Planning Storage Strategies

     The three most important services a file server can provide are security, capacity, and speed. Security is a big topic and covers many aspects of server management, including fault tolerance, fault recovery, network security, and file system security. The overall performance and reliability of a file server is directly related to the performance and reliability of the disk subsystem, which is both the slowest and most likely to fail component in a file server. File servers are often "disk bound," or speed limited by their hard drives because the servers spend so much time accessing files and servicing multiple file requests from users simultaneously.

     Disk striping and mirroring provided by NTFS are software implementations of RAID technology. For the sake of brevity, however, when we refer to RAID, we are refering to RAID implemented in hardware as a hard disk controller.

     Note that cost increases dramatically with with both speed and fault tolerance because RAID adapters are expensive and because replication requires two file servers configured similarly. Also note that all forms of fault tolerance reduce speed somewhat.

Storage Planning Matrix
	ONE DAY DATA LOSS	NO DATA LOSS	NO DOWN TIME
NORMAL SPEED	Tape Backup	NTFS mirror	Replication + RAID 1
FAST	NTFS Stripe Set	NTFS Stripe Set with Parity	NTFS Replication + NTFS Stripe Set with Parity
VERY FAST	RAID Level 0	RAID Level 5	Replication + RAID 5

How Much Disk Space Is Enough?

     If you implement mirroring, duplexing, or RAID Level 1, you will need twice as much disk space. If you implement striping with parity or RAID Level 5, you will need one additional hard disk to compensate for space taken by parity information.

---

Storage Options

Partitions and Volumes

     Hard disks, like file cabinets, have a well-defined structure of partitions and volumes that allows the various file systems to coexist peacefully on the same disk.

• Partitions
  
  
  Hard disks are divided into partitions. Each partition contains one volume maintained by any file system. Volumes are partitions with file systems in them. When you format a partition, you are creating a volume. Partitions allow one hard disk to appear as many hard disks. Changing the partition of a disk destroys all the data in the partitions being changed.
  
  Master Boot Record
  
  
  The master boot record is the first portion of data on a hard disk. It is reserved for the BIOS bootstrap routine, which contains low-level code that redirects the loading of the operating system to the partition marked as active. This indirect method allows up to four bootable partitions to exist on the same disk and, therefore, multiple operating systems.
  
  
  The Partition Table
  
  
  The partition table is the index of partitions used by the bootstrap routine to identify distinct portions of the disk. File systems can create volumes only within partitions. Partition entries in the partition table contain a simple starting position on the disk and the length (or number of clusters) of the partition.
  
  
  
• Volumes
  
  
  A volume is the structure imposed upon a partition to allow files to be indexed. A volume contains the table of contents of the disk, called the root directory. The volume also maintains allocation tables that show what space in the volume is occupied, thus ensuring that no files overwrites another by being stored in the same location. Some file systems, such as Windows NT, can create logical partitions containing volumes that extend across more than one physical partition. Volume sets and stripe sets are both examples of logical partitions that extend across multiple physical partitions.
  
  Drive Letters
  
  
  Drive letters are assigned to volumes for easy identification when referring to the exact location of a file. MS-DOS assigned drive letters as it found devices when booting, starting with C for physical disks. Windows NT allows the administrator to assign drive letters as necessary.
  
  
  Volume Sets
  
  
  Up to 32 partitions can be combined into one volume known as a volume set. That volume set receives a single drive letter and can be treated as a single partition for the purposes of formatting and file system use by the operating system. The partitions combined into a volume set do not have to be the same size or on the same drive or the same type of drive (SCSI, ESDI, or IDE).
  
  When data is stored to a volume set, the data is stored to the partitions one partition at a time. When one partition is full, the next partition is used.
  
  Volume sets do not provide fault tolerance to your file system; they do not implement fault or error recovery information beyond that already provided by the file system. In fact, if a fault occurs with any one partition or hard disk in a volume set, the whole volume set may be unusable.
  
  
  Stripe Sets
  
  
  A stripe set is like a volume set except that all the stripe set's partitions must be on different drives and should be the same size. In a stripe set, the data is stored evenly across all the partitions of the stripe set instead of to one partition at a time.
  
  Because of the way data is stored to a stripe set, stripe sets can be much faster than volume sets or even single partitions. A stripe set with multiple disk drive controllers (or sophisticated disk drive controllers that con communicate with several disks drives at a time) can transfer more information than a single drive controller connected to a single drive can transfer.
  
  
  Stripe Sets with Parity
  
  
  Stripe sets with parity add an extra twist to stripe sets that make them much more fault tolerant than normal volumes. A stripe set with parity reserves an amount of space equal to the size of one hard disk drive for parity information. The parity information is used to re-create information in the event that one of the hard disks in the stripe set fails.
  
