Why's the network running so slow? Where's the bottleneck? Is it hardware, software or a user? You can find answers quickly if you know your network's infrastructure.
- By Bill Heldman
As a server administrator you probably spend your day taking care of
and worrying about the server farm. There are logs to read, applications
to attend to, MP3s to delete off of users' home drives and so forth
a busy person.
So what happens when you come in one sunny Monday morning and you're
practically attacked by users who want to know why the network is so slow?
You've got servers to attend to and how should you know what's going on
with the network? But you can't tell your users that, especially if you're
the lone administrator in your company. You have to dig in and figure
out what the problem is.
You might at first suspect that somebody's trying to download a huge
file from the Internet or perhaps running a big report from an application
server. Upon checking, you find that there aren't a lot of users attached
to the networkthere can't be because the
network is so slow!
This is where knowledge of the infrastructure comes in handy. Your problem
may not lie at all with the servers or users; it may be in the guts of
the network: the cabling (and associated wiring closets), protocols, switches
and hubs. It's key that administrators understand, almost intimately so,
the nature of the infrastructure because so many problems occur there.
If you're not familiar with the infrastructure's makeup, you may overlook
a very important place where you can find clues in your troubleshooting
Candidates, Listen Up
Because of the importance of infrastructure the CompTia
A+ and Network+ tests are good to have under your belt.
Both of these tests will help you understand how a PC
with a network interface card (NIC) inside it connects
to the network world. The CompTIA tests are designed
to be agnostic toward any one personal or server operating
system (OS) or hardware component, so they're a great
way of getting an overall flavor for what PCs, networks
and servers are all about without having to delve into
any one company's offerings. See www.comptia.com
for more details regarding their tests.
Microsoft indirectly sanctions these exams, allowing
MCSA candidates who have passed a combination of CompTIA's
A+/Network+ exam to waive the MCSA track's elective
requirement. See www.microsoft.com/trainingcert/
Microsoft used to offer a test called Networking Essentials but has discontinued
that requirement when it revamped the MCSE program. In its place you might
pursue the CompTIA offerings. While you can learn some things from Exam
70-221, Designing a Microsoft Windows 2000 Network Infrastructure, that
exam assumes that you already know something about network infrastructures
to start with (and it's brutal). Windows 2000 has heavily raised the bar,
in terms of the underlying infrastructurebandwidth
requirements, Windows 2000 service offerings, and so forth.
So, let's take a look at the things involved in your network's infrastructure
in the next few months, understanding that you'll have to take a closer
look at each in order to really feel comfortable with the network's overall
operation. I'll also give you some pointers on what tests to study for
in order to augment your overall administrative experience (and make yourself
even more employable than you already are).
First off, let's look at bandwidth issues and how the cables your network
uses can affect network performance. In the coming months, we'll discuss
the differences between hubs, switches and routers, and the types of internetworking
protocols that you're likely to encounter.
Bandwidth of The Network
Most electronics engineers disdain the word "speed" when used to describe
the bandwidth you have on your network, but that's what it really comes
down tothe amount of bits per second that
can be transmitted through the cabling, which is really the network's
data transfer rate. (Never, ever refer to the network's data transfer
rate as speed when you're talking to people who work with infrastructures
all the time. They'll lecture you for an hour on how it's not speed, it's
the transfer rate. As for me, I'm perfectly fine with the use of the word
speed as I think it adequately describes what's really going on.) Older
networks function at 10 million bits per second (Mbps10Base-T),
newer networks run at 100 Mbps (100 Base-T) and the newest networks operate
at 1000 Mbps (gigabit Ethernet"Gig-E"
or 1000Base-T). Note that all of the data transfer rates given above are
based upon Ethernettoday's king of network
architecture. There are other data transfer rates that are used when you
consider token ring, an older and at one time very popular architecture.
