I know RAID is overkill for most workstations, but so is a DDS drive and seperate home and system drives. If you want fault tolerance, (the stated reason for two drives) having one system drive and one home drive with no RAID means you spend your money only to become twice as vulnerable to downtime due to drive failures.
If you want to avoid downtime, especially if money is no object, get a RAID controller and have a single filesystem mirrored over two physical disks. Not only will it be more reliable, it will be faster too.
This is the perspective of someone with a couple of years experience in "enterprise" development, i.e. apps that have a UI that is probably a bunch of data entry forms, a database at the back, and a "business logic" layer that looks at the input, massages it, acts on it, sends it to the db, etc.
These sorts of apps are not very sexy compared to writting an OS or a game or something, but they're what 90% of developers in the world work on.
A typical enterprise app will need to have a bunch of different front ends (e.g. web interface, a simple gui for data entry clerks, a bunch of reports) etc. Also, you'll have lots of people working on different bits and peices, and lots of changes going on all the time, e.g.say a sales manager comes up with a new promotional scheme that gives volume discounts you need to update your app to handle the calculations.
So with all these people working on apps that are evolving rapidly, you can end up with spaghettii code pretty quickly. If you want to have something that's maintainable, you really need to use OO and come up with an object model that seperates the user interface, the business rules, and the data layer.
The problem with this is that it is hard to have an object model that works well over distributed systems, and hard to have an object model that can be used by developers using multiple languages.
COM is a start to allowing objects to be used from multiple objects, and over distributed systems, but it has limitations, largely related to the fact that different languages don't have the same idea of basic data types.
.NET solves this by making all languages share a virtual machine that defines a bunch of basic data types, and a base 'object'. This means that any object created in one.NET language can be accessed by another.NET language.
So you can have a your web front end people write ASP (VB) pages that interact with business logic written in c# without having to compromise your object model.
Having admined both Linux and NT/W2K for a few years, I don't think there is much difference in the ability to script in Linux or W2k.
E.g. here's how to setup a new user in VBScript.
-----
dim strUserName
set strUserName = "TheNewUser"
set adsDomain = GetObject("WinNT://MYDOMAIN")
set adsUser = adsDomain.Create("user",strUserName)
adsUser.SetInfo
------
BUT this is obviously nothing like how you create a new user through the GUI. So it probably seems more complicated.
It's a much smaller cognitive jump typing 'adduser foo' at a #prompt to using that command in a bash script.
The article goes out of it's way to be seen to be 'fair' to NT, and then ends with a comparison that shows the support cost for adding a 5000 mailbox solution to an existing mainframe is exactly $0. Presumably this is because it is assumed the site allready has mainframe support resources?
But by the same token a site that uses NT for file & print servers (and therefore has an existing NT support team) should be able to use the same support resources for looking after their Exchange servers.
I'm not saying that there would be NO increase in support demands going from NT file & print to NT file & print plus Exchange, but then I don't believe that adding 5,000 whinging email users won't affect the workload of a mainframe support team either.
So to do a comparison, you should either add support costs to both NT and Mainframe, or neither. Doing it to just one is very misleading.
The only bit of information that is stored only on the Passport server is the user's password.
So if a site wants to back out of passport, they could just email all their users a new temporary password.
No big deal. All the other bits of info (full name, age, whatever the user wanted to tell the site) are stored on the site's servers, not Passport. Info put into passport is merely used to prefill forms the first time the site requires that info.
So if a hypothetical MS tax is too much, backing out means nothing more than a more cumbersome login for new users, and a change of password for existing users.
Actually I changed my mind - Passport does change things slightly.
The problem with aggregagating user transactions across multiple sites is matching up user accounts on one site with user accounts on another. DoubleClick solved this by using cookies, but (at least on single user Win9x boxen) identify a machine only, not a user, i.e. they can't detect multiple users of one machine or someone who uses lots of machines.
