|







|
|
|
Here
you will find some basics, it is by no means complete. |
|
Hardware
101 - Quick Jump |
- Belarc
Advisor - Free Personal PC Audit (version 6.1f)
- Chipsets
101
- The
Intel 440BX AGPset
- Universal
Serial Bus (USB)
- Sockets
versus Slots
- The
World of DVD
- Firewire
IEEE-1394
- Bandwidth:
The Need For Speed
- The
Future of Network Bandwidth
- The
History and Future of Storage
- A
Introduction to AGP
- AMD
K6-2 and 3D Now!
- Moving
Beyond SDRAM
- Intel
440GX AGPset & 450NX PCIset
- Intel
Pentium II Xeon Processor
- Inside
Intel's Celeron(tm)/333A
- RAID
Explained
|
| Chipsets
101 |
| If
the processor is the brain of a system, then the chipset is the
heart of a system. In the past, third-party vendors (OPTi, VIA,
SiS, etc) created the chipset for Intel-based processors. Today,
Intel dominates the chipset market, while other vendors are
constantly playing catch up with Intel.
Click here for a complete
listing of each chipsets' features and specifications.
Notice that Intel chipsets are at the 100MHz bus, Pentium II
class, while other vendors are mostly doing the socket 7 class.
This is not by design; the slot 1 architecture is patented and
heavily protected by Intel.
|
| The
Intel 440BX AGPset |
| The
440BX AGPset is the third generation Pentium II class chipset from
Intel. (The first and second generation being the 440FX PCIset and
440LX AGPset, respectively.) Along with support for SDRAM (up to
1GB), AGP 2x at 133MHz, and UltraDMA/33, the 440BX is the first
chipset to support 100MHz memory bus speed. Pentium II CPUs from
350MHz to 400MHz (currently) will be able to run at 100MHz memory
bus speed. One practical advantage of the 100MHz memory bus is
that all processor speeds, starting with 350MHz, will be
increments of 50. CPU speeds such as 233, 266, 333, etc. will be
gone forever. Now we can have easy to remember numbers such as
350, 400, 450, etc.
The increase from 66MHz to 100MHz memory bus will definitely
help overall system performance as CPU speeds are increasing by
50MHz almost every four or five months now. There's a limitation
in CPU speed in which the memory bus becomes the bottleneck.
Running the memory bus at 100MHz requires a new type of SDRAM DIMM
called PC100; typically, PC100 SDRAMs are rated at 8ns or faster.
The BX also have a new feature called Open Page Architecture (OPA).
This allows the Pentium II processor to leave "tabs" on
virtual memory page addresses. So instead of wasting valuable CPU
cycles looking for a specific memory page, OPA will allow the CPU
to index the page immediately. Intel claims this will net a 3-5%
performance improvement for graphics intensive applications.
Initially, the BX will be paired with the Intel PIIX4-E I/O
controller. (The PIIX4-E controller governs the system memory and
I/O such as the two USB ports and the UltraDMA/33 IDE ports.)
Later on, this will be changed to the PIIX6 controller, which will
add support for Firewire or IEEE 1394 and four USB ports.
Click here
for the block diagram of the 440BX AGPset.
|
| Universal
Serial Bus (USB) |
| Universal
Serial Bus (USB) is an initiative originally proposed by Intel,
Compaq, and Microsoft back in 1995. The primary purpose of USB was
to make adding devices and peripherals to the PC easier. And to
centralized all the I/O connectors in the back of a typical PC.
Currently, there are connectors for keyboard, mouse, printer,
joystick, serial ports, etc. Why not just have one universal
connector in which user can connect all their devices?
In addition to standardizing on one port, the USB design also
have the following advantages:
Speed - The maximum bandwidth of USB is 12Mbps, which is
faster than most ethernet cards. Compare this to the maximum speed
of a typical serial port, which is 115Kbps.