  The data in the parity blocks stored on the other disk will be equal to the data on the failed drive when run through the mathematical operation that created the parity information. In effect, when a single drive fails, the other drives can pick up the slack until the first drive is replaced. Because drives tend to fail one at a time, in theory your server need never go down due to hard disk failure; the administrator will be able to replace disks as they go bad without affecting the data on the disk set.
  
  
  
  
  

---

Fault Tolerance and Fault Recovery

     Fault tolerance encompases all measures taken to protect data and preserve accessibility in the presence of a hardware failure. Tape backup is actually not fault tolerance but fault recovery. The difference between the two is that fault tolerance preserves accessibility to data online, while fault recovery makes the recovery of data possible but does not preserve accessibility.

The minimum level of fault tolerance for servers include the following:

• Tape Backup
  
  
  Tape backup (also called archiving) is the minimum level of fault tolerance that should exist for all servers. If tape backup is the only fault tolerance system in use, the worst-case scenario for data loss is the amount of information created or changed from the time of the last backup until the server failed. Tape backup systems are most effective when backups occur nightly. This schedule makes the worst-case data loss equal to one work day, which is acceptable for most organizations. When this degree of loss it not acceptable, other fault tolerance systems should be used because the load created by a tape backup operation can be excessive during working hours.
  
  Tape Backup Systems
  
  
  Tape backup systems can be implemented either locally or over the network.
  
  Local Tape Backup
  
  
  Local tape backup refers to placing a tape drive in every file server and copying files from that server's hard disk drive to its tape drive.
  
  Benefits of Local Tape Backup
  
  
  • Data is not transmitted over the network.
    
  • Backups can occur during periods of use without causing network load.
    
  • Data closely matches the structure of the hard disk drives.
    
  • Easy access to the Registry.
    
  • Restoration possible without a functioning network.
    
  • Works with Windows NT tape backup software.
    
  
  Drawbacks of Local Tape Backup
  
  
  • Additional administrative burden for larger networks.
    
  • Additional hardware cost.
    
  
  
  Networked Tape Backup
  
  
  Networked tape backup refers to copying files from every server on the network to a single tape server that copies all the files to tape in a central location.
  
  Networked tape backup is generally used in large multi-server networks, especially when servers are distributed physically throughout the organization. A common arrangement is for network administrators to use their own administration workstations as a tape server.
  
  Benefits of Networked Tape Backup
  
  
  • One tape server can back up multiple servers.
    
  • Less administrative work for larger networks.
    
  
  Drawbacks of Networked Tape Backup
  
  
  • Massive archiving operations can take longer than overnight, thus causing a network load during operating hours.
    
  • All files are transmitted over the network, creating a possible security loophole for extremely sensitive information.
    
  • Single point of archive failure - if the tape server fails, no archiving occurs.
    
  • Network failure preempts backup operation.
    
  • Additional software costs.
    
  
  
  The type of tape system you install depends upon your needs. However, because of the additional software costs and the reduced fault tolerance of networked tape backups, it is recommended using local tape backups where possible.
  
  
  Tape Backup Styles
  
  
  The two most common tape backup methods are full system backups and incremental backups.
  
  Full System Backups
  
  
  Full system backups copy all the server hard disk drives on a daily basis. This method is simple to setup and easy to restore.
  
  Unfortunately, full system backups requre tape drives that are larger than all the data stored on your server and fast enough to transfer all that data during idle periods.
  
  Full system backups also require the use of tape drives and tapes that are larger than the entire capacity of the network. These devices are very expensive, as are the tapes they use.
  
  
  Incremental Backups
  
  
  Incremental backups are an alternative to using large expensive tapes and tape drives. An incremental backup uses a file attribute bit called the archive bit to mark files as having been backed up.
  
  When the file system modifies or creates a file, the archive bit is set, indicating that the file needs to be backed up. Tape backup programs clear the archive bit when the file is backed up to indicate that it no longer needs to be archived. An incremental backup simply copies those files on the file server that have the archive bit set.
  
  Generally, a full system backup is performed when free time is available on the network, like weekends, and someone is on hand to change tapes if the tape device capacity is smaller than the capacity of the server. Then daily incremental backups are performed over night.
  
  Incremental backups are more administrative hassle than full system backups. The location of a single server file is usually impossible to find unless the tape backup software keeps a database of files and the tape volumes they've been written to. Restoring a system completely requires restoring the contents of a number of tapes.
  
  
  Tape Rotation
  
  
  Tape rotation refers to the reuse of tape media. You can reuse tapes on a rotating basis when the tapes are so far out-of-date that you no longer care about the data.
  
