Lesson 09

---

Chapter 1

Introduction

You’ll remember from the last lesson that a networked PC has four significant functions. First, it must connect to the network. Second, if the PC shares data, the PC needs to protect that shared data by creating more than one copy of the data. Third, and again only if the PC shares data, it needs specialized hardware that enables it to share the data as quickly as possible. The fourth and last function unique to a network PC is reliability: the shared resources of the network must be available whenever another system accesses them.

I covered the connectivity issues such as connectors, cable types, NICs, and modems. In this lesson I’m going to talk about data protection, speed issues, and reliability in networks and servers.

Chapter 2

Protection of Data

The single most important part of most networks is the shared data. The main motivation for building networks, after all, is the ability for many users to access shared data. This shared data might be as trivial as pre-made forms, or as critical as accounts receivable. The sudden loss of all data from a network would cripple most organizations; computers can be replaced and new employees hired, but the data is what keeps most organizations functioning.

Certainly, any good network must include a solid backup plan, but restoring backups takes time and effort. Unless the data is being continually backed up, the backups will always be a little dated. For this reason, restoring data from backups should be looked upon as a last resort option.

Businesses have failed after the loss of data, even with relatively good backups. The shared data of a network, therefore, should have better protection than the fallback of laboriously having to restore potentially dated backups! A good network must have a method of protecting data such that if a hard drive fails, a network technician can bring the data instantly, or at least very quickly, back online. This requires some sort of instant backup or automatic copy of the data stored on a second drive.

OK, so you need to come up with a way to make data redundant on the serving system. How do you do this? Well, first of all, you could install some fancy hard drive controller that reads and writes data to two hard drives simultaneously, as shown in Figure 9-1. The data on each drive would always be identical; one drive would be the primary drive and the other drive, called the mirror drive, would not be used unless the primary drive failed. This process of reading and writing data at the same time to two drives is called drive mirroring.

Fig. 09-01. Mirrored drives

If you really want to make data safe, you can use two separate controllers, one for each drive, as shown in Figure 9-2. With two drives, each on a separate controller, the system will continue to operate, even if the primary drive’s controller stops working. This super-drive mirroring technique is called drive duplexing. Drive duplexing is also much faster than drive mirroring since one controller does not write each piece of data twice.

Fig. 09-02. Duplexing drives

Even though drive duplexing is faster than drive mirroring, they are both slower than the classic one drive/one controller setup. The third and most common way to create redundant data is by a method called disk striping with parity.

Disk striping, without parity, means to spread the data among two or more drives. Disk striping by itself provides no redundancy. If you save a small Microsoft Word file across two striped drives, for example, the file is split into two pieces; half of the pieces go on one drive and half go on the other.

Fig. 09-03. Disk striping

The one and only advantage of disk striping is speed—it is a very fast way to read and write to hard drives. But if either drive fails, all data is lost. Regular disk striping is not something you ever want to do—unless you simply don’t care about data. Nobody does disk striping!

Disk striping with parity, in contrast, protects data. Disk striping with parity adds an extra drive, called a parity drive, that stores information that can be used to rebuild data should one of the data drives fail.

Let’s look at that same Microsoft Word again. The data is still stored on the two data drives, but this time a calculation is done on the data from each equivalent location on the data drives to create parity information on the parity drive. This parity data is created by a simple, but very accurate, calculation. It’s similar to dividing two numbers and storing the result of the division. The calculation itself is not important; the fact that the parity data can be used to rebuild either drive is very important.

Disk striping with parity requires at least three drives, but it is very common to use more than three. Unfortunately, the more drives you use, the higher the chance that one might fail. Note that disk striping with parity can only recover data if one drive fails. If two drives fail, you’re heading for the backup tapes!

Disk striping with parity combines the best of disk mirroring and plain disk striping. It protects data and is quite fast. The majority of network servers use a type of disk striping with parity.