- Hot Swap - USB is mainly designed for external
devices. Wouldn't it be nice to add and sremove devices to
your system without having to power it off and on? USB allows
user to add a device while the system is turned on; USB will
then automatically enumerate the new device, load the driver,
and allow the user to use it.
- Flexibility - With traditional serial ports, users
were limited to the type of device they can add. With USB,
there's almost no limitation. A USB device may be almost
anything: mouse, joystick, modem, printer, scanner, etc. Also,
a user can add up to 127 USB devices per system. Of course,
practical limitations may reduce this number, but it is still
better than having the standard "two serial ports per
system."
So you're thinking, "Hey, this is a great idea! But if
USB was introduced in 1995, where are all those wonderful USB
devices today?" USB suffers from the classic
chicken-and-egg problem. Peripheral makers were waiting for
system manufacturers to integrate the USB port into their
system, and PC makers were waiting for peripheral designers to
create USB devices first. And to compound the problem,
Microsoft didn't release USB drivers for Windows 95 until late
1997. If you've purchased a new computer system in late 1997
or now, chances are you'll have two USB ports with the USB
drivers loaded.
USB devices should take off once Windows 98 hits the
market. For more information about USB, see the following
links:
|
| Sockets
versus Slots |
| Pentium
class processors (including AMD K5/K6 and Cyrix M1/M2) are paired
with socket 7 motherboard. Likewise, Intel's IA-32
family of CPU is paired with motherboard with slot 1. The slot
1 design was conceived by Intel to make changing/upgrading the CPU
easily. Instead of inserting the processor into a socket, the user
simply add the CPU as he or she would an add-on card.
The socket 7 that exists on Pentium class system boards today
is able to accept Intel Pentium MMX, AMD K5/K6, and Cyrix M1/M2
CPU. In the next few months, AMD introduce a new super socket 7
(or super 7) for their AMD K6-2 processor. Super 7 will allow the
system board to run at 100MHz memory bus speed. This is necessary
for AMD to keep up with Intel's faster IA-32
family.
As far as Intel is concerned, the socket 7 design is a
dead-end. They will be focusing on slot 1 and slot 2. Because the
slot 1 design is patented solely by Intel, the only company that's
able to make slot 1 CPU is Intel. AMD may attempt to physically
clone the slot 1 design with their next generation K7, but the
electrical characteristics will be patterned after Digital's Alpha
processor. For the high-end Xeon CPUs, Intel will introduce the
slot 2. Physically, the slot 2 will be the same as slot 1, but the
number of pins on the slot 2 will be greater; this is to
accomodate the complexity of the Xeon processor.
And as a transition to the Merced, Intel may introduce yet
another slot called slot M in 1999.
|
| The
Multimedia World of DVD |
| Current
CD-ROM discs are able to storage a maximum of 640MB. DVD, the
successor to CD-ROM, discs are able to store up to 4.7GB of
information! And that's just the beginning. Using dual layer
technology, the storage capacity will double to 9.4GB per disc. In
addition, information may be encoded on both side of the
DVD disc, giving us an incredible 18GB per DVD in the near future.
So what can we do with 4.7GB of information? We can put a
full-length movie (2+ hours) encoded in MPEG2 format with Dobly
Digital 5:1 (or AC-3) on a single 4.7GB DVD disc today with room
to spare! Because of the extra free space, most DVD titles comes
with different language tracks: English, Spanish, French, etc. A
little known fact about DVD is that it really doesn't stand for
anything, but people have associated DVD as Digital Video Disc.
MPEG2 is a processor intensive compression / decompression
scheme; click here to learn more
about MPEG2 decoding. The visual quality of a typical DVD movie is
much sharper and better than super VHS or even a laser disc.
Unfortunately, quality comes with a price: an Intel Pentium II/333
or faster processor is required in order to perform software MPEG2
decoding. Because of this, most PC DVD kits rely on a dedicated
MPEG2 decoder PCI card. Due to current technology, this is an
inexpensive solution since it frees up processor for other use.
