Desktop Fragmentation
Increased Connectivity Opens Multiple Endpoint Possibilities
By Roger L. Kay
Time was when a client was a client was a client. We had desktops and notebooks, basically
stationary and mobile PC clients, and you either got one or the other. There were choices of
environment (Windows or Mac), choices of color (beige or black), and choices of vendor (Dell,
HP, Gateway, and so on), although this last was less of a choice than it looked because
essentially the same group of companies — Intel or AMD, ATI or nVidia, Seagate or Western
Digital —supplied the technology behind them all.
These clients (or endpoints) were sometimes called “fat” clients, but Microsoft has always
objected to this term, perhaps because it implies bloat, not exactly a brand-polishing concept.
“Rich” clients sounds better.
Originally, these rich clients were just complete computers, albeit microcomputers, that allowed
the user to do all computing task right there at the desk. All the rich resources required were
already in the box. But preceding the rich client was another paradigm, sometimes called
“terminal server,” which dated even farther back in the mists of time to the days of the mainframe
and minicomputer. Under this scheme, the terminal didn’t do much, just accepted keystrokes
from the user and passed them up by data communications to a big central computer, perhaps
shared by a number of users. The big computer, when the user’s slice of time came up, sent
back down the line the appropriate pixels to be painted on the screen. Most of the actual
computing was done in the big computer.
The terminals at the end of the line didn’t have much smarts and were thus sometimes called
“dumb terminals.” But a more elegant term evolved, particularly after the advent of client-server
computing: “thin clients.” So, now you had two parallel tracks, two computing schemas, one
involving rich clients and one involving thin clients — because after all, you can never be too rich
or too thin, as Wallis Simpson, the Duchess of Windsor, may once have said (whereas it’s easy
to be too fat and too dumb).
As the PC market developed, rich clients proliferated, outnumbering thin clients by more than
100 to 1. And there it might have rested, except for a few trends.
The most important was the evolution of bandwidth. As data communications became cheaper,
faster, more reliable, and widely available, various alternative schemas became more possible,
ones that would divide the computing task between the client at the edge and some computer or
computers at the center without as much of a performance penalty as in the past.
Also, security started to be more of an issue. Since the dawn of the 21st century, data owners
have become increasingly aware of the potential loss of data, loss of value, breach of trade
secrets, and theft of identity that can happen, either via malicious intent or by accident, with all
that data passing back and forth through the network and lying out at its edges in unprotected
hard drives.
Finally, there was the problem of complexity. Networks got built ad hoc. This and that were
added on as they became available or the need arose. Different architectures, generations of
technology, and applications were layered on like so much adobe, and things just got a lot
harder to manage.
So, with good connections, security problems, and too much complexity, folks began to take a
second look at computing models to see if there weren’t a better way. The need had arisen for
one or more schemes that could mitigate these issues, and various entrepreneurs and
established vendors rose to meet the challenges and take advantage of the opportunities.
Alternative Client Computing Paradigms
Over the past few years, a spectrum of alternative solutions has cropped up, such that it is no
longer appropriate to simply dismiss the category as "thin clients" and be done with it. The
flavors of thin, thin-ish, and semi-rich computing are many. The following paragraphs attempt to
sort them into major categories.
Terminal Server
The traditional thin client model, the oldest of the lot, involves the client doing very little other than
accept input from the user and display output from the server on the user’s local monitor.
Computing is done at the server end. Hardware at the client end is supplied by Wyse, NeoWare
(recently acquired by HP), and several others. Communications takes place via Remote
Desktop Protocol (RDP) to Microsoft’s Windows Terminal Services on the server. The good
news is all the valuable stuff is back at the server end. The bad news is performance isn't so
great. Terminal server is good for things like data entry. Terminal server can provide a good,
cost-effective answer for a situation, for example, in which a large number of task workers are co-
located in a service facility.
Diskless Computing
A semi-thin solution is typified by Ardence, which was recently acquired by Citrix, a supplier of
thin client software solutions. Diskless computing involves what amounts to a PC at the client
end, but the storage lies at the server end. The software redirects disk access requests to the
network transparently to the user. Data comes down the line as if were being summoned from
the disk. The bad news is that some degradation of performance occurs, but with gigabit
Ethernet, this shortcoming can be minimized. The good news is that, when the client is
powered down, all the data stays up at the server end, where it can be physically protected. A
typical customer for this type of solution would be a government lab, where data security is
paramount.