What passport does is make people use the same account ID at all sites (i.e. their email address).
Passport sites aren't the only sites that do this, e.g. safari.oreilly.com uses your email address as the login, as does amazon. So if Oreilly and Amazon wanted to match up the userbase to see what other books safari users purchased, they could quite easily. It would be a bit harder for Oreilly and SlashDot to match users however, since the login on slashdot is NOT your email address. But slashdot, like most sites, does still collect an email so matching would still be possible.
They way passport changes things a little is that people with multiple emails are more likely to use the same address on all sites, and less likely to give bodgey email addresses. So matching will be (a little bit) more reliable.
Passport doesn't collect transactions from affiliated sites.
There is no way that MS will know that you bought Rat Poison from one passport using site, and Lingerie from another.
Well, let me rephrase that. There are plenty of ways that that kind of information can be collected (i.e. through doubleclick and similar user-info-swapping deals) but Passport doesn't alter the equation.
There is a common misunderstanding here, passport is not the sole repository of all data for all sites who want to use passport. Each site collects and maintains it's own info.
If you have a look at the passport SDK, you'll see that the affiliated sites don't have direct access to any of the user's data.
A site that wants to use Passport for SSO generates an URL that redirects to the passport website. Then the user logs in, and passport redirects back to the original site. The original site can then access the authenticated username, but that's it.
When the site wants to get some data from the user, say the user's age or address, they don't query passport directly. What they do is redirect back to passport, passport generates a form with the values prefilled in. Then the user can edit those values, or just click submit, and the values are posted back to the original site.
So as a user you still get full control over what data a site you visit has. And you can tell a particular site info that is different to what is stored in passport. But it does save you typing in the same old boring gumpf into site after site.
Australian companies (webcentral.com.au, secureip.net.au) will let you resell website hosting on a per site basis. i.e. you pay them say $25/month for a single website. Then you build a site for a customer, charge them for the development, and say $35/month to host it. Or maybe $100/month to host it and do some regular updates to it.
or you can buy the hardware and run it yourself. That would be more fun, but the economies of scale are all against you.
but if you want to do this, what you really need to do is work out a marketing plan. Whether you decide it's cheaper to do stuff in house, or outsource, you still need to find some customers, otherwise you'll find yourself using your shiney new server farm for nothing but reading slashdot all day, waiting for the phone to ring.
I've worked for a few small ISP's in Australia, so this is my take on the economics.
There are two ways of setting up as an ISP. The 'real' isp way means buying a pipe to telstra (ISDN/Frame Relay), and a port concentrator (or just a bunch of modems hanging of a linux box) plus at least one box running mail & RADIUS. Apart from the hardware, your costs will be about $300/month to lease a 64k ISDN line (this is for a metro link, could go up to $900/month for rural areas) plus 18c/MB of traffic downloaded by your users. Then you need about $25/month for each inbound line.
Budget ISP's can often get away with 10 inbound lines sharing a single 64K uplink.
As you can see, there are some pretty high fixed costs, so you need to have a reasonable number of users to amortise those costs over. And even then the margins are pretty slim.
An alternative is to become a 'virtual' ISP. This means going to a company like Vivanet (www.vivanet.com.au) or PSINet who allready have a POP in your area. You then pay around $100-$150/month for each modem channel. If you offer unlimited dialup accounts, you will probably need about 1 modem channel for each 8 accounts you sell. You still need to provide a RADIUS/Mail server somewhere. Hosting one of these boxes in a datacenter will be around $200-$500/month depending on the size of the box, how much traffic goes through it etc.
Re broadband, if telstra offers ADSL in your area, then any one of a number of wholesale ISP's (including VIVANet) will also let you resell ADSL there.
If you do the numbers you'll see there really isn't much money to be made selling bandwidth. But
one factor I haven't mentioned is that for many small ISP's, selling bandwidth is not the primary profit maket. For Small businesses, their ISP often becomes their defacto IT department, so ISP's can generate good revenue from setting up LAN's, buildings website's, basic DB development etc.
