Guaranteeing 95% uptime of most anything is a rather easy task. React in less than 12 hours and you got your lower back covered for three outages per month. No need for extra staff and that kind of stuff – almost free.
Raising the bar to 99% makes the thing quite a bit more difficult. Still, seven hours of reacton time at one outage per month is realistic even for motivated individuals with somewhat reliable software – some work necessary.
Jack that up to 99.9% and it’s a bit more troubling. A workday of outages per /year/ isn’t that easy to accomplish; you’ll probably need someone on pager duty all the time and might want a second server to run your stuff on – both cost money.
Skipping four and going right for five nines is a whole different beast. We’re talking about five minutes of downtime per year. A decade of uptime with less than an hour of unavailability. You aren’t going to accomplish this with a single guy, a pager and two or three servers; instead you’ll need competent staff all around the clock monitoring your redundant server clusters distributed throughout several continents connected to various tier 1 providers. If you’re not seeing a pattern, let me give you a clue about the implications of that: Money. Cold hard cash. You’ll not only pay your staff and their offices but your uplink providers’ staff, your datacenters’ staff, their respective bosses’ boss’ boni and whatnot.

As a provider of anything, you’ll pass your expenses (eagerly) and savings (not so eagerly) on to your customers, creating what’s called a market. As such a customer, I, for one, usually don’t require five nines and will happily save 75% (or more) by reducing to 99.9% (or less). I’ll gladly trade a day of cell phone down time for another week of holidays and a day of not having a rental car available for a nice meal. There’s always landlines, phone booths, taxis and hitchhiking.

Guaranteeing 95% uptime of most anything is a rather easy task. React in less than 12 hours and you got your lower back covered for three outages per month. No need for extra staff and that kind of stuff – almost free.
Raising the bar to 99% makes the thing quite a bit more difficult. Still, seven hours of reacton time at one outage per month is realistic even for motivated individuals with somewhat reliable software – some work necessary.
Jack that up to 99.9% and it’s a bit more troubling. A workday of outages per /year/ isn’t that easy to accomplish; you’ll probably need someone on pager duty all the time and might want a second server to run your stuff on – both cost money.
Skipping four and going right for five nines is a whole different beast. We’re talking about five minutes of downtime per year. A decade of uptime with less than an hour of unavailability. You aren’t going to accomplish this with a single guy, a pager and two or three servers; instead you’ll need competent staff all around the clock monitoring your redundant server clusters distributed throughout several continents connected to various tier 1 providers. If you’re not seeing a pattern, let me give you a clue about the implications of that: Money. Cold hard cash. You’ll not only pay your staff and their offices but your uplink providers’ staff, your datacenters’ staff, their respective bosses’ boss’ boni and whatnot.

As a provider of anything, you’ll pass your expenses (eagerly) and savings (not so eagerly) on to your customers, creating what’s called a market. As such a customer, I, for one, usually don’t require five nines and will happily save 75% (or more) by reducing to 99.9% (or less). I’ll gladly trade a day of cell phone down time for another week of holidays and a day of not having a rental car available for a nice meal. There’s always landlines, phone booths, taxis and hitchhiking.

Guaranteeing 95% uptime of most anything is a rather easy task. React in less than 12 hours and you got your lower back covered for three outages per month. No need for extra staff and that kind of stuff – almost free.
Raising the bar to 99% makes the thing quite a bit more difficult. Still, seven hours of reacton time at one outage per month is realistic even for motivated individuals with somewhat reliable software – some work necessary.
Jack that up to 99.9% and it’s a bit more troubling. A workday of outages per /year/ isn’t that easy to accomplish; you’ll probably need someone on pager duty all the time and might want a second server to run your stuff on – both cost money.
Skipping four and going right for five nines is a whole different beast. We’re talking about five minutes of downtime per year. A decade of uptime with less than an hour of unavailability. You aren’t going to accomplish this with a single guy, a pager and two or three servers; instead you’ll need competent staff all around the clock monitoring your redundant server clusters distributed throughout several continents connected to various tier 1 providers. If you’re not seeing a pattern, let me give you a clue about the implications of that: Money. Cold hard cash. You’ll not only pay your staff and their offices but your uplink providers’ staff, your datacenters’ staff, their respective bosses’ boss’ boni and whatnot.

As a provider of anything, you’ll pass your expenses (eagerly) and savings (not so eagerly) on to your customers, creating what’s called a market. As such a customer, I, for one, usually don’t require five nines and will happily save 75% (or more) by reducing to 99.9% (or less). I’ll gladly trade a day of cell phone down time for another week of holidays and a day of not having a rental car available for a nice meal. There’s always landlines, phone booths, taxis and hitchhiking.

Guaranteeing 95% uptime of most anything is a rather easy task. React in less than 12 hours and you got your lower back covered for three outages per month. No need for extra staff and that kind of stuff – almost free.
Raising the bar to 99% makes the thing quite a bit more difficult. Still, seven hours of reacton time at one outage per month is realistic even for motivated individuals with somewhat reliable software – some work necessary.
Jack that up to 99.9% and it’s a bit more troubling. A workday of outages per /year/ isn’t that easy to accomplish; you’ll probably need someone on pager duty all the time and might want a second server to run your stuff on – both cost money.
Skipping four and going right for five nines is a whole different beast. We’re talking about five minutes of downtime per year. A decade of uptime with less than an hour of unavailability. You aren’t going to accomplish this with a single guy, a pager and two or three servers; instead you’ll need competent staff all around the clock monitoring your redundant server clusters distributed throughout several continents connected to various tier 1 providers. If you’re not seeing a pattern, let me give you a clue about the implications of that: Money. Cold hard cash. You’ll not only pay your staff and their offices but your uplink providers’ staff, your datacenters’ staff, their respective bosses’ boss’ boni and whatnot.

As a provider of anything, you’ll pass your expenses (eagerly) and savings (not so eagerly) on to your customers, creating what’s called a market. As such a customer, I, for one, usually don’t require five nines and will happily save 75% (or more) by reducing to 99.9% (or less). I’ll gladly trade a day of cell phone down time for another week of holidays and a day of not having a rental car available for a nice meal. There’s always landlines, phone booths, taxis and hitchhiking.