2011 June 15 archive at Robert Pogson

Archive for June 15th, 2011

Slavery to M$

Published by Robert Pogson June 15th, 2011 in technology. 2 Comments

There are some articles on the web describing what businesses who feel locked in to M$ have to pay to adopt “7″. Some highlights:

For every 10 users of IT in the system there is approximately one application that needs to be checked for compatibility with “7″ or replaced.
Typical costs for automated checking are $100-200 per application and much more for manual testing.
Molex found 650 applications in their system and eventually chose 341 old and new applications to have on the new system with “7″.
Sifting through the applications can take months, perhaps a few days for each application.
Sifting through all the compatibility issues and details can take months.
Total cost of software compatibility work can cost $100-$200 per application per PC.

I am glad schools have simpler IT, just a few apps per PC and they can be all identical. This example shows the huge hole businesses have dug for themselves by depending on M$ for a platform. Any time M$ and its “partners” want money from businesses, they only have to wreck compatibility of apps and money will flow in with new hardware purchases and new software purchases. What a racket. With FLOSS they could keep their software indefinitely if they wish and just recompile it if some parts are changed.

In the case of Molex, they found almost half the apps that locked them in to M$ were not even required. Worse, this game must be repeated at almost every medium to large business on the planet. With FLOSS, the distros do it once for the whole world of users. That’s much more efficient.

You can read that sorry tale, 3 tips for migrating applications to Windows 7

To get a measure of the lockin of a US State see State of Montana – Biennial Report for Information Technology – 2009

Only 13% of their applications are “commercial”. 31% are custom-made for their unique processes and 49% are databases. They have 14814 PCs under a policy of replacement every four years. They have 850 IT staff. They plan to spend $107million over the next two years for a long list of major projects in IT. They have to sanitize or destroy every hard drive before disposal.

They actually considered migrating just file and print services to GNU/Linux and decided against it in 2006, reasoning, We have the expertise onboard to support the Windows product. We know we can attract technicians with Windows-based skills. Recruitment of highly skilled Linux support technicians would prove to be more of a challenge than recruiting Windows technicians…
and they chose to be dependent on Active Directory rather than OpenLDAP.

Meanwhile, the Legislative Branch has no problem using GNU/Linux. They are rolling it in for file/print serving. They are also migrating to “7″ without much question. Their budget is $2.5million and 1/3 of that is hardware/software replacement costs on a four year cycle. They are considering thin clients, though.

So, the three examples show that businesses and large organizations have been working slavishly for decades making dependence on M$ and that other OS stronger. For all the money they spend globally, they could produce FLOSS and keep their software in-house. They seem to have the gambler’s addiction and a belief that repeating mistakes can bring a successful conclusion. Lock-in in IT is a mistake. It costs flexibility and higher costs indefinitely. FLOSS may have a serious cost for migration but the on-going costs are trivial in comparison. Changing PCs every 4 years whether they are worn out or not is the height of stupidity. Thin clients have no moving parts these days and can last 10 years. They can run any software on a server that these folks need to run. The Legislative Branch of Montana has 380 PCs. If they used thin clients for 10 years it would cost them about $30/year/PC. They are spending 8000=$800000/380/4 times as much as they need and they plan to do that forever…

- Robert Pogson

Donovan Colbert: Microsoft lost mostly on perception over the last 10 years.

Published by Robert Pogson June 15th, 2011 in technology. 11 Comments

Just when I begin to think technological evangelism from M$ has faded, a new wave of even larger lies emerges. We’ve had the browser wars, Get the FUD, SCOG v World, and now we have Donovan Colbert writing, “Microsoft lost mostly on perception over the last 10 years.” Continue reading ‘Donovan Colbert: Microsoft lost mostly on perception over the last 10 years.’

- Robert Pogson

The “Wrong” Way to do Virtual Desktops

Published by Robert Pogson June 15th, 2011 in Uncategorized. 17 Comments

The concept of virtualizing desktops is wonderful. Putting everything in one place so that it is easier to manage and uses less hardware/power/maintenance overall is great.

There is a “wrong” way to do it, though. The form of virtualization of the desktop often promoted by businesses who will benefit by massive sales of hardware and software is wrong, IMHO.