RAID

The many different techniques of using multiple drives for data protection and increasing speeds were organized by a couple of sharp guys at Berkeley back in the eighties. This organization was presented under the name RAID—Random Array of Inexpensive Devices, or Random Array of Independent Devices. There are seven official levels of RAID, numbered 0 through 6.

RAID 0 is disk striping. RAID 1 is disk mirroring and disk duplexing. RAID 2 is disk striping with multiple parity drives; this is unused, so you can safely ignore it. RAID 3 and 4 are both disk striping with parity. The differences between these two are trivial.

RAID 5 is disk striping with parity, but parity information is placed on all drives. This is the fastest way to provide data redundancy, and the most common RAID implementation.
RAID 6 is basically RAID 5 with the added capability of asynchronous and cached data transmission; you might think of it as a Super RAID 5.

No network tech worth her salt says things like, “We’re implementing disk striping with parity.” Use the RAID level. Say instead, “We’re implementing RAID 5.” It’s more accurate and very impressive to the folks in the Accounting department!

Drive Technologies

Talking about RAID levels is like singing about football. You may sound good, but that doesn’t mean you know what you are talking about! Remember that RAID levels are a general framework; they describe methods to provide data redundancy and enhance the speed of data throughput to and from groups of hard drives. The RAID levels you use do not indicate how you implement these methods.

There are literally thousands of different methods for actually setting up RAID. The method you use depends largely on the desired level of RAID, the operating system you’re dealing with, and the thickness of your wallet. Before I delve into these solutions, let’s do a quick run-through of hard drive technologies to make a few terms more clear.

EIDE

If you peek into most desktop PCs, you’ll find hard drives based on the ultra-popular EIDE, or Enhanced Intelligent Device Electronics, standard. EIDE drives are always internal; modern PCs are designed to use up to four EIDE drives. EIDE drives can be identified by their unique 40-pin ribbon cable connection.

Fig. 09-04. EIDE connections

The price, performance, and ease of installation explain the tremendous popularity of EIDE drives. IDE, the predecessor to EIDE, was exclusively a hard drive technology. EIDE can accept any type of storage device, including CD-ROMs, tape backups, and removable drives. Even with the ability to handle diverse devices, the PC cannot handle more than the maximum of four EIDE devices.

SCSI

SCSI, which stands for Small Computer System Interface, accomplishes much the same goal as EIDE: making hard drives and other devices available to the PC. SCSI, however, is not a hard drive technology. Instead, think of SCSI as a mini network that connects many different types of devices. Any type of storage device that you can imagine comes in a SCSI version. SCSI manifests itself in PCs via a card called a host adapter. The host adapter then connects to SCSI devices in a daisy chain. An installed set of SCSI devices is called a SCSI chain.

Fig. 09-05. SCSI chain

Each SCSI device on the SCSI chain must have a unique SCSI ID. SCSI devices are numbered 0 through 7, with 7 usually reserved for the host adapter itself. More advanced versions of SCSI can support up to 16 devices including the host adapter.

SCSI devices can be internal or external. Better host adapters come with an internal and an external connector, enabling both types of devices to exist on the same SCSI chain. Figure 9-6 shows a SCSI chain with both internal and external devices. Note that each device gets a unique SCSI ID.

Fig. 09-06. Typical SCSI chain with internal and external devices

SCSI Connections

Fortunately, the Network+ exam isn’t interested in your ability to configure SCSI. The exam does demand that you know the many connections unique to SCSI devices. No other one class of device has as many connections as SCSI. SCSI has been in existence for a long time and has gone through four distinct standard upgrades, and many variations within each standard over the years.

The fact that SCSI devices can be both internal and external gives SCSI drives an advantage over EIDE. There are two types of internal SCSI connections, both of which are inserted into a ribbon cable, just like EIDE: the 50-pin narrow connection and the 68-pin wide SCSI. Figure 9-7 shows a typical 50-pin narrow connection with a ribbon cable attached.

