1298-1300
Page 1298
Table 56.2. continued
| Disk/Tape Storage Media Type | Storage Maximum, Gigabytes | Average Access Time, Milliseconds | Average Write Time, Milliseconds | Media Device Driver Cost |
| DEC TZ-80 tape cartridge | 40GB | 10 S | 333M@min. | $3,500.00 |
| Exabyte EXB-8505 8mm tape | 5-10GB | 67.5 S | 10-35M@min. | $4,125.00 |
| Mirrored Disks Ultra/Wide 40MB@Second | 4GB | 4-9 MS/2 disks | 4-9 MS/2 disks | $995.00 |
| RAID 5 Array Ultra/Wide 40MB@Second | 4GB | 12-16 MS | 12-16 MS | $995.00 |
| Microtek Floptical Compact Disks | 120MB | 60k @Sec | $5,000.00 | |
| SONY minidisk Read/write CD | 120MB | 600k @Sec | $595.00 |
RAID Versus the Speed of Disk Mirroring
Commonly referred to as fault tolerant disk systems, Redundant Arrays of Independent Disks (RAID) and "mirrored disk arrays" both use different RAID levels to provide redundant duplication of data disk storage devices.
In RAID5, when a single disk drive failure occurs, the redundant duplicate media spread across the remaining disks are used to recreate the data that was stored on the now unavailable disk drive. A "scratch pool" contains extra disk drives reserved for each RAID5 Array. In the event that failure occurs to any of the RAID5 Arrays, an extra disk from the scratch pool is temporarily added to redefine that RAID5 stripe.
This recovery of data should be considered complete only after a new disk is installed and mounted onto the file system.
Performance of common RAID, other than RAID Level 1 and 5, is significantly slower as the data from each drive is duplicated among an average of three to five other disk drives. This impacts not only the number of physical read/ writes but degrades CPU and I/O performance across the processor bus as well.
Page 1299
The database implications here are twofold in that file writing is significantly slower due to duplication across three (3) to five (5) disks on average. Another common RAID drawback is that the real-time reads are only possible from the single master disk, and not from the redundant copy.
What are RAID Levels 0_5? Each RAID level offers a different approach to data redundancy and fast access to that data. Fault tolerance, read/write performance, and storage capacity are unique to each RAID level.
- RAID Level 0: Striping of data across multiple disks to increase performance. This increased performance sacrifices fault tolerance, which RAID Level 0 completely lacks.
RAID Level 0 does not dedicate any disks to error checking or disk mirroring. Storage capacity is maximized and the up-front cost is minimal. - RAID Level 1: Commonly referred to as mirroring. Fault tolerance is implemented by creating a twin for each data disk. Read performance is improved but write time may be increased as data must be written twice. This RAID level requires double capacity as 2 gigabytes are required to store each gigabyte of data in fault-tolerant redundancy. This is the most expensive and one of the most popular forms of disk mirroring. Best for use as an operating system disk, commonly referred to as a "system disk" and/or an Oracle source code location.
- RAID Level 2: Error correction is used to recover data instead of data duplication storage on extra disks.
- RAID Level 3: Error correction is implemented at the drive controller hardware and at a parity or duplicate drive, for the most efficient use of large, similar blocks of data. Only one write is allowed to the disk array at one time.
- RAID Level 4: Better performance and less fault tolerance than RAID Level 3. Similar in that reads from multiple drives occur simultaneously and the same storage efficiency as RAID Level 3.
- RAID Level 5: Software striping of data and the backup copy across the disk array for the best performance for security, cost advantages, and Oracle database servers. Boosts performance with simultaneous reads and writes. This is the best and most popular form of RAID fault tolerance.
Mirrored Disk Drives
RAID Levels 1 and 5 enable a noticeable performance increase while simultaneously providing a consistent data mirror, one for one, across each mirrored disk drive.
Performance is optimized by the operating system as queries for data across mirrored disks are split among the duplexed physical media. This optimizes the physical read/write ratio because when two files are accessed from the same disk which has been mirrored, both disks are used to retrieve the data, thus speeding up reads, enabling faster throughput.
Page 1300
The data actually travels from the disks, through the host adapter and finally across the CPU bus for I/O.
Mirroring also allows for separation of a live system into two segments, one of which remains active to the users. This momentarily "breaks" the mirror while data can be copied via any high-speed means to an archive disk or compressed tape for backup and recovery purposes.
The mirror is then "re-silvered" as it is re-attached and rolls forward to update any file server activity since the initial separation from the master disk drive.
This breaking of the mirror procedure momentarily eliminates the second disk mirror recovery procedure from your contingency plan and should therefore only be done after other alternative backups are verified .
Parallel Processor Types, Smart Controller Cards,
and Bus I/O
If the processor is a 32-bit Intel i486 and no more than two processors are connected to a 64-bit bus via the host adapters, then no input/output contention will result, if the MHz rating of the bus is verified to be identical to, or greater than the CPU clock speed rating. This may also be refereed to as a "PCI" or "local" bus.
As other more powerful processors enter the marketplace , specific attention to the architecture, processor type, the number of processors per bus and their maximum throughput rates per clock tick cycle (the MHz rating) must be addressed.
When a single 64-bit processor is connected to a 64-bit bus, having the same clock speed or MHz rating, then no contention or delay of information processing results. If, however, two such 64-bit processors share the same single 64-bit bus with the same clock speed, each individual processor must wait a full cycle while the other processor is active.
This also means that when both processors are active, each processor can only receive input/output half of the time, from the common bus peripheral devices such as disks, tape devices, or network host adapter interfaces.
It is therefore wise to verify that your expansion processor bus is rated "Ultrawide and fast" for 40MB@second throughput, or "fiber channel" 200M@second processor throughput. This will help to eliminate the parallel processor contention for the CPU bus and to allow for the speediest of computational zest.
If the Intel DX processor is a DX2, the processor internal clock speed is doubled or, in the DX4, tripled. This is relative only to the internal CPU operations, NOT the input/output across the common bus. The MHz rating of the bus is actually the speed or bandwidth at which the information input/output travels.
- Reduced Instruction Set Chip: A Reduced Instruction Set Chip (RISC) can actually process efficiently many CPU operations per cycle more than the older Complex
EAN: 2147483647
Pages: 391