If Wine or DOSEmu can't handle it alone, I'd wager that having an install of Win9x or MS-DOS handy for these programs to access would fix the problem
Depends what the problem is:-) The problem RIS solves is how to make installing an OS image (inluding applications) onto a box so easy that any user can be guided through the process over the telephone, AND the process can be completed in 15 minutes, AND the end user doesn't need to have any boot floppies etc.
So when a user messes up their PC, telephone helpdesk dudes spend 5 minutes starting the rebuild process instead of 2 hours trying to untangle whatever the user has done, and 15 minutes later the user can log in to their PC. And (if you are using some of the other nifty features of W2k) all their apps will be available and configured for the user also.
You can certainly do unattended setups from any distribution media (CD-ROM, NT server, Samba, even a Netware server if you want) but there are some big headaches with scaling these approaches.
Windows 2k Server allows unattended installation of windows 2k clients without the need for install media or boot floppys. The service is called RIS (Remote Installation Service) and it uses PXE compatable NIC's. These NIC's have a dhcp/tftp client. The theory is you boot one of these computers, during the boot up you press F8, you are prompted for a user name & password, then the an OS image is TFTPed to the local machine.
Although there may be linux distro's that support unattended setup of file servers, and you can also use PXE to boot diskless linux workstations, as far as I know there is no linux distro that will support unattended setup of windows 2k pro workstations in this way.
....I have found (at least in.au) is the Intel 1100 series. This is actually a 1RU case + motherboard + 2xIntel Pro 100 NIC's. Add your own HD/RAM/CPU/OS. You can also add a video card or CDROM if you need to, but these things are designed for headless deployment (Ghost/PXE/etc) and they have a serial port on the front so you can plug in a vt100 console if you want to see the bios or text mode/dos stuff.
I have built some linux servers, PIII plus 128MB ram plus 9GB IDE drive came to about $2300 AUD so I would guess given exchange rates etc would be around $1K US.
Slashdot Mirror
Am I missing something?
I know RAID is overkill for most workstations, but so is a DDS drive and seperate home and system drives. If you want fault tolerance, (the stated reason for two drives) having one system drive and one home drive with no RAID means you spend your money only to become twice as vulnerable to downtime due to drive failures.
If you want to avoid downtime, especially if money is no object, get a RAID controller and have a single filesystem mirrored over two physical disks. Not only will it be more reliable, it will be faster too.
This is the perspective of someone with a couple of years experience in "enterprise" development, i.e. apps that have a UI that is probably a bunch of data entry forms, a database at the back, and a "business logic" layer that looks at the input, massages it, acts on it, sends it to the db, etc.
.NET language can be accessed by another .NET language.
These sorts of apps are not very sexy compared to writting an OS or a game or something, but they're what 90% of developers in the world work on.
A typical enterprise app will need to have a bunch of different front ends (e.g. web interface, a simple gui for data entry clerks, a bunch of reports) etc. Also, you'll have lots of people working on different bits and peices, and lots of changes going on all the time, e.g.say a sales manager comes up with a new promotional scheme that gives volume discounts you need to update your app to handle the calculations.
So with all these people working on apps that are evolving rapidly, you can end up with spaghettii code pretty quickly. If you want to have something that's maintainable, you really need to use OO and come up with an object model that seperates the user interface, the business rules, and the data layer.
The problem with this is that it is hard to have an object model that works well over distributed systems, and hard to have an object model that can be used by developers using multiple languages.
COM is a start to allowing objects to be used from multiple objects, and over distributed systems, but it has limitations, largely related to the fact that different languages don't have the same idea of basic data types.
.NET solves this by making all languages share a virtual machine that defines a bunch of basic data types, and a base 'object'. This means that any object created in one
So you can have a your web front end people write ASP (VB) pages that interact with business logic written in c# without having to compromise your object model.
