What system writes data on 2 or more disks simultaneously?

Configuring Disk Mirroring

The Nokia disk mirroring feature [RAID Level 1] protects against downtime in the event of a hard-disk failure on your Nokia appliance [for platforms that support the feature], but you must have a second hard disk drive installed on your appliance.

Disk mirroring allows you to configure a mirror set composed of a source hard disk drive and a mirror hard disk drive that uses Nokia Network Voyager. The source drive is where IPSO is installed. When configuring a mirror set, the hard disk drives are synchronized [the source hard disk drive is fully copied to the mirror hard disk drive] and all new data written to the source drive is also written to the mirror hard disk drive. If the source hard disk drive fails, the mirror hard disk drive automatically replaces your source hard disk drive without an interruption of service on the Nokia appliance.

The source and mirror hard disk drives can be warm-swapped on appliances other than the IP500 series, meaning you can replace a failed hard disk drive without shutting down your Nokia appliance. You can also monitor the status of a mirror set, the synchronization time, and the system log entries. Figure 4.9 shows the Disk Mirroring Configuration page.

To create a mirror set or enable disk mirroring, perform the following steps:

In the system tree, click Configuration | System Configuration | Disk Mirroring.

Under Mirror Set, select Create. [The source and mirror hard disk drives should have identical geometries.]

Click Apply.

A message indicating that a mirror set was created appears. Numbers indicate which hard disk drive is the source and which is the mirror, and that mirror synchronization is in progress.

The synchronization percent value in the Mirror Set table indicates the percentage of synchronization zones that are copied from the source to the mirror disk. A zone is equivalent to contiguous disk sectors. When all zones are copied to the mirror disk, the synchronization percent value is 100. For a 20GB disk, the synchronization time is approximately 20 to 30 minutes.

To delete or disable disk mirroring, on the Disk Mirroring page, next to the item for removal or deactivation, select Delete. Click Apply and then click Save to make the changes permanent.

You can only delete a mirror set that is 100-percent synchronized.

RAID-1

RAID-1 is more commonly referred to as disk mirroring or disk duplexing, and involves the use of two disk drives. It can be implemented through hardware, or through software using the disk management features built into Windows Server 2003. A RAID-1 volume created using the Windows software is referred to as a mirrored volume. With RAID-1, each drive is a “mirror image” of the other. Disk duplexing refers to mirroring when each drive is attached to its own separate controller or communication channel.

Note

Fault tolerant software RAID [RAID-1 and RAID-5] can only be implemented in Windows Server 2003 on disks that have been converted from basic to dynamic status. Dynamic disks offer more flexibility in management options, but cannot be accessed locally by operating systems prior to Windows 2000. Because it is recommended that production servers not be installed in a dual boot configuration, this should not be an issue with your DCs.

Duplexing can improve both read and write performance. In a duplexed environment, both disks can be read from or written to simultaneously. In a mirrored environment, only one of the drives can be read from or written to at a time. Because data must be written to both drives, write performance can be slightly degraded in a mirrored environment. For the reasons stated previously, this is less true in a duplexed environment. Data can be read from either disk in a mirrored environment and from both disks simultaneously in a duplexed environment, thereby increasing read performance. Fault tolerance is enhanced because if one of the drives malfunctions in a mirrored environment, or one of the drives, controllers, or communications channels malfunctions in a duplex environment, the server remains functional using the other hardware in mirrored configuration.

A duplexed volume appears to the operating system as a mirrored volume. The difference is in the hardware configuration; the software does not recognize any difference between mirrored and duplexed volumes. If a disk in a RAID-1 volume fails, you should break the mirror, add a new disk, and recreate the mirror to restore fault tolerance.

Test Day Tip

When using software RAID, the Windows Server 2003 boot and system files can only exist on a RAID-1 [mirrored or duplexed] volume. You cannot use a Windows RAID-5 volume for disks that hold the boot or system volumes. This limitation does not exist when using hardware RAID, in addition, hardware RAID offers higher performance and is preferred if the extra cost is acceptable to the organization.