PC Blades
Next, we have PC blades, pioneered by ClearCube in 2001 and taken up in 2003 by HP as part
of its Consolidated Client Infrastructure (CCI) initiative. The two architectures differ somewhat,
but both have full PCs on blades in a rack in the datacenter and thin clients out in the field. The
two types of communications are brute force coax over short distances, which delivers decent
performance, and TCP/IP over Ethernet, which can handle long distances but gives poor
performance. CCI envisions a storage area network (SAN) behind the server pool, while
ClearCube virtualizes storage in the rack on the PC blades themselves. However, each side is
moving toward the other for competitive reasons. More recently, ClearCube, which is privately
held, has been having financial difficulties, which may cause it to abandon it solutions approach
for lack of economies of scale, and focus more directly on the software that manages the
blades. Avocent also plays in this space with a PCI card client insert that enables a standard
PC to function as a client-side device in a blade schema. Avocent has teamed with HP and Apro
to offer a full solution. The good news here is that the architecture is familiar to most IT
managers and has advantages in manageability and security. The bad news is that the cost of
the blade plus the thin “puck” at the client end is generally higher than a comparable PC and
doesn’t perform as well.
Hardware Compression
Most recently, a new wrinkle has been introduced: hardware graphics compression.
Essentially, working silicon now exists that takes output from a PC blade’s graphics accelerator
subsystem, compresses it, delivers it over standard TCP/IP links, and on the client side
recomposes it for display. This fix allows good performance over remote links for rich data,
giving the blade solutions a boost. More processing can be done at the server end and rich bits
can be sent to the client without sacrificing performance. This silicon is supplied by Teradici, a
Canadian company based in Vancouver, which has announced a number of partnership in the
past year, including tie-ups with ClearCube, IBM, Verari, and Leostream, and partnerships with
AMD, Ericom Software, nVidia, and Devon IT. Teradici calls its solution “PC over IP” or PCoIP.
Virtualization
All of this is getting muddied by virtualization, which has many benefits as well as some
associated Microsoft licensing issues. For example, virtualization allows blade PCs to be used
for more than one end user, improving the economics, but some of the flexibility implied by such
a scheme is subject to licensing restrictions Microsoft places on its customers. Virtualization
also has an impact on the storage pools that can either be aggregated on the blade rack and
allocated to users as needed or held in massive SANs behind the rack in the datacenter.
Winners and Losers
On the winner/loser scorecard, Wyse is pretty well positioned on the hardware end as the
largest independent supplier of thin clients. Ardence has done well for itself, including being
acquired by Citrix. Teradici has some key intellectual property. ClearCube has a good software
portfolio related to failover switching and remote client management as well as virtualization, but
is struggling with scale on the hardware end. HP seems to be treading water with CCI, although
it has announced a few wins from time to time. Dell recently entered the market with its Flexible
Computing Solutions offering, based on the Citrix-Ardence technology. No word on how that’s
going yet. Microsoft is moving cautiously over licensing in this environment; case in point, the
company changed its policies earlier this year, issuing a cryptic note that appears to relate to
permitting Ardence-type solutions. IBM is happy to promote thin computing because all it really
wants to do is sell servers. Lenovo has licensed ClearCube or is a reseller, but does not seem
to be running with the ball very hard. Avocent also seems to be dancing around the edge of this
space, but is weakened by its dependence on partners.
Where to From Here?
Thin computing has been forecast to grow for eons, but has consistently under-delivered. The
main reason is inertia. People like to work with what they are familiar with, and Microsoft and
Intel have them pretty well trained on PCs. However, the theory goes: the argument in favor of
thin and thin-ish computing gets better as bandwidth becomes more reliable, more available,
and faster. From a marketing perspective, thin clients work best in a self-contained installation
where the vendor "guarantees" that it will work. IBM has done fairly well with this services-
oriented approach. HP has tried it as well.
Given the dynamics, no reason exists for thin clients not to become increasingly popular. In the
extreme case, why not have all your computing done safely in the glass house if users are none
the wiser? The logical conclusion of this thinking leads to a cloud-computing world such as the
one envisioned by Google.
But mobile computing necessarily involves some intermittent communication, and people still
want the full experience even when they're not connected. Thus, rich clients are likely to
dominate for years to come, particularly as the tide turns in favor of mobility. But as even mobile
clients become connected most of the time through advanced, high-bandwidth wireless
connections like WiMAX, the pendulum may swing again in favor of thin computing. The growth
rate of alternative client shipments has exceed that of PCs for years. It's just that the base is so
small that alternative client shipments are still two orders of magnitude smaller than PC
shipments.
