you create a stripe set with parity within NT, you have a couple of things working against you.

First, the parity algorithm must be computed with the CPU that’s also trying to execute application code. This work is better offloaded to the hardware controller in a RAID box below the operating system.

Second, if one drive fails, the recovery process requires that you reboot the server. With the proper hardware RAID system, you can remove the bad drive and replace it with the new drive without rebooting the system. You really should monitor the health of the drives (usually this software comes with the server hardware), because you might not actually notice the performance decrease after a drive fails. When a second drive fails in the RAID array, you’re sunk. The software RAID is an interesting aspect of NT, but the best way to implement disk fault-tolerant systems is with external hardware.

Storing Files and Giving Access

Once you’ve worked out your drive subsystems, you need to choose a format in which to store your files. With Windows NT 4.x, you can choose between FAT and NTFS. The only time to use FAT is if you need to build a dual-boot system.

First of all, don’t build dual-boot systems unless you’re using them for testing purposes. NTFS, on the other hand, offers many features that just aren’t available under FAT, including hotfix capabilities and smaller sector size in large disks. Never use FAT to change sector sizes, such as particular record sizes. FAT also doesn’t offer NTFS transaction rollback capabilities, local security. Meanwhile, NTFS has transaction functionality whereby the file system rolls back to a stable state. NTFS can also audit disk resources, a capability that FAT lacks. Essentially, FAT in Windows NT is a migration tool, and it should be treated as such.

Once you choose NTFS as the file format, you introduce new problems. When you format a disk with NTFS, for example, the default permission is Everyone—Full Control. Several issues surrounding this option can cause problems down the road. One is that with Full Control every user will have access to any files created or copied to this directory or any directory created below the root that hasn’t had its permissions modified. They also will be able to change the permissions of the files and directories. Change the Full Control to Change so that all directories that are created afterward will inherit this permission. This creates a better starting point for specifying more granular permissions as you move forward.

The other issue is with the group Everyone. The SID S-1-1-0 behind this group exists on all NT systems. This leaves a significant security hole in your system. SP3 added a group to the system called Authenticated Users. This group only includes users that actually exist in that particular database, because the SID behind it is unique to your installation. The other group you want to make sure has access to each entire drive is Administrators and System with Full Control. This allows the operating system to have access to the drive for such things as the paging file, if you happen to place it on a particular drive.

When setting up permission on any resource, the accepted practice is to use the following guidelines within Domains. First, create a local group on the machine where the resource, such as a directory or printer share, resides. Give the permissions that you want for each individual group. Create a Global Group and add the users to whom you want to allocate the resource. Then add the Global Group into the Local Group to assign permissions. This may sound like a lot of work to give permissions that could easily be given directly to individual accounts. However, this method allows system administration to scale across a large domain and, more important, across multiple domains. Always use groups to manage accounts and permissions.

Invaluable Extras

Everytime you make a significant change to the system, run RDISK /S and update the Emergency Repair Disk. This simple but valuable tool allows the administrator to replace corrupted files on the system. If you don’t update the repair disk frequently, you can set the system back to an earlier configuration, and you also might lose important information. The /S switch also saves part of the Security Accounts Manager database from the Registry to the \WINNT\REPAIR directory. You can also use this directory as the source for repairing the system instead of the ER disk that’s normally used. The directory can also contain information that won’t fit on the ER disk.

Finally, use Performance Monitor. This tool has the distinction of being the most talked-about but seemingly least-used utility. Its ability to track performance over time with the logging feature lets you create a baseline for your system. You can then use Performance Monitor’s alert capabilities to notify the administrator when system performance drops below your norm and before a component actually fails.

These are some of the best practices that I think are important. Undoubtedly there are more good ideas out there. I encourage you to send me your best practices. I’ll compile them and publish them for the benefit of others rolling out new systems.

About the Author

Michael Chacon, MCSE, MCT, is a directory services architect, focusing on the business and technical issues surrounding identity management in the enterprise. He is the co-author of new book coming from Sybex Publishing that covers the MCSA's 70-218 exam.

comments powered by Disqus
Most   Popular