Having admined both Linux and NT/W2K for a few years, I don't think there is much difference in the ability to script in Linux or W2k.
E.g. here's how to setup a new user in VBScript.
-----
dim strUserName
set strUserName = "TheNewUser"
set adsDomain = GetObject("WinNT://MYDOMAIN")
set adsUser = adsDomain.Create("user",strUserName)
adsUser.SetInfo
------
BUT this is obviously nothing like how you create a new user through the GUI. So it probably seems more complicated.
It's a much smaller cognitive jump typing 'adduser foo' at a #prompt to using that command in a bash script.
The article goes out of it's way to be seen to be 'fair' to NT, and then ends with a comparison that shows the support cost for adding a 5000 mailbox solution to an existing mainframe is exactly $0. Presumably this is because it is assumed the site allready has mainframe support resources?
But by the same token a site that uses NT for file & print servers (and therefore has an existing NT support team) should be able to use the same support resources for looking after their Exchange servers.
I'm not saying that there would be NO increase in support demands going from NT file & print to NT file & print plus Exchange, but then I don't believe that adding 5,000 whinging email users won't affect the workload of a mainframe support team either.
So to do a comparison, you should either add support costs to both NT and Mainframe, or neither. Doing it to just one is very misleading.
The only bit of information that is stored only on the Passport server is the user's password.
So if a site wants to back out of passport, they could just email all their users a new temporary password.
No big deal. All the other bits of info (full name, age, whatever the user wanted to tell the site) are stored on the site's servers, not Passport. Info put into passport is merely used to prefill forms the first time the site requires that info.
So if a hypothetical MS tax is too much, backing out means nothing more than a more cumbersome login for new users, and a change of password for existing users.
Actually I changed my mind - Passport does change things slightly.
The problem with aggregagating user transactions across multiple sites is matching up user accounts on one site with user accounts on another. DoubleClick solved this by using cookies, but (at least on single user Win9x boxen) identify a machine only, not a user, i.e. they can't detect multiple users of one machine or someone who uses lots of machines.
What passport does is make people use the same account ID at all sites (i.e. their email address).
Passport sites aren't the only sites that do this, e.g. safari.oreilly.com uses your email address as the login, as does amazon. So if Oreilly and Amazon wanted to match up the userbase to see what other books safari users purchased, they could quite easily. It would be a bit harder for Oreilly and SlashDot to match users however, since the login on slashdot is NOT your email address. But slashdot, like most sites, does still collect an email so matching would still be possible.
They way passport changes things a little is that people with multiple emails are more likely to use the same address on all sites, and less likely to give bodgey email addresses. So matching will be (a little bit) more reliable.
Passport doesn't collect transactions from affiliated sites.
There is no way that MS will know that you bought Rat Poison from one passport using site, and Lingerie from another.
Well, let me rephrase that. There are plenty of ways that that kind of information can be collected (i.e. through doubleclick and similar user-info-swapping deals) but Passport doesn't alter the equation.
There is a common misunderstanding here, passport is not the sole repository of all data for all sites who want to use passport. Each site collects and maintains it's own info.
If you have a look at the passport SDK, you'll see that the affiliated sites don't have direct access to any of the user's data.
A site that wants to use Passport for SSO generates an URL that redirects to the passport website. Then the user logs in, and passport redirects back to the original site. The original site can then access the authenticated username, but that's it.
When the site wants to get some data from the user, say the user's age or address, they don't query passport directly. What they do is redirect back to passport, passport generates a form with the values prefilled in. Then the user can edit those values, or just click submit, and the values are posted back to the original site.
So as a user you still get full control over what data a site you visit has. And you can tell a particular site info that is different to what is stored in passport. But it does save you typing in the same old boring gumpf into site after site.