In addition to the high quality visual aspect of DVD, it also
incorporates Dobly Digital 5:1 or AC-3. This is an audio encoding
scheme developed by Dobly Labs. Click here
for more details.
|
| Firewire
IEEE-1394 |
| Let's
take a closer look at a faster version of USB, dubbed Firewire
or IEEE-1394.
Firewire will have
all the benefits of USB, including more bandwidth. The maximum
bandwidth of USB is 12Mbps (1.5MB/sec); Firewire's
initial speed will be 200Mbps (25MB/sec), which will be increased
to 1Gbps in the future. But Firewire
will not replace USB; in fact, it is designed to be a supplement
to USB. Low speed devices (such as keyboard, mouse, joystick, etc)
will be connected to the USB port. While high speed,
bandwidth-demanding devices (such as digital camcorder, CD burner,
etc) will be attached to the Firewire
bus.
And like USB, Firewire
hasn't really taken off yet. Currently, the only mainstream
manufacturer of Firewire
host controller is Adaptec.
The primary function of Firewire
today is digital video editing. Sony, Panasonic, and other
camcorder makers already have Firewire-enabled
camcorders available today. In addition, Firewire
may eventually replace IDE as the primary interface for
consumer-level hard disk drives. The current IDE interface is
limited to four devices on two channels (primary and secondary),
with a maximum cable length of about 12 inches per channel. Firewire
is limited to 64 devices with a cable length of 13.5 feet
between each devices
|
| Bandwidth:
The Need For Speed |
| One
of the main factor that's preventing the Internet from become an
all pervasive tool is bandwidth, or the lack of it.
If you're a casual user who uses the Internet for work and
play, then you'd wish that website would load just a bit faster.
And if you're a hardcore user who basically spends eight or more
hours per day on the net, then you can never have enough
bandwidth! New technologies on the horizon promise to alleviate
the bandwidth bottleneck for consumers. The bad news is, they are
not widely available yet. Here's a quick rundown on each of them
(starting with the slowest):
POTS Modems - POTS stands for plain old telephone
system; most households today probably have two or more phone
lines. Modem technology have increased dramatically over the past
few years. Consumers have witnessed the increase from 14.4K to
28.8K to 33.6K to today's standard of 56K. Today's copper phone
wires are limited to 53K due to an existing FCC regulation. This
is about as fast as you can push it with POTS modems. And even
with 56K technology, depending on where you are, your speed may
vary from 38K - 52K. Even with these limitations, a POTS modem is
still the most cost-effective solution for the majority.
- Multilink - So if you have two phone lines at home,
wouldn't it be great if you can somehow link those together
and double your bandwidth with POTS modem? You can with the
multilink feature that's available in Windows 98. Multilink
allows you to bond two (or more) modems to create a single
data channel. For example, if you have two 56K modems and your
connection speed for each is 44K, then your overall bandwidth
is 88K. A few things to keep in mind though. You'll need two
different phone lines, and your Internet Service Provider
(ISP) also have to support multilink. And most ISPs will
charge you double for using this feature. Also, keep in mind
that multilink will use the slowest speed of the modem.
It doesn't make sense to multilink two 14.4K modems (for a
total of 28.8K), when you can use a single 56K modem instead.
- ISDN - Integrated Service Digital Network never
really caught on for a few reasons. ISDN's bandwidth is
128Kbps using two 64K D channels. ISDN is still not widely
available; the setup cost and usage of ISDN is relatively high
compared to POTS modems.
- Satellite - Satellite's bandwidth is around 400Kbps.
This is more than enough for most users. But the setup fee of
a satellite connection is high, and this technology isn't
widely available yet. But if you need the speed and can afford
it, then it's a good alternative to POTS modems.
- ADSL - Asymmetrical Digital Subscribe Lines is one of
the technology to watch for in coming years. The downstream
bandwidth is 8Mbps. Both Intel and Microsoft have recently
jumped on the ADSL bandwagon. The monthly fee isn't very high,
but ADSL's availability is very limited right now.