  For some organizations, that date might be yesterday, and they may rotate between only two tape sets. Other organizations will keep backup tapes for up to a month. Most organizations rarely need to go longer than a week with incremental backup tapes, which is the rotation recommended by most archive specialists and tape backup companies.
  
  To implement a fail-safe network, you must keep a recent full system copy off-site.
  
  
  
  
• Mirroring
  
  
  Disk mirroring is the process of performing every write operation in a file server to two identical disks. In the event of a primary disk failure, the secondary disk is ready to take over instantly as an exact copy.
  
  Disk mirroring and duplexing is RAID level 1. Microsoft uses the terms mirroring and duplexing to refer to RAID level 1 as performed by Windows NT.
  
  Disk mirroring is the most basic form of fault tolerance. It preserves accessibility in the event of a hard disk failure. Disk mirroring preserves accessibility only in the event of hard disk failure.
  
  
  
  
• Duplexing
  
  
  Disk duplexing is related to disk mirroring but adds another level of fault tolerance: the hard disk controller. Disk duplexing stipulates that two hard disk controllers are used, with two disks attached, to perform the mirroring function. If one of the hard disk controllers fail, the system will fall over to the other controller, preserving accessibility. Disk mirroring is not tolerant of hard disk controller failure.
  
  Hard disk controller failure is far less common than it once was, so disk duplexing is not as important as it once was. Windows NT does not differentiate between disk mirroring and disk duplexing. To duplex two disks on different controllers, simply install the necessary hardware and mirror the partitions as you would for disk mirroring.
  
  
  
  
• Striping with Parity
  
  
  Striping with parity is a fault tolerance measure added to shore up disk striping, which alone actually reduces the fault tolerance of a disk subsystem by multiplying the probability of disk subsystem failure by the number of disks in the stripe set. Adding a parity stripe to a stripe set allows any one disk in the stripe set to fail without denying access to the disk set becuase the data on the failed drive can be calculated from the parity stripe, which exists across the remaining drives.
  
  Striping with parity is the Windows NT term for its software implementation of RAID level 5.
  
  
  
  
• Hardware RAID
  
  
  RAID refers to the simultaneous use of multiple hard disk drives to increase the fault tolerance and / or speed of a disk subsystem.
  
  Hardware RAID controllers are, in effect, special hard disk controllers that control all the hard disk drives independantly on the same bus and calculate or derive the stripe data. RAID controllers can improve performance significantly because they are not distracted by the other processing chores of the CPU must perform. Think of RAID controllers as disk coprocessors.
  
  Although the term RAID also applies to software-only systems like those implemented by Windows NT, people usually mean hardware RAID controllers when they use the term.
  
  Windows NT implements the following RAID levels in software:
  
  
  • RAID Level 0 (striping)
    
    
    RAID level 0 is not implemented for fault tolerance, as it actually decreases the fault tolerance of the disk subsystem. Rather RAID level 0 is the ultimate in disk performance: parallel drives without the performance penalty of deriving a parity stripe. RAID level 0 should not be used without another fault tolerance or fault recovery system in place.
    
    
  • RAID Level 1 (mirroring)
    
    
    RAID level 1 is disk mirroring. Many less expensive disk controllers offer RAID level 1 as an option, which frees Windows NT from dealing with the overhead of providing the mirrored disk functions. Unfortunately, RAID level 1 hardware controllers cannot implement disk duplexing, which requires more than one controller and, therefore, must be controlled by Windows NT.
    
    
  • RAID Level 5 (striping with parity)
    
    
    RAID level 5 is the most commonly implemented form of hardware RAID: parallel drives with parity stripe. The level offers a good mix of speed and safety.
    
    RAID controllers are both very fast and fault tolerant for the following reasons:
    
    • Their only purpose is to split data transferred to and from the computer into the optimal size for transfer to the disks attached.
      
    • They usually have robust caches.
      
    • They use a dedicated microprocessor to calculate and derive the stripe set.
      
    
    Unfortunately, they are also very expensive. The price of a RAID controller can account for one-third of the total cost of a server.
    
    
    
  
  
• Replication
  
  
  Replication is the ultimate form of fault tolerance. Replication provides a backup in the event of any hardware failure on the primary server by copying data between the primary and the backup server continually.
  
  Any component failure is protected by replication, not just disk or disk controller failure as protected by the fault tolerance systems discussed earlier. Replication can be the only fault toerance system you implement, replacing disk mirroring, duplexing, or RAID. Because both servers can be operational servers running Windows NT, replication can be performed very inexpensively IF you already have two servers.
  
  The directory replicator service replicates files only after they are closed. As a result, files that are constantly open, such as databases, will not be replicated until every user has closed them, which may not occur in the case of database files. Therefore, you should consider using a true real-time replication package from a third party (Vinca and Octopus) if you intend to rely on server replication for fault tolerance.