© 2007 Endpoint Technologies Associates, Inc. All rights reserved.
stationary and mobile PC clients, and you either got one or the other. There were choices of
environment (Windows or Mac), choices of color (beige or black), and choices of vendor (Dell,
HP, Gateway, and so on), although this last was less of a choice than it looked because
essentially the same group of companies — Intel or AMD, ATI or nVidia, Seagate or Western
Digital —supplied the technology behind them all.
These clients (or endpoints) were sometimes called “fat” clients, but Microsoft has always
objected to this term, perhaps because it implies bloat, not exactly a brand-polishing concept.
“Rich” clients sounds better.
Originally, these rich clients were just complete computers, albeit microcomputers, that allowed
the user to do all computing task right there at the desk. All the rich resources required were
already in the box. But preceding the rich client was another paradigm, sometimes called
“terminal server,” which dated even farther back in the mists of time to the days of the mainframe
and minicomputer. Under this scheme, the terminal didn’t do much, just accepted keystrokes
from the user and passed them up by data communications to a big central computer, perhaps
shared by a number of users. The big computer, when the user’s slice of time came up, sent
back down the line the appropriate pixels to be painted on the screen. Most of the actual
computing was done in the big computer.
The terminals at the end of the line didn’t have much smarts and were thus sometimes called
“dumb terminals.” But a more elegant term evolved, particularly after the advent of client-server
computing: “thin clients.” So, now you had two parallel tracks, two computing schemas, one
involving rich clients and one involving thin clients — because after all, you can never be too rich
or too thin, as Wallis Simpson, the Duchess of Windsor, may once have said (whereas it’s easy
to be too fat and too dumb).
As the PC market developed, rich clients proliferated, outnumbering thin clients by more than
100 to 1. And there it might have rested, except for a few trends.
The most important was the evolution of bandwidth. As data communications became cheaper,
faster, more reliable, and widely available, various alternative schemas became more possible,
ones that would divide the computing task between the client at the edge and some computer or
computers at the center without as much of a performance penalty as in the past.
Also, security started to be more of an issue. Since the dawn of the 21st century, data owners
have become increasingly aware of the potential loss of data, loss of value, breach of trade
secrets, and theft of identity that can happen, either via malicious intent or by accident, with all
that data passing back and forth through the network and lying out at its edges in unprotected
hard drives.
Finally, there was the problem of complexity. Networks got built ad hoc. This and that were
added on as they became available or the need arose. Different architectures, generations of
technology, and applications were layered on like so much adobe, and things just got a lot
harder to manage.
So, with good connections, security problems, and too much complexity, folks began to take a
second look at computing models to see if there weren’t a better way. The need had arisen for
one or more schemes that could mitigate these issues, and various entrepreneurs and
established vendors rose to meet the challenges and take advantage of the opportunities.
Alternative Client Computing Paradigms
Over the past few years, a spectrum of alternative solutions has cropped up, such that it is no
longer appropriate to simply dismiss the category as "thin clients" and be done with it. The
flavors of thin, thin-ish, and semi-rich computing are many. The following paragraphs attempt to
sort them into major categories.
Terminal Server
The traditional thin client model, the oldest of the lot, involves the client doing very little other than
accept input from the user and display output from the server on the user’s local monitor.
Computing is done at the server end. Hardware at the client end is supplied by Wyse, NeoWare
(recently acquired by HP), and several others. Communications takes place via Remote
Desktop Protocol (RDP) to Microsoft’s Windows Terminal Services on the server. The good
news is all the valuable stuff is back at the server end. The bad news is performance isn't so
great. Terminal server is good for things like data entry. Terminal server can provide a good,
cost-effective answer for a situation, for example, in which a large number of task workers are co-
located in a service facility.
Diskless Computing
A semi-thin solution is typified by Ardence, which was recently acquired by Citrix, a supplier of
thin client software solutions. Diskless computing involves what amounts to a PC at the client
end, but the storage lies at the server end. The software redirects disk access requests to the
network transparently to the user. Data comes down the line as if were being summoned from
the disk. The bad news is that some degradation of performance occurs, but with gigabit
Ethernet, this shortcoming can be minimized. The good news is that, when the client is
powered down, all the data stays up at the server end, where it can be physically protected. A
typical customer for this type of solution would be a government lab, where data security is
paramount.