Fig. 09-07. 50-pin narrow SCSI connection

The oldest external SCSI connection is a 50-pin Centronics. Although it’s dated, a large number of SCSI devices still use this connector. It looks like a slightly longer version of the printer Centronics connector.

Fig. 09-08. 50-pin SCSI Centronics connection

Many host adapters use a female DB-25 connector. DB-25s have been on Apple computers for many years, but they are fairly new to PCs. This SCSI connector is identical to a PC parallel port. If you plug your printer into the SCSI port, or a SCSI device into the printer, it definitely will not work—and in some cases it may damage devices!

Fig. 09-09. Parallel and SCSI connections: both DB-25

Most modern SCSI devices now come with the special, SCSI-only, High Density DB connectors. High Density DB connectors look like regular DBs at first, but have much thinner and more densely packed pins. High Density DB connectors come in 50-pin and 68-pin versions. The 50-pin version is much more common.

Fig. 09-10. High Density DB-50

They Both Work!

Both EIDE and SCSI drives work beautifully for RAID implementations. People who are new to RAID often assume that RAID requires some special, expensive stack of SCSI drives. Such is not the case.

You certainly can spend the money on fancy RAID boxes, but you do not have to go that route. You can easily implement RAID using nothing but cheap EIDE drives and cheap, sometimes free, software. Further, RAID can use combinations of EIDE and SCSI—although trying to keep track of combinations of SCSI and EIDE drives is not recommended!

Most people prefer SCSI drives for RAID. SCSI drives tend to be faster than EIDE drives, and you can put more drives into a system: 7-15, rather than the limit of four for EIDE. The only drawback with SCSI is cost; the hard drives are more expensive and you often must purchase a host adapter as well. When speed outweighs cost as a factor—and it usually does for servers—SCSI implementations win. Finally, if you need serious speed and extra bells and whistles, you can install any number of expensive stack-of-SCSI-drives solutions.

RAID Implementations

All RAID implementations break down into either hardware or software methods. Software is often used when price takes priority over performance. Hardware is used when you need speed along with data redundancy.

The most famous software implementation of RAID is the built-in RAID software that comes with Windows NT/Windows 2000. The NT Disk Administrator can configure drives for RAID 0, 1, or 5, and it works with both EIDE and SCSI.

Fig. 09-11. NT Disk Administrator

NT is not the only software RAID game in town. There are a number of third-party software programs available that can be used with other operating systems. There are even third party software RAID solutions for NT that add a number of extra features above what Disk Administrator provides.

Most techs and administrators prefer hardware RAID. Software RAID works for very small RAID solutions, but tends to run quite slowly and usually requires shutting down the PC to reconfigure and replace drives. When you really need to keep going, when you need RAID that doesn’t even let the users know there was ever a problem, hardware RAID is the only answer. Since most organizations fit into this latter category, most RAID in the real world is hardware-based.

There are a large number of hardware RAID solutions, and almost all these solutions rely on SCSI. Assuming that you have the right type of host adapter, you can pull out a bad SCSI drive—and in some cases, a bad EIDE drive—and replace it with another one without even rebooting the server. This hot-swapping process is very common in hardware RAID.

Fig. 09-12. Hot-swapping a drive

RAID provides data redundancy. To implement RAID, you have to first decide the level of RAID you want to use and whether you want to go the hardware or software route. For the exam, make sure you can quote the different levels of RAID, and know your SCSI connections!

Tape Backup

Various RAID solutions provide data redundancy to a certain degree, but to secure your server data fully, nothing beats a tape backup. If the RAID solution works properly, that tape backup can happily collect dust on an off-site shelf somewhere. In the event of a catastrophic hardware crash, however—such as when two drives in your disk stripe with parity simultaneously go to that hardware heaven in the sky—only that tape can save the day.