The CompTIA Network+ test will delve into all of the various ways that
a network can be wired (its topology) as well as the assortment of data
transfer rates that can be derived from today's network architectures.
Ethernet is a network architecture, and you can wire it in a bus or star
topology (in a line, one PC after another, or all PC cables directed to
a single hub or switch, respectively; see Table 1). Token ring is also
a network architecture and is always wired in a ring. The Institute of
Electrical and Electronics Engineers (IEEE) alluded to in Table 1 is an
international body that regulates standards regarding network architectures
and data transfer rates. See http://computer.org
for more info.
|Table 1. A comparison of architectures,
topologies and data transfer rates.
Network bandwidth is predicated largely on the cabling you have installed.
Category 4 (Cat 4) wiring cannot play in the 100 Mb/s sandboxyou
have to have Category 5 (Cat 5) wiring to make this happen. While some
well-installed Cat 5 wiring installations might work in the gigabit environment,
Category 6 cabling is generally recommended. So the bandwidth you desire
for your network depends on the health of your building's cabling. Re-cabling
a building can be an incredibly expensive proposition and one that you
should never undertake yourself unless you've been trained in it. Leave
the re-cabling job to professional cabling companies.
There are three kinds of cabling you have to consider:
- Cable that runs from your data center to the wall jack in a user's
- Cable that runs between wiring closets (we'll discuss wiring closets
in more detail later on in this article)
- Cable that connects PCs or servers to the wiring infrastructure
Generally, the cable that runs through your ceilings or floors to user's
offices and is then attached to their wall jacks is copper, has a plenum
to keep it from giving off harmful fumes in the event of a fire, and is
stranded (wound). Cables that run from PCs or servers and somehow connect
to the wiring infrastructure are typically solid wire (or should be if
they're not) instead of stranded and don't necessarily need to have a
plenum. (Stranded cable has a tendency to weaken at points where it is
excessively bent so solid cable makes more sense in cables that run from
the user's NIC to the wall jack. Stranded cable is less expensive than
solid, so it makes sense to use it when running it through a ceiling breezeway.)
Cables that run from one wiring closet to another can be fiber-optic or
copper (and typically would be stranded if copper).
Take a look at a cable next time you're in a wiring closet. It will be
clearly marked as to the category it is rated for. Fiber-optic cable (which
is not marked with a category) is always orange, flat and easy to recognize.
Ethernet cable comes in many different colors, is thinner and more rounded
than Ethernet cable and must be labeled as to the category it supports.
When studying for a network exam, you'll likely be tested on the different
kinds of topologies and cables that you might run into. While it's good
to study these kinds of cabling environments for the test, it will be
an extremely rare case in which you run into anything other than CAT5
or CAT6 in an Ethernet environment. So strong is the Ethernet standard
and so adopted is 100Base-T with CAT5 cabling, that you should consider
it a lucky break to work on old cable such as coaxial 10Base-2 or Token
Ring with its Multiple Access Units (MAUs) and other funky architectures.
(OK then, maybe not. Consider it lucky that you have Ethernet instead.
And content yourself to study other cabling for the network examsplanning
on never running into any other kind…seriously.)
Every once in a great while, you'll run into an instance where you're
having problems with a user trying to work on a 100Base-T network and
exhaust all troubleshooting avenues, only to find out that the user has
an old CAT4 cable. It's easy to tellthe
cable's well marked. Sub it out and the user is on his way.
Why are companies so enamored with Ethernet and CAT5 or CAT6 cable? Because
it's easy to understand, standardized, quick and painless. It works! Where
will you run into trouble? We've already alluded to the CAT4 cable trying
to work on a 100Base-T network issue. But you can also run into badly
crimped connectors, stranded cable that's come loose, cable, jacks or
receptacles that have failed, Ethernet cable length violations and cables
that are not properly seated in their receptacles. You can purchase cable
troubleshooting devices that will assist you in troubleshooting cabling
problems, but a watchful eye will be of great help for administrators
who don't have bucks to throw at expensive testing gear.