- Cable Modems - The Holy Grail for all home Internet
users today. Cable modem's downstream is an incredible 30Mbps.
The monthly cost is around $40 for unlimited usage. (You must
have cable for TV in order to use cable modem; so the $40
charge is in addition to your monthly TV cable bill.) Consider
yourself lucky if your neighborhood offers cable modem today.
It is being deployed in some major metropolitan US cities.
To give you an idea how fast cable modem is, most
businesses are running with fractional T1 (from 256K to
1.5Mbps) or a full 1.5Mbps T1 line. 1.5Mbps is for the entire
office consisting of, depending on the size of the business,
perhaps 30-40 users. Larger corporations with a DS3 or T3
connection have a bandwidth of 45Mbps.
Of course, cable modem's 30Mbps bandwidth is only
achieveable if you're the only person using it. Most often
this is not the case, as the entire neighborhood shares the
30Mbps bandwidth. But even if it delivers only 1.5Mbps, it's
still more than enough for a single user.
|
| The
Future of Network Bandwidth |
| Many
businesses today are still using 10Mbps ethernet as their primary
network connection between desktop computers and hubs. But new
companies are starting out with 100Mbps fast ethernet connections
for every segment of their network: desktop, hub, backbone, etc.
With the explosive growth of the Internet, it seems that we can never
have enough bandwidth to satisfy our needs. Transfering multiple
video streams will even bring 100Mbps LAN to its knees.
As such, there are always newer and faster networking
technologies that promise to deliver more bandwidth. Here's a
quick summary of each technology:
| Type |
Speed |
Pros |
Cons |
Available? |
| Ethernet |
10Mbps
or 1.25MB/sec |
Inexpensive |
Showing
its age now |
Everywhere |
| Fast
Ethernet |
100Mbps
or 12.5MB/sec |
Fast |
Hubs
are still expensive |
Everywhere |
| Gigabit
Ethernet |
1,00Mbps
or 125MB/sec |
Extremely
Fast |
Still
an emerging technology; expensive |
Limited |
| Fast
Gigabit Ethernet |
10Gbps
or 1.25GB/sec |
Smoking! |
Being
developed |
No |
| Fiber
Optic or FDDI |
Up
to 100MB/sec |
Fast
and long distance |
Very
expensive |
Limited |
| ATM |
25Mbps
to 655Mbps |
Fast
and quality of service |
Expensive |
Limited |
ATM is the only technology that has quality of service (QOS).
This means if a video stream requires x amount of bandwidth, ATM
will guarantee x amount of bandwidth. Fast Gigabit Ethernet is
still in the development stage now, and probably won't be out
until late 1999. FDDI and ATM is still too expensive for every
single desktop systems. The fastest and cost-effective solution
today is Fast Ethernet. Fast Ethernet network adapters are
relatively inexpensive now; the biggest investment is still
probably the hub.
| The
History and Future of Storage |
| Since
the advent of computers, users have been searching for the
most efficient way to store their data.
As a sidenote, the high cost of storage in the past is partly
responsible for the famous (or infamous) Year 2000 problem
the computer industry is facing now. In the past, data
storage on magnetic media was extremely expensive. Instead
of storing all four digits to represent a year, people
choose to store only two digits, to save space and cost.
For example, 1998 would be stored as 98 only.
Here's a summary of the storage technologies used in
the past and what's coming up in the near future. This
article will focus primarily on magenetic and optic media:
Floppy Disk - Like the ISA slot, the floppy disk
drive is still present in most PC systems today. Even
though alternatives such as Iomega ZIP and LS-120 are
readily available now. The floppy drive is kept mainly for
legacy reasons today; some people still have important
data stored on their 3.5" 1.44MB disks. This drive
probably won't be phased out until the year 2000.
Hard Disk Drive (HDD) - The primarily storage
for most people today. The size of the average HDD have
increased dramatically over the past five years. And the
price of the average HDD have also dropped considerably.