Putting a virtual machine per desktop anywhere wastes RAM everywhere,
putting those virtual machines on a server and shoving it all over to the PC when it boots is insane, and
keeping everything on the server but running the software on the thin client chokes networks at boot time.

Why do people keep trying to invent the wheel? Good old-fashioned X and GNU/Linux offer a great solution:

Use a single OS image for each class of users on the server/cluster,
run all processes on the server running GNU/Linux, a true multi-user/multi-tasking OS that caches files like a beautiful woman gathers admirers, and
boot clients PXE for lowest CAPEX, or
use a minimal embedded GNU/Linux OS on the thin client for fastest boot-up on large deployments.

That’s so simple and it works. Why make life more complicated? GNU/Linux can handle the complexity of any organization. User permissions and user/group classification sort everything out. Use OpenLDAP for large deployments. Use a common login screen for everyone and let the users’ permissions and $HOME directory sort out their desktop layouts, menus, and resources.

The arguments against this tend to fall along a few lines:

Computing on the server is a bit more expensive in terms of hardware but it turns out to be a tiny price, $25 per user, for tiny systems and perhaps a bit more or less for large systems. You can run 400 users on a $40K server as Largo, FL, does but you can run 40 users on a $1K server as I do. That’s still tiny compared to the price difference between a thick client/VDI thin client ($hundreds) and a light thin client ($50).
There is a load on the network. For point, click and gawk it amounts to less than 1 M bytes/s per client. For video, the sky’s the limit. Do you really need full-screen video at work on every client? Not in a school. Not in an office. You can budget for this and provide the network that fits your users/loads, or you can use thicker clients for the few that need video or you can use a TV… Keeping data and executables off the LAN more than compensates for a humane rate of screen refreshes (often partial screen refreshes).
There are still those who claim thin client systems are slow. I can make a thin client system slow by using 10base-T or having too many users on a CPU or using a single old ATA drive for all users’ files, but why would I do that? Just eliminate bottle-necks and the system flies. Use RAID 1 with multiple heads seeking and your users will not have to wait. Cache all the files needed to login, open an application or window and the system will scream. My users typically see 10s logins to a usable desktop and less than 2s to open the largest application, OpenOffice.org/LibreOffice. They get this performance from thin clients with 400 MHz CPUs. They get the performance of a GNU/Linux terminal server with 64bitness, huge RAM and RAID and multiple gigabit/s NICs. Put your money on the server, not the clients and the performance can far exceed what the COTS thick client will do.
Believe it or not there are still a few people who claim thin client systems cost more than thick client systems… I can build a thin client system that costs more than a thick client system by using new ATX PCs with M$’s tax as thin clients, using gigabit/s to each thin client, using an over-sized server/cluster, or wasting the resources of the server using a mess of virtual machines and that other OS on it, but why would I? Use minimal RAM on each thin client and save $20 or so per client. Use no hard drive, just a flash memory at most on a thin client, or even no storage at all and boot PXE, and save $40 per client. Use an ARMed CPU instead of x86/AMD64 and save $50 per client. Use a tiny case or one that bolts onto the back of a monitor to save $50 or more per client. Replace the thin client every 10 years instead of 3 years and save a bundle. I can easily build a better-performing system for half the cost using GNU/Linux and thin clients. The typical point, click and gawk crowd are utilizing their CPUs less than 10% of the time. It makes sense to put that CPU utilization on the server and using a much less expensive client. Savings multiplied over a few clients or thousands adds up to important savings. Done right, the overall system will be much cheaper. Suppose you install 100 thick clients and a file server. With that other OS, you can easily spend $300 or more per client and $7000 on the server ($5K licence + $2K hardware), a total of $37000. With GNU/Linux thin clients ($100) and a more robust server ($7000 for hardware), you would spend only $17000 and get better performance. You save power, too. Thick clients = 400W for server and 100W * 100 for clients = 10.4KW. Thin clients = 1KW for server + 10W*100 for clients = 2KW. You save 8KW all day long. You also get to throw away much less at decommissioning, about 10Kg less per client.

Any place that has bunches of users with similar requirements can be served well with good old X and GNU/Linux thin clients. It costs less and performs better than many other solutions.

To read about how badly virtualized desktops can get done read Steer clear of the desktop virtualisation bootstorm

- Robert Pogson