RAID1

RAID1 is another fundamental RAID level that keeps and maintains the exact same content of a disk drive on one or more additional disks. RAID1 is often referred to as disk mirroring, and carries the great advantage of duplicating the contents of a given disk drive onto another drive. Should the initial drive fail [or any member of the RAID1 set], operations can still continue by getting data on other valid members of the RAID1 set [Figure 2-6].

The advantage of RAID1 is the high level of data redundancy and availability. In fact, it carries other advantages, especially in the area of point-in-time backup operations, which is discussed later. The disadvantage of RAID1 is that you lose capacity. If you combine two 72 GB disk drives in a RAID1 array, the resulting raw capacity is 72 GB. You've just lost the capacity of the mirroring drive. The same thing applies if you arrange three drives together: The resulting capacity is still the size of one disk.

From a request rate perspective, the deal is somewhat different depending on whether you mostly read from the volume or write to it. In read operations, the RAID controller can read from either of the two volumes [in a two-volume RAID1 set] and yield a performance level of 200 I/O per second, given that your individual disk drives can sustain 100 I/O per second.

When it comes to write operations, the data have to be written to both volumes at the same time, and in some cases, the resulting response time is dependent on the slowest drive to complete the operation. Therefore, a two-disk RAID1 set has a write performance of 100 I/O per second. This is the first instance in which we see that the read/write ratio gains importance in the actual performance of the volume. When we compare the RAID levels, you will see that the R/W ratio can weigh significantly in choosing the most appropriate level for your environment.

Figure 2-6 describes the block layout of a RAID1 volume. Note that for two disks, the contents are the same. As mentioned earlier, you could combine more than two drives in a RAID1 set. A three-member RAID1 volume can handle the loss of two disk drives [really, the probability of losing two drives in a three-disk RAID1 set is quite small]. The other advantage is that you can “break” the mirror [without necessarily affecting the current operation of the drive, although it's best to do this with a “quiesced” volume], and in a matter of seconds you have a duplicate copy of your original volume. This is a fast way to back up data. Given the decreasing price of storage these days [in terms of $/GB], this can be an interesting approach, and we'll see that in high-end environments, this becomes a sound data duplication strategy for backups of the replica database of Exchange 2007 CCR passive nodes.

Striping and RAID

Striping, or RAID, is the ability to spread the I/O requests across multiple disks in parallel. RAID is an acronym, which stands for Redundant Arrays of Inexpensive Disks, and the basic concept of RAID is to combine small, inexpensive disks to yield improved performance and reliability and have them appear to the server as a single, larger disk. We touched upon this problem earlier in the chapter when we looked at adding more disks into our warehouse architecture and where RAID provides a solution for better throughput and reliability by using multiple disks.

RAID can be described as levels, which progress from 0 to 10. As you progress through the levels, you get different performance and reliability. For example, rather than storing all of the data on one file on a single disk, the data is striped across multiple disks, which yields better performance during reading, because all of the disks potentially get read in parallel rather than just one.

Striping is the process where the disks are split into separate sections and the sections are utilized round-robin style on each disk. So, for example, if your stripe size is 1M, then the first 1M of your file is written to disk 1, the second 1M stripe is written to disk 2, and so on. This splitting of the data into stripes that are written to different disks is done either at the byte level or the block level and is performed automatically and completely transparently to the applications, whether it is a file for a word processor or a sophisticated database. Striping has the effect of reducing the contention for disk areas and improving I/O throughput to the server. Figure 3.8 shows a four disk RAID, where the stripe width is four and stripe size is the size of the individual amounts of data written to each disk.

The process of mirroring the data across multiple disks involves duplicating the information that we write to one disk by writing a complete copy to a second disk so we always have a copy of our data.