This is mainly due to advancement in magnetic media
storage technology. Nothing beats HDD for storage: it's
the fastest storage technology available and it's
relatively cheap.
CD-ROM - The capacity of the CD-ROM (roughly
640MB) hasn't increased since its introduction. But the speed
of CD-ROM drives have increased geometrically. We have
seen drives that were rated for 1x, 2x, 4x, 8x, 16x, 24x,
and finally stopping at 32x. Although most CD-ROM drives
that claim 32x actually average about 12x to 20x. Only
data stored on the outer ring of the CD-ROM disc is read
by the drive at 32x. The inner rings cannot sustain 32x
throughput. 1x is approximately 150K/sec. So a 32x drive
is capable of reading data at 4.8MB/sec (150K/sec * 32 =
4,800K/sec). Today, CD-ROM is the most popular for
software publishing to distribute their application.
Removable HDD - HDDs are nice and fast, but you
can't move them from system to another easily. Removeable
HDDs solve this problem. The most popular media of this
type is the Iomega JAZ (1GB/2GB) drive. JAZ drives have
comparable performance to normal HDDs. The LS-120 and
Iomega ZIP drives also fall into this group, even though
they only store 120MB and 100MB per disk, respectively.
The best thing about the LS-120 is its ability to read
normal 3.5" 1.44MB floppies; when LS-120 drives come
down in price, it may completely replace the floppy drive.
Removable Optical Drive - This technology never
really caught on in the consumer market. Optical media
allows the user to write to a disc as many times as she
wants. Pinnacle
Mirco is the main player for optical drives. Its Apex
optical drive is able to store 4.6GB on both side of a
disc. But it was too expensive and too late. The JAZ drive
was faster and cheaper.
CD-R - These are CD-ROM drives with the ability
to record or "burn" a CD-ROM disc. A lesser
known name for CD-R is CD-WORM drives (write once, read
many). If you need to
store data that doesn't change, then CD-R is the way to
go.
CD-RW - This takes CD-R to the next logical
step: CD-RW (read/writeable) is able to write to a CD disc
more than once. CD-RW hasn't really caught on yet because
CD-RW discs are more expensive than blank CD-R discs. And
if a user wants to repeatedly write over the media,
there's always tape backup drives, which is slower but
much cheaper. In addition, a CD-RW disc may not be
readable to a regular CD-ROM drive.
DVD-ROM - The next generation of CD-ROM drives.
DVD-ROM discs can store up to approximately 4.6GB of
information. (See the article "The Multimedia
World of DVD" for more details.) Because of the
higher density of DVD-ROM discs, the speed rating of
DVD-ROM drives is different from CD-ROM drives. Most
DVD-ROM drives today are rated at 2x while reading DVD
discs and 20x while reading CD-ROM discs. 1x for DVD-ROM
drives is about 1.5MB/sec. DVD-ROM drives should replace
CD-ROM drives by the first half of 1999.
DVD-RAM - The future of optical storage is
DVD-RAM. This will allow users to write to DVD discs as
they would a normal HDD. Currently, there's no standard
for DVD-RAM. DVD-RAM created discs may not be readable on
all DVD-ROM drives. This technology probably won't take
off until second half of 1999.
With all these choices, it's easy for the average
consumer to get confused. My recommendation would be, get
a large IDE HDD (greater than 6.0GB in size) and a DVD-ROM
drive (which can read DVD-ROM, CD-ROM, and CD-R discs).
Add a CD-R drive if you're interested in creating your own
music and data CDs. Don't spend the money on a CD-RW or
DVD-RAM now. And finally, get a JAZ drive if you need to
move large (>100MB) files around from one system to
another daily. Otherwise, get a LS-120 drive, which can
read normal 3.5" 1.44MB floppy disc and 120MB floppy
disc.
|
|
| A
Introduction to AGP |
| With
today's demanding games and multimedia applications, even the
well-proven PCI bus cannot meet the video bandwidth requirements.