Cabling cannot run parallel to electrical wires because you'll get a
crosstalk problem. Cables cannot be longer than the Ethernet length rule
or you'll run into performance or dropped data issues. Cables shouldn't
really come out of a switch in the IDF or MDF and then run into a hub
in a user's office off of which the user has subsequently hung a bunch
of devices. That's a great way to generate performance issues. Better
to run multiple jacks to this user in such a circumstance. Poorly terminated
or mistreated (bent at a 90 degree angle to make way for a desk) cables
will give you fits.
| Well, it's a big topic. Suffice to say
for now that you'll have to set up at least one Wireless
Access Point (WAP) where the wireless devices can connect
to the network and that typical building stuff like concrete,
steel girders and other things will severely limit a device's
ability to get to its WAP, in spite of the distances the
vendor says the wireless device can go. Test, test, test!
One of my server administrators set up a WAP and then
took her Compaq iPAQ around the building to see how well
she could connect. In some cases she couldn't hit the
WAP at all; in others she got the full 100 yards the literature
said she could.
Wiring ClosetsMDF and IDF
We give an interesting name to the wiring closets in your building. The
main wiring closet, typically where the routers, telephone gear, and perhaps
even some of your servers are located, is called the Main Data Facility
(MDF). Most MDF have a wiring patch panel in themon
the one side are jacks, on the other are cables leading from the patch
panel out into the building to the various offices. There are Ethernet
and fiber patch panels (token-ring requires a little different hardware
configuration). The patch panels and associated wiring running to offices
are generally installed by cabling experts. I've run into network performance
issues that involved shoddy crimping of the connections between the patch
panel and the wiring running to office wall jacks.
You'll run an Ethernet patch cable from your servers, routers and other
gear to a hub or switch (I cover hubs and switches in more depth next
time) or directly to the patch panel itself. Each patch panel jack is
numbered, giving you the ability to trace where your wires are going.
If you have a feeder wire going to another wiring closet somewhere in
the building, we call that other closet an Intermediate Data Facility
(IDF). In many configurations, the wire connecting the two closets is
fiber-optic because it was at one time able to handle higher speeds than
conventional Ethernet cabling. However, today's Gig-E standards have allowed
Ethernet to play in the high-speed arena. Figure 1 shows an MDF connected
to an IDF with a fiber-optic cable.
|Figure 1. A typical MDF/IDF configuration. This
shows two wiring closets in different sections of your building. Server
A is connected by Ethernet cable to the backbone, Server B by fiber.
Note that the same switch can host a Cat 5 and a fiber-optic connection
as well as different data transfer rates.
Note that the patch panel cables can go to different parts of a buildingyou
might have some offices fed off of the patch panel in the IDF, while others
are fed off of the patch panel in the MDF. Fiber patch panels are easy
to differentiate from standard cable panels. There's an orange jumper
cable running from the fiber patch panel to a switch. It's possible to
equip servers with fiber-optic NICs and connect them directly to the fiber-optic
ports on a switch. The patch panels, switches and associated cabling between
the IDF and MDF comprise the network's backbone. Any time a server is
hooked directly to a switch that is high-speed and communicates directly
with the MDF and IDF, it is said to be hooked to the backbone. Servers
can hook to the backbone through conventional Ethernet or fiber-optic
In Figure 2 you can see that you have two wiring closets (I call them
closets and they might well be closet-sized, but they could also be full-sized
rooms) each of which has two patch panels, for fiber-optic (an eight-port
panel) and one for Ethernet (a 16-port panel) as well as a switch in each
closet. The MDF and IDF are connected by a fiber-optic cable running between
them. One server, Server A, is connected to the backbone by an Ethernet
cable, the other, Server B, is connected by a fiber-optic cable. Because
they're both connected to the backbone, you've reduced slightly your chances
for failure (because you don't have a wall jack introduced to the system)
and you can run the servers at a higher data transfer rate than the workstations.