Magnetic tape is the oldest of all methods for storing data with computers. Who hasn’t seen an episode of an old TV program showing old reel-to-reel tapes spinning merrily in the background? The reel-to-reels are gone, replaced by hard drives, and tapes are now relegated to the world of backup. Nothing can beat magnetic tape’s ability to store phenomenal amounts of data cheaply and safely.

Every properly designed network uses tape backups, so every network tech must learn to use them. The type of tape backup you implement will vary from network to network, as do the methods for backing up data. This section covers the types of tape backup; refer to the Lesson 12 for the actual backup methods.

There are a dizzying number of tape backup options, each with different advantages and disadvantages. They basically break down into three major groups: QIC, DAT, and DLT. All of the groups similarly use cartridge tapes—square tapes like fat audio cassettes—but the physical cartridge size, capacity, recording method, tape length, and speed vary enormously.

QIC

QIC, or Quarter Inch Tape, is an old standard that’s rarely used in any but the smallest of networks. QIC was one of the first standards used for PC backups, but it’s gone through many evolutions in an attempt to stay up with the demand for increased capacities over the years.

The earliest versions of QIC could store about 40 megabytes—which is fine for the days when tiny hard drives were the rule, but unacceptable today. There have been a number of increases in QIC capacities, as high as 2 gigabytes, but QIC has fallen out of favor as a desired tape standard.

Imation Corporation created an improved QIC format called Travan that is quite popular, again on smaller networks, with capacities up to 8 gigabytes. Under the Travan banner, QIC lives on as a tape backup option. Older QIC/Travan drives used a floppy connection, but EIDE or SCSI connections are more common today.

DAT

DAT, or Digital Audio Tape, was the first tape system to use a totally digital recording method. DAT was originally designed to record digital audio and video, but has easily moved into the tape backup world. DAT tapes have much higher storage capacities than QIC/Travan tapes—up to 24 gigabytes—and are popular for medium-sized networks. DAT drives use a SCSI connection.

DLT

DLT, or Digital Linear Tapes, are quickly becoming the tape backup standard of choice. DLT is a relatively new standard that has massive data capacity—up to 70 gigabytes. DLT is very fast, incredibly reliable, and quite expensive compared to earlier technologies. When the data is critical, however, the price of the tape backup is considered insignificant. DLT drives use a SCSI connection.

Data Redundancy Is the Key

Data redundancy provides networks with one of the most important things they need: security. Improper preparation for the day a server hard drive dies can lead to many quickly prepared résumés as you find yourself suddenly out of work! When the data is important enough—and when isn’t it?—providing data redundancy via RAID solutions is a must.

Chapter 3

Speed

A system providing a resource to a network has a tough job. It needs to be able to handle thousands, millions, even billions of transactions over the network to provide that shared resource to other systems. All of this work can bring a standard desktop PC to its knees.

If you’ve ever taken a regular desktop PC and shared a folder or a printer, and then watched the PC act as though it just shifted into first gear, you can attest to the fact that sharing resources is a drain on a PC. Systems that share resources, and especially dedicated servers, require more powerful, faster hardware to be able to respond to the needs of the network.

There are a number of methods for making a serving system faster. Making a good server isn’t just a matter of buying faster or multiple CPUs. You can’t just dump in tons of the fastest RAM. Fast CPUs and RAM are very important, but there are two other critical areas that tend to be ignored: a good server needs fast drives and fast NICs.

Fast NICs

The first place to look for making a server faster is the NIC. Placing the same NIC in your server that you place in your workstations is like putting a garden hose on a fire truck: it just isn’t designed to handle the job. There are a number of methods for making the NIC better suited to the task. You can increase the megabits—that is, the data throughput; make the NIC smarter; and make the NIC do more than one thing at a time.

Increase the Megabits

Mixing 10-megabit and 100-megabit Ethernet on your network can optimize network performance. The trick is to have the server part of the network run at a faster speed than the rest of the network.