The RAID levels are:

RAID 0: striping [i.e., striping our files so that each successive stripe occurs on a different disk in a group of disks]

RAID 1: disk mirroring. Where an exact copy of each disk is automatically maintained. If one disk fails, then the data is available on the other disk until a new disk can be inserted and the copy made while the system is on-line. The cost of this is that it doubles the number of disks that are required.

RAID 3: striping with parity at the byte level. Uses an additional parity disk with other data, where parity data is simply a calculated numerical value from using the data on the actual data disks, so that if any one of the data disks fail, then the remaining data in conjunction with the parity data can rebuild the lost data. The parity information is on a separate, dedicated disk, for example, five data disks and one parity disk.

RAID 4: block data striping with parity. Same as RAID 3 but working at the block and not byte level.

RAID 5: block striping rotated parity. Both the parity and the data are striped across the disks. This removes the need for a dedicated parity disk, because the parity information is rotated across all disks. Read performance is better, but write performance can be poor. At least three disks are needed for a RAID 5 array.

RAID 10: [also known as 0+1]: mirrored stripe sets providing better security from the mirroring and better performance from the striping.

RAID can be implemented either in hardware, via a RAID controller for the disks, or in software, via a tool called a logical volume manager [LVM]. An LVM is a piece of software that could be provided either as part of the operating system, from a third-party vendor [e.g., Veritas], or from the disk storage vendor [e.g., from EMC]. The LVM combines areas on separate disks and makes them appear as a single unit to the operating system, and in the process it provides software striping and mirroring. Performing RAID in hardware may be easier, but there may be the disadvantage that it can be more difficult to change at a later stage—for example, if more disks are to be added; performing RAID in software can take advantage of a number of powerful volume managers available. Regardless of the mechanism, the I/O request from the server is transparently split by the controller or LVM into separate I/Os for the individual disks.

From a performance perspective, the DBA and systems administrator must carefully consider not just the type of RAID that they wish to employ and on which disks, but they must also consider items such as the RAID stripe size [i.e., the byte size of each stripe on the disks] and the number of disks in the stripe set.

A high-level perspective of the operations needed to use a volume manager to prepare the disks for Oracle can be seen in the following text. Different hardware vendors will have their own volume managers and there are third-party ones, and in addition, different operating systems may have a slightly different approach, so the steps listed here are intended to be very generic. The DBA and the system administrator must:

Create or initialize the physical volumes. Define the disk as a physical volume and write certain configuration information to it.

Create the physical volume groups. This is the process where the physical volumes are collected together to create the physical volume group. A physical volume can only belong to one volume group, but a volume group may contain many physical volumes. During this step the physical volume is partitioned into units of space called physical extents: These are the smallest units of storage used on the disk, and an extent is a contiguous section of the disk.

Create logical volumes in the volume groups. This is the step when striping and mirroring are implemented.

Create the file systems on the logical volumes. The file system is the structure on the disk necessary for storing and accessing the data. Without the file system it is not possible to create directories and files. For example, the file systems can be FAT32 or NTFS, which will be familiar to Windows users, or Linux Ext2, Linux Ext 3, and Linux Swap, which will be familiar to Linux users.

Mount the file systems. Make the file systems known and usable to the operating system.

Then, finally, use the file systems for the database files [i.e., create the database tablespaces using files on these file systems].

8.6.2.1 Hardware Considerations for Data Vault Layer

Because the Data Vault is the permanent storage for the enterprise data, the data needs to be secured against hardware faults. The first set of options is the RAID setup used for the physical drive. The best economical options should be either RAID-1 for disk mirroring or RAID-5 for adding parity bits to the data written to disk. The second option is often favored because it presents a compromise between available disk space and fault tolerance. Keep in mind that the Data Vault layer stores the history of all enterprise data that is in the data warehouse. Therefore, the size of the database could become very large.