PC Blades
Next, we have PC blades, pioneered by ClearCube in 2001 and taken up in 2003 by HP as part
of its Consolidated Client Infrastructure (CCI) initiative. The two architectures differ somewhat,
but both have full PCs on blades in a rack in the datacenter and thin clients out in the field. The
two types of communications are brute force coax over short distances, which delivers decent
performance, and TCP/IP over Ethernet, which can handle long distances but gives poor
performance. CCI envisions a storage area network (SAN) behind the server pool, while
ClearCube virtualizes storage in the rack on the PC blades themselves. However, each side is
moving toward the other for competitive reasons. More recently, ClearCube, which is privately
held, has been having financial difficulties, which may cause it to abandon it solutions approach
for lack of economies of scale, and focus more directly on the software that manages the
blades. Avocent also plays in this space with a PCI card client insert that enables a standard
PC to function as a client-side device in a blade schema. Avocent has teamed with HP and Apro
to offer a full solution. The good news here is that the architecture is familiar to most IT
managers and has advantages in manageability and security. The bad news is that the cost of
the blade plus the thin “puck” at the client end is generally higher than a comparable PC and
doesn’t perform as well.
Hardware Compression
Most recently, a new wrinkle has been introduced: hardware graphics compression.
Essentially, working silicon now exists that takes output from a PC blade’s graphics accelerator
subsystem, compresses it, delivers it over standard TCP/IP links, and on the client side
recomposes it for display. This fix allows good performance over remote links for rich data,
giving the blade solutions a boost. More processing can be done at the server end and rich bits
can be sent to the client without sacrificing performance. This silicon is supplied by Teradici, a
Canadian company based in Vancouver, which has announced a number of partnership in the
past year, including tie-ups with ClearCube, IBM, Verari, and Leostream, and partnerships with
AMD, Ericom Software, nVidia, and Devon IT. Teradici calls its solution “PC over IP” or PCoIP.
Virtualization
All of this is getting muddied by virtualization, which has many benefits as well as some
associated Microsoft licensing issues. For example, virtualization allows blade PCs to be used
for more than one end user, improving the economics, but some of the flexibility implied by such
a scheme is subject to licensing restrictions Microsoft places on its customers. Virtualization
also has an impact on the storage pools that can either be aggregated on the blade rack and
allocated to users as needed or held in massive SANs behind the rack in the datacenter.
Winners and Losers
On the winner/loser scorecard, Wyse is pretty well positioned on the hardware end as the
largest independent supplier of thin clients. Ardence has done well for itself, including being
acquired by Citrix. Teradici has some key intellectual property. ClearCube has a good software
portfolio related to failover switching and remote client management as well as virtualization, but
is struggling with scale on the hardware end. HP seems to be treading water with CCI, although
it has announced a few wins from time to time. Dell recently entered the market with its Flexible
Computing Solutions offering, based on the Citrix-Ardence technology. No word on how that’s
going yet. Microsoft is moving cautiously over licensing in this environment; case in point, the
company changed its policies earlier this year, issuing a cryptic note that appears to relate to
permitting Ardence-type solutions. IBM is happy to promote thin computing because all it really
wants to do is sell servers. Lenovo has licensed ClearCube or is a reseller, but does not seem
to be running with the ball very hard. Avocent also seems to be dancing around the edge of this
space, but is weakened by its dependence on partners.
Where to From Here?
Thin computing has been forecast to grow for eons, but has consistently under-delivered. The
main reason is inertia. People like to work with what they are familiar with, and Microsoft and
Intel have them pretty well trained on PCs. However, the theory goes: the argument in favor of
thin and thin-ish computing gets better as bandwidth becomes more reliable, more available,
and faster. From a marketing perspective, thin clients work best in a self-contained installation
where the vendor "guarantees" that it will work. IBM has done fairly well with this services-
oriented approach. HP has tried it as well.
Given the dynamics, no reason exists for thin clients not to become increasingly popular. In the
extreme case, why not have all your computing done safely in the glass house if users are none
the wiser? The logical conclusion of this thinking leads to a cloud-computing world such as the
one envisioned by Google.
But mobile computing necessarily involves some intermittent communication, and people still
want the full experience even when they're not connected. Thus, rich clients are likely to
dominate for years to come, particularly as the tide turns in favor of mobility. But as even mobile
clients become connected most of the time through advanced, high-bandwidth wireless
connections like WiMAX, the pendulum may swing again in favor of thin computing. The growth
rate of alternative client shipments has exceed that of PCs for years. It's just that the base is so
small that alternative client shipments are still two orders of magnitude smaller than PC
shipments.
© 2007 Endpoint Technologies Associates, Inc. All rights reserved.
Too Fat or Too Dumb?
 |
| |
 |