The 32 bit PCI slot running at 66MHz have a peak bandwidth of
133MB/sec. And in most cases, the 133MB/sec is only for burst
mode; the PCI slot cannot sustain 133MB/sec indefinitely.
Because of this limitation, Intel introduced the Accelerated
Graphics Port (AGP) architecture for video cards. It's important
to note that the AGP slot is for video cards only, unlike
the multi-purpose PCI slot. AGP is just beginning to take off this
year. The first Intel chipset to support AGP was their 440LX
AGPset, which is now being displaced by the newer 440BX
AGPset. The AGP slot is
able to run at 66MHz with the LX AGPset and a full 133MHz with BX
AGPset. Currently, AGP's full speed is 2x mode; Intel will
increase this to 4x in the very near future.
With the AGP slot in place, games and applications are able to
move and copy texture from system memory to video memory at a very
high speed, a feat that wasn't possible with the PCI architecture.
For now, 3D technology is known mainly in the gaming and graphics
designing world. But this may change when Microsoft release their
"Chrome" technology for web browser sometime in 1999.
"Chrome" will move 3D graphics and animation into your
browser.
For more technical information, go to the AGP
Forum website.
|
| AMD
K6-2 and 3D Now! |
| Intel
dominates the processor market with an 80% market share. By
leveraging their processor and chipset advantage, no company is
even close to challenging Intel. Both AMD and Cyrix have tried
unsuccessfully in the past.
But with the FTC lawsuit pending against Intel, and the growing
segment of sub $1,000 systems, AMD may be in a position to do some
real damage to Intel's market share. It's rare that Intel would
misjudge the market. But it did with the sub $1,000 market, and
now Intel is trying to regain grounds with its low-cost Celeron
CPU.
AMD recently launched their next generation AMD K6 processor
named K6-2. The MMX instruction set in the K6-2 has been optimized
so it performs about the same the Pentium II CPU. In addition, AMD
has added additional MMX instructions for floating point
operations; this MMX extension is named 3D Now! Current Intel MMX
instruction sets only works with integer. AMD have actually
managed to leapfrogged over Intel technologywise with the K6-2.
And K6-2 will level the playing field by bringing the 100MHz
memory bus to socket 7 motherboards.
But the K6-2 is still a socket 7 processor. (See the Sockets
versus Slots for a discussion of sockets and
slots.) Does the K6-2 have a chance in the market place when
everything seems to be transitioning to slot 1? Yes, it does
indeed have a fighting chance. Ironically, Intel has just recently
announced the socket 470 for their upcoming Celeron with L2 cache
processor! So the socket market is not as dead as Intel wants you
to believe.
Initial benchmarks have shown that the K6-2 is very close to an
equivalent Pentium II at the same clock speed. AMD won't be
stopping with the K6-2. They plan to introduce the K6-3 by the end
of this year. The K6-3 will have 256K of cache onboard.
|
| Moving
Beyond SDRAM |
| Moore's
Law states that processing power will double every 18 months. And
since the introduction of the Intel 8088 processor, this law has
never been broken. But increasing the performance of a processor
also requires increasing the bandwidth of the memory subsystem.
Here's a brief look at the past, present, and future memory
technologies.
Fast Page - The standard type of memory for 486, and
early Pentium 60/66/75/90 (still remember those?) systems. Back
then, fast page memory was sufficient to keep up with the
processor.
- EDO - Extended Data Output. EDO delivers about
10%-15% performance increase from fast page memory. The read
cycle from memory and CPU is shorten. EDO became the dominant
memory when the Pentium 100/133/150/166 (non-MMX) appeared on
the market.
- SDRAM - Synchronous DRAM. With the advent of the
Pentium MMX and Pentium II processors, memory technology took
another leap in terms of raw speed. SDRAM coordinates the
input and output of the memory and the processor. Makers of
SDRAM claim a performance improvement of 50% or more over EDO
memory. But in the real world, the performance is only about
20%. Currently, SDRAM is the dominant memory type, for both
the 66MHz and the new 100MHz (PC100) bus.