If you have a 10BaseT network, you can purchase a switch that has a couple of 100-megabit ports. Put a 100BaseT NIC in the server and connect it to one of the 100BaseT connectors on the switch. The server runs at 100 megabits while the workstations run at 10 megabits, as shown in Figure 9-13. This optimizes the server speed and, since the server does most of the work in the network, optimizes your network as well.

Fig. 09-13. Server at 100 megabits per second; workstations at 10 megabits per second

Smarter NICs

Most NICs still need the CPU to handle most of the network job, but several companies make powerful NICs with onboard processors that take most of the work away from the CPU. Every NIC manufacturer has a different method to provide this support and those methods are way outside the scope of this lesson. From a network person’s standpoint, just buy a special server NIC, plug it in, and enjoy the benefits of faster response times.

Full-duplex NICs

Most network technologies consist of send and receive wires, and most NICs can handle only sending or receiving at a given moment. Full-duplex NICs can both send and receive data at the same time, which practically doubles the speed of the network card. Make sure that your server NICs are full duplex, but be warned that you may need to upgrade the server’s hub to take advantage of full duplex!

Making the NIC better is one of the easiest upgrades to a server, as it usually means simply yanking out an inferior NIC and replacing it with something better. At worst, you may even replace a hub or switch. Make your NIC better and you’ll see the results.

Make the Drives Faster

The other big way to make a server faster is to make the process of getting the data to and from the shared drives faster. There are two main options here, the first of which is to get fast drives. Using run-of-the-mill EIDE drives in a busy serving system is not smart. Try using high performance SCSI drives on a fast controller; this should make a big difference. The second option is to use RAID 5; you probably need it anyway for data protection anyway, and you’ll also enjoy the speed.

It’s Not Just Hardware

The demands of networking require servers to have better hardware than your run-of-the-mill stand-alone PC. Improving CPUs, adding RAM, using powerful NICs, and running fast hard drives all combine to make your serving PC more powerful.

But hardware is not the only answer. Good maintenance, such as defragging and setting up good disk caches, also plays an important role. Many times slow resource access is due to poor network design and is not the fault of the serving system. Be careful about throwing hardware at the slow access issues; it can often be money not well spent!

Chapter 4

Reliability

The last network function, primarily for serving systems, is reliability: the shared resource must be there when the user needs it. You can achieve reliability by providing a secure environment for the server and by adding redundant hardware to compensate for failed components.

There is a nasty tendency to confuse reliability with data protection. Don’t confuse the two. All the pretty RAID systems aren’t going to do you any good if somebody steals the server. Tape backups are useless if the power supply dies. Clearly, other technologies are needed to keep the serving system reliable. There is no logical order to explaining these technologies and safeguards, so I’ll cover them in no particular order.

Good Power

All of the components in the PC run on DC current electrical power. Without clean, steady, DC power, the components stop working.

There are a number of steps that electrical power must take between the power company and those components. At any given moment, if one of those steps fails, the PC will no longer work. You can take several actions to safeguard your hardware to make sure this doesn’t happen, starting with the power company.

Electrical power in the US is a wonderful commodity. Electrical service is pretty reliable, and the electricity is generally of high quality. Most folks in the US can count on a good electrical service 98% of the time—but even so, that 2% can really get you!

Electrical power sometimes stops, resulting in power outages, and sometimes goes bad, resulting in electrical spikes and sags. Additionally, techs can screw up perfectly good electricity on their own by overloading circuits with too much equipment.

You can protect the servers from problems of power outages, electrical spikes, and overloaded circuits with several important technologies: dedicated circuits, surge suppressors, UPSs, and backup power.

Dedicated Circuits

A dedicated circuit is a circuit that runs from the breaker box to only certain outlets. In most homes and offices, a circuit might have many jobs. The circuit that runs your PC might also run the office water cooler and the big laser printer in the next room.

Using too many devices on one circuit causes the power to sag. A sag can cause your computer to do nothing, lock up, or spontaneously reboot. It all depends on how lucky you are at that moment!