|Figure 2. Server A is connected to an Ethernet
port on a switch in the MDF. Server B is connected to a fiber port
on the same switch. Server C is connected to a fiber port on a switch
in the IDF. The two switches in the IDF are connected to one another
by a jumper cable on the uplink ports. Incoming wiring from offices
passes into the patch panel. Patch cables then run from the patch
panel jacks to the switches. In this way users can communicate with
the servers. Switch ports can be mixed and set for various data transfer
rates. Servers are connected to the backbone.
I should make a few points before we continue this discussion. Fiber-optic
cable uses strands of glasstwo strands
making up a pair. Fiber is typically installed and you're charged by the
number of pairs that are put in. For redundancy's sake, you'll want a
second pair of fiber to run between your MDF and IDF. That way, in case
the first pair fails, you can very easily and simply snap in the replacement
pair, and your network would be running again. You should follow the same
redundancy procedures for Ethernet cabling running at Gig-E speeds. If
your backbone cable connecting the MDF and IDF goes out, then users cannot
communicate with one another or the servers. You always want to consider
redundant links between your MDF and IDF backbone.
You can have more than one IDF, so you can understand how the design
might get a little tricky. One company I worked for had 16 IDFs connected
to the data center MDF. Generally, the MDF is the place where the servers
live and all of the building's network wiring terminates. It's Grand Central
Station, if you will. You will typically find that your WAN connectionsT1
Frame Relay, etc.also terminate inside
this room. A WAN circuit (or any telephony circuit, for that matter) termination
point in a building is called the demarcation point and is referred to
by internetworkers (router and WAN circuit folks) as the "demarc" (and
which I've also referred to in some of my books as the "d-mark"). As a
general rule of thumb, your routers are also located inside the MDF.
Finally, when considering the MDF and IDF backbone, you'll want to take
into account the aggregate bandwidth of the backbone. What I mean by that
is if you have a 1000Base-T backbone (1000 Mbps) and you attach a lot
of servers at 1000Base-T and dozens or hundreds of users at 10Base-T or
100Base-T, it's possible that you could saturate the network with the
collective bandwidth being used by all the computers. You should note
that it's only remotely possible because it's very seldom that a user's
workstation will take up anywhere near the bandwidth the NIC is capable
of putting out, nor do servers typically operate near the upper end of
Gig-E bandwidth limitations.
However, in large client/server installations that generate a lot of
activity and reports, or where large files are regularly dragged across
the wire (as in the case of backups, for example), or some distance learning
is going on and users are downloading video from the Web, it's possible
for the backbone to get saturated and for the network to slow to a crawl.
(The three instances referred to in the previous sentence aren't the only
ones which might introduce excessive aggregate bandwidthjust
examples.) I've seen this happen time and time again. If you suspect an
aggregate bandwidth saturation issue have the network sniffed by a professional
during the times when you suspect that this is happening. It's the only
good way to figure out which stations are causing the problem and to whom
I hope I haven't confused you and instead have helped you with all of
the dialog regarding MDFs, IDFs, cabling and so forth. It is key that
you understand how your cabling, MDF and IDF hook together because a poor
design or shoddy workmanship can be very effective in slowing a network
down (and make problems very difficult to find). If you have a problem
and you've considered all things associated with the server software (including
TCP/IPa simple protocol suite but one
that can make so much trouble) and hardware, then your next stop is the
Above all, don't be afraid of the infrastructure. If you've ever set
up a tent, wired a basement, worked a maze in a book, done macramethen
you have the basic ideas needed to understand a building's cabling. It's
all about lines connecting to one thing, leading somewhere and then connecting
Next time we'll talk about the devices that allow users to connect to
the IDF and MDF and to communicate with one another and with serversswitches
and hubs. It's fun interesting stuff, so stay tuned.