- SDRAM II (DDR) - Double-data rate SDRAM aims to
double the performance of current SDRAM. DDR allows the
processor to read from memory on both the rising and falling
edge of the clock. Currently, no Intel or third party chipset
fully supports DDR.
- RDRAM - Rambus DRAM. This is based on the technology
developed by Rambus, Inc.
Intel have selected RDRAM for their future processors, IA-32 (Katmai)
and beyond. RDRAM uses a narrow data channel clocked at
extremely high speed to achieve high performance. The Nintendo
64 game console uses Rambus as its primarily memory interface.
|
| Intel
440GX AGPset & 450NX PCIset |
| Intel
will be officially launching its next generation Pentium II
processor this month. The Xeon
processor promise higher performance while maintaining Pentium II
compatibility. The Xeon CPU will require a new slot 2
architecture. This means Xeon system board will using new
chipsets.
440GX AGPset - The only difference between the 440GX and
the current 440BX is the 440GX will be able to address up to 2GB
of memory (440BX can only do 1GB). The 440GX will be able to
support both the slot 1 and slot 2 architecture. Since the 440GX
will only support up to two Xeon processor, this AGPset is
intended for highend workstations and mid-range servers. All the
features of the 440BX (100MHz FSB support, USB, UltraDMA/33, AGP
2x, etc) will be present on the 440GX.
- 440NX PCIset - Notice the 450NX is classified as a
PCIset instead of an AGPset. This is because the 450NX does
not support the AGP slot. Intel intentionally left out AGP
support because the 450NX is meant for an enterprise server
system board. Another interesting aspect of the 450NX is that
it only supports EDO memory. The 450NX can address up to 8GB
of main memory. Intel recently annouced an errata
for the 450NX and their Xeon processor. Due to this errata,
four way Xeon CPU systems will be delayed a couple of weeks.
Intel will be releasing a software fix for this problem.
Both the 440GX and 450NX represent Intel's first generation
Xeon/slot 2 chipset. As expected, there's not too many new
features in these two chipsets. But the second and third
generation slot 2 chipsets will probably have AGP Pro (AGP
4x), Firewire, and RDRAM (RAMBUS)
support. Lastly, the front side bus (FSB) between the
processor and memory may be increased to 133MHz.
|
| Intel
Pentium II Xeon Processor |
| Intel
have recently launched their next generation server-class, highend
workstation processor named Xeon (tm). In keeping with the Pentium
II family of names, this processor's full name is Pentium II Xeon.
With the IA-64 Merced processor pushed back to late 1999 or
even early 2000, the Xeon is more than a stepping stone from IA-32
to IA-64.
The first thing you'll notice about the Xeon is its sheer size.
It's about twice the size of the Pentium II processor. In
addition, the Xeon is mated for a slot 2 motherboard, instead of
the typical Pentium II slot 1 board.
The most important change in the Xeon is the speed and size of
its L2 cache. Pentium II's L2 cache runs at half the CPU clock
speed and the size is fixed at 512K. The Xeon uses a special type
of cache memory called CSRAM (customized SRAM) which allows it to
run at full CPU clock speed. The Xeon is available with
512K, 1MB, or 2MB (450MHz or faster) of L2 cache. The Xeon is
still based on the P6 architecture and the MMX instructions are
still the first generation, not the new KNI (Katmai New
Instructions).
Current Pentium II processors and chipsets are limited to a
dual processor configuration. The Xeon is able to scale up to quad
and even eight processor later. With a quad configuration, 2MB of
full speed cache, and the 100MHz FSB (front size bus), the Xeon is
well suited for a server or highend workstation environment.
This level of performance comes with a cost. In order Intel to
boost their profit margin, they need to market and sell the Xeon
as a highend processor.
Related Links:
|
| Inside
Intel's Celeron(tm)/333A |
| Intel released a new addition to their Pentium II family.