Dedicated circuits keep this from happening. In most cases, dedicated circuits have outlets with bright orange faceplates to let you know that they are dedicated. This is meant to prevent some uninformed person from plugging a photocopier into the circuit.

Surge Suppressors

It almost sounds silly to talk about suppressors these days, doesn’t it? Does anyone really need to be convinced that all PCs, network and stand-alone, need surge suppressors? An electrical surge—a sudden increase in the voltage on a circuit—can and will destroy an unprotected computer. Every computer should plug into a surge suppressor!

UPS

A UPS, or Uninterruptible Power Supply, is standard equipment for servers. Many UPSs also provide protection from power spikes and sags. Most UPSs provide only a few minutes of power, enough to enable the server to shut down cleanly. All servers should have a UPS!

Backup Power

When you really want to be reliable, get a backup power supply. No, you don’t have to put a diesel generator in the basement! There are a number of small battery-based backup systems that will provide a few hours of protection. If you want something that can last for a few days, however, you will need a gasoline/diesel backup system.

Environment

Keep the server room locked at all times. Get a card lock or combination lock doorknob and make sure that only trusted people have access. Keep the humidity low, but not too low—around 40% is about right for most electronics. Keep the room a little on the cool side; right around 68 degrees Fahrenheit is just about perfect, although most PCs can handle up to 80-85 degrees before overheating becomes a problem. If in doubt, check with the system’s manufacturer for recommendations.

Redundant Components

Many components inside the system can be made redundant. It is very common to find servers with redundant power supplies where a power supply can be removed without even shutting down the PC. You can buy NICs that work together in the same PC, covering for one another if one dies; there are even NICs that can be replaced without rebooting the PC! Placing hard drives on separate controllers—like the drive duplexing discussed earlier in this lesson—provides excellent redundancy.

Finally, there are methods for making the entire server redundant. There are a number of methods where two or more servers can be mirrored, providing the ultimate in reliability, assuming the cost is bearable!

NAS

A special note needs to be made about Network Attached Storage, or NAS. A NAS device is simply a server that has been dedicated to file sharing. It is not used for e-mail, authentication, file management, or any other of the roles of standard servers. NAS devices communicate using mainly NFS for UNIX environments or Common Internet File System or CIFS for Microsoft Windows environments; they can also use FTP, HTTP, and other networking protocols.

The reason that NAS devices are of note here, is that they are frequently used to handle such tasks as web cache and proxy, firewall, audio-video streaming, tape backup, and data storage with file serving. They can be set up as, essentially, PnP appliances dedicated to file sharing with their own redundancy components. By freeing your other servers up to take care of e-mail, authentication, and so on, you can go a long way toward increasing your network’s efficiency.

How Much Reliability Do You Need?

Reliability is like any security system: expensive, boring, a pain to administer; and you never have enough when you need it. Measure the cost of being down versus. the cost of reliability to make your decision. You might be surprised to find that it’s a lot cheaper to be safe than sorry.

Chapter 5

Conclusion

There is no such thing as The Perfect, Complete Network PC. Certainly every network PC needs to connect to the network; but data protection, speed, and reliability are functions that vary tremendously depending on network size, types of data and applications, the existing network cabling system, demands of growth, and of course, your pocketbook.

The Network+ exam does not assume that you can build the perfect network PC, but it does expect you to have a feel for the options you have. That way, when it comes time to build or buy that system, you can act as an advocate for your network, to ensure that you get as close to perfect as possible.

Supplementary Material

Webopaedia Definition of RAID http://www.webopedia.com/TERM/R/RAID.html

Here’s a definition of RAID, with related links, from the PC Webopaedia site.

How SCSI Works http://www.howstuffworks.com/scsi.htm

This definition, from the How Stuff Works site, explains the ins and outs of SCSI, including the numerous variations of SCSI standards over the years.

Copyright 2003 by ed2go.com. All rights reserved.
No reproduction or redistribution without written permission.