That's because the site has been hacked by a worm (similar to, but predating, Code Red). the CERT advisory on this is at http://www.cert.org/advisories/CA-2001-11.html
Australian companies (webcentral.com.au, secureip.net.au) will let you resell website hosting on a per site basis. i.e. you pay them say $25/month for a single website. Then you build a site for a customer, charge them for the development, and say $35/month to host it. Or maybe $100/month to host it and do some regular updates to it.
or you can buy the hardware and run it yourself. That would be more fun, but the economies of scale are all against you.
but if you want to do this, what you really need to do is work out a marketing plan. Whether you decide it's cheaper to do stuff in house, or outsource, you still need to find some customers, otherwise you'll find yourself using your shiney new server farm for nothing but reading slashdot all day, waiting for the phone to ring.
There are two ways of setting up as an ISP. The 'real' isp way means buying a pipe to telstra (ISDN/Frame Relay), and a port concentrator (or just a bunch of modems hanging of a linux box) plus at least one box running mail & RADIUS. Apart from the hardware, your costs will be about $300/month to lease a 64k ISDN line (this is for a metro link, could go up to $900/month for rural areas) plus 18c/MB of traffic downloaded by your users. Then you need about $25/month for each inbound line.
Budget ISP's can often get away with 10 inbound lines sharing a single 64K uplink.
As you can see, there are some pretty high fixed costs, so you need to have a reasonable number of users to amortise those costs over. And even then the margins are pretty slim.
An alternative is to become a 'virtual' ISP. This means going to a company like Vivanet (www.vivanet.com.au) or PSINet who allready have a POP in your area. You then pay around $100-$150/month for each modem channel. If you offer unlimited dialup accounts, you will probably need about 1 modem channel for each 8 accounts you sell. You still need to provide a RADIUS/Mail server somewhere. Hosting one of these boxes in a datacenter will be around $200-$500/month depending on the size of the box, how much traffic goes through it etc.
Re broadband, if telstra offers ADSL in your area, then any one of a number of wholesale ISP's (including VIVANet) will also let you resell ADSL there.
If you do the numbers you'll see there really isn't much money to be made selling bandwidth. But one factor I haven't mentioned is that for many small ISP's, selling bandwidth is not the primary profit maket. For Small businesses, their ISP often becomes their defacto IT department, so ISP's can generate good revenue from setting up LAN's, buildings website's, basic DB development etc.
If Wine or DOSEmu can't handle it alone, I'd wager that having an install of Win9x or MS-DOS handy for these programs to access would fix the problem Depends what the problem is :-) The problem RIS solves is how to make installing an OS image (inluding applications) onto a box so easy that any user can be guided through the process over the telephone, AND the process can be completed in 15 minutes, AND the end user doesn't need to have any boot floppies etc.
So when a user messes up their PC, telephone helpdesk dudes spend 5 minutes starting the rebuild process instead of 2 hours trying to untangle whatever the user has done, and 15 minutes later the user can log in to their PC. And (if you are using some of the other nifty features of W2k) all their apps will be available and configured for the user also.
You can certainly do unattended setups from any distribution media (CD-ROM, NT server, Samba, even a Netware server if you want) but there are some big headaches with scaling these approaches.
Although there may be linux distro's that support unattended setup of file servers, and you can also use PXE to boot diskless linux workstations, as far as I know there is no linux distro that will support unattended setup of windows 2k pro workstations in this way.
....I have found (at least in .au) is the Intel 1100 series. This is actually a 1RU case + motherboard + 2xIntel Pro 100 NIC's. Add your own HD/RAM/CPU/OS. You can also add a video card or CDROM if you need to, but these things are designed for headless deployment (Ghost/PXE/etc) and they have a serial port on the front so you can plug in a vt100 console if you want to see the bios or text mode/dos stuff.
I have built some linux servers, PIII plus 128MB ram plus 9GB IDE drive came to about $2300 AUD so I would guess given exchange rates etc would be around $1K US.