This low end processor is called Celeron(tm). Essentially, the
Celeron is a Pentium II CPU without any L2 cache onboard. It comes
with support for MMX and fits into any slot 1 system board.
But because it did not have any L2 cache, the Celeron lags in
performance when compared to the AMD K6 or K6-2 or Cyrix M2. Intel
launched the Celeron at 266MHz, which they quickly moved to
300Mhz. Even at 300MHz, the Celeron was no match for the lower
cost AMD K6-2. Perhaps Intel released the Celeron prematurely, but
they had to release a CPU that was price competitive with the
AMD/Cyrix processor.
Even before Intel was putting the finishing touches on the
Celeron/300, the next generation Celeron with 128K L2 cache was in
the works. Intel's initial plan was to release this new Celeron in
Q1/1999. But most likely due to intense competition from AMD and
Cyrix, Intel will be release their Celeron/333A by the end of Q3
of this year or beginning of Q4. The "A" at the end of
the clock speed will allow consumers to differentiate between the
cache and non-cache version.
The important thing to note about the 128K L2 cache is it is
running at full CPU speed. The L2 cache on Pentium II
processors, not including the Xeon, only runs at half the CPU
speed. This mean that the performance of a Celeron/333A with
128K L2 cache is almost equivalent to a Pentium II/333 with 512K
L2 cache! The speed of the Celeron L2 cache makes up for its lack
of size. This will allow Intel to compete at a price point similar
to the AMD K6-2 and Cyrix M2.
|
| RAID
Explained |
| Unlike
desktop systems which use IDE disk drives as their primary storage
subsystem, servers usually utilize a SCSI storage subsystem
coupled with an high-end RAID controller.
A dedicated SCSI controller can lower processor utilization,
which is critical in a server environment. And most servers
require some sort of fault tolerance against hard drive crashes or
failures.
RAID stands for Redundant Array of Inexpensive
(or Independent) Disks. The concept of RAID is
simple: instead of having a single large capacity hard disk (which
is expensive and prone to failure), gather a collection of smaller
and inexpensive disks and call it a RAID array. For fault
tolerance, split the data between the drives, instead of storing
all the data on one physical drive. In addition, by splitting the
data across different physical drives, I/O performance may be
increased.
There are different levels of RAID technology; the three most
commonly used are:
RAID level 0 (striping)
- Level 0 offers maximum disk I/O performance at the expense of
fault tolerance. Data is stored (or striped) across the array of
drives. This can greatly enhance I/O performance by reducing the
latency time present in a single drive. The downside to level 0
is, if one drive fails, then all your data would be lost. Level 0
is ideal for a workstation system, which requires optimal I/O, but
it's definitely not suitable in a server.
- RAID level 1 (mirroring)
- Offers excellent fault tolerance at the expense of cost.
Level 1 requires two drives: a primary and the backup drive.
The size of the backup drive must be at least the same size or
greater than the primary drive. Any data that is written to
the primary drive is also written to the backup drive. In
effect, the backup drive merely sits there waiting for a
potential crash on the primary drive.
- RAID level 5 (striping with parity)
- Most servers will use this RAID level. Level 5 is simliar to
level 0, except, in addition to data, parity information
is also striped across the RAID array drives. Parity
information allows the other drives in the array to rebuild
the information on the crashed drive. Level 5 offers the best
of both world: it has excellent fault tolerance and
performance.
In addition to the various levels of RAID, there's two
concepts in a RAID environment that can further simplify drive
management. The first is known as drive hotswap. This allows
the user to replace a crashed drive in a RAID array, without
having to power the system down. The user takes the bad drive
out of the drive bay, replace it with a new drive, and insert
the drive bay back into the system. All the time while the
server is up and running.
The other concept is spare pooling for a RAID level 5
array. This allows the user to set up a number of spare
drive(s). When a drive in the array crashes, the RAID
controller will automatically take a drive from the spare pool
and add it to the existing array.
Related links:
|
|
|