While RAID-1 and RAID-5 are not the fastest available RAID options, it is worth favoring them for the Data Vault layer, due to the added fault tolerance of the physical disk. If one disk fails, it might affect the availability of the data warehouse [because the RAID array might be rebuilt], but the data is not necessarily at risk. However, if more than one disk fails at the same time, this is not the case any longer. In such a case, for example during a large-scale disaster that affects the whole server infrastructure, the collected enterprise data is still at risk.

To cope with such a fatal scenario, it is possible to replicate the data warehouse between multiple data centers. In this case, the data centers are usually geographically distributed and connected via a network [either a Virtual Private Network [VPN] over the Internet or a private network using satellite connections, etc.]. In order to make this scenario work, the network connection has to be fast enough to replicate all incoming data. This scenario might be required if multiple geographically distributed enterprise data warehouses [EDWs] should be made available around the world and a central EDW is not sufficient. Note, however, that the requirements for replicated data warehouses are high.

Developing your risk mitigation strategy

The steps in developing your risk mitigation strategy are as follows:

Gather your recovery data.

Compare cost, capability, and service levels of options in each category.

Determine if the options remaining are risk acceptance, avoidance, limitation, or transference and which, if any, are more desirable.

Select the option or options that best meet your company’s needs.

Now that you’ve gathered these data on various recovery options, you can review them in relation to cost, capabilities, and service levels.

These data can now be compiled into a document in whatever format is suitable for your needs. Some people like to use a grid or matrix, and others prefer an outline format. The key is to create a highly usable document that delineates the choices you’ve made. Let’s look at two examples. In our first sample, we look at a small segment of data that might be included with regard to backups. This uses a grid or matrix style, and it should give you an idea about what data to include and how you might approach it. In our second sample, we’ll use text without a grid, so you can compare which method might work better for you.

Warning When Using Data Destruction Tools

From the previous section, we can note that software antirecovery tools can offer a high level of security for personal users and companies; however, if you are dealing with top secret data, physical destruction is still the safest solution. To achieve the maximum security possible, it is important to be aware of the following issues when using wiping software tools:

If you are deleting data that is scattered across many disk drives [RAID or mirroring disks], make sure to wipe all other locations to avoid leaving copies in other places.

Some files [like MS Office® documents] can have a previous version file stored somewhere other than the main file location that has been erased. To counter this issue, delete all previous restore points and file history. More details to follow.

Make sure to delete all instances of a file; for example, a user may destroy a specific file and forget other copies in other locations.

Hibernation and virtual memory [page files] files can contain IM chat, browser history, encryption passwords, decrypted documents, and other important data.

RAM memory can be captured from a running PC, resulting in recovering important files and even passwords from it.

Windows® by default stores thumbnails of graphics files and certain documents and movie files in the thumbnail cache file. Although thumbnail images are small, they still give sufficient information of any prior existence of the deleted file[s].

Volume shadow copies and file history may contain copies of your deleted files.

Windows event logs and registry can contain information about your PC usage. Such logs can give important evidence of your previous usage of Windows®. Registry antiforensics will be thoroughly discussed later.

As we note, data can be scattered in different locations on our disk drive; deleting a file and using a shredding tool may not be enough alone to securely destroy important data.

Disks

No one will discount the importance of RAM, but it very important to note that the most common hardware bottleneck for database systems is related to disk input and output [I/O]. That might seem somewhat intuitive, because after all, a database is about its data. However, even in simple data transactions, many more files are involved than the files that contain the source data, and all of them are stored and retrieved from the disk system. Among the most important files you should be aware of when trying to optimize your database server are these:

Table files

Nonclustered index files

Transaction log files

Full-text indexes

TempDB database files

Windows operating system files

Windows paging files

One of the most fundamental things you need to choose when you first start optimizing your database server is the type of disk system to use. Most often, the only appropriate choice is a redundant array of independent disks [RAID]. Defined in 1987, RAID not only provides better performance, but it offers fault tolerance. When choosing a RAID solution, you need to balance your needs for data redundancy, speed, and cost.