How to increase web server uptime with DNS failover

[Guy Macon]

How to increase web server uptime with DNS failover:

For illustration, let's start with some really bad hosting...

Find four free web hosts that are each 90% reliable
-- in other words they each are down on average one
day out of ten. Use a failover DNS system with monitoring
to switch servers if one fails. On average:

One server will be down one day out of ten

Two servers will be down one day out of a hundred

Three servers will be down one day out of a thousand

All four servers -- and thus your website -- will be
down at the same time one day out of ten thousand.

Two servers with 99% uptime will also be down at the same
time one day out of ten thousand. Three at 99% will be
down at the same time one day out of a million.

In addition, there will be a delay for some users before
the new server is switched in. One website measured this
as follows:

After 5 minutes, 3% of visitors see the new server.
After 10 minutes, 20% of visitors see the new server.
After 15 minutes, 37% of visitors see the new server.
After 20 minutes, 59% of visitors see the new server.
After 25 minutes, 69% of visitors see the new server.
After 30 minutes, 76% of visitors see the new server.
After 35 minutes, 80% of visitors see the new server.
After 45 minutes, 90% of visitors see the new server.
After 90 minutes, 95% of visitors see the new server.
After 18 hours, 99% of visitors see the new server.

With any number of servers at 90% uptime, this will
happen on average every ten days. With any number of
servers at 99% uptime, this will happen on average
every hundred days.

Needless to say, this works even better if you start
with high-uptime paid servers.

References:
http://blog.pyromod.com/2007/09/viab...-failover.html
http://www.webhostingtalk.com/showthread.php?t=524788
http://www.dnsmadeeasy.com/s0306/prod/dnsfosm.html
http://www.georedundanthosting.com/p...t-hosting.aspx
http://edgedirector.com/app/back.htm
http://www.webstrikesolutions.com/Page_ID.aspx?IDTXT=36


--
Guy Macon <http://www.GuyMacon.com/Guy Macon <http://www.GuyMacon.com/>
Guy Macon <http://www.GuyMacon.com/Guy Macon <http://www.GuyMacon.com/>
Guy Macon <http://www.GuyMacon.com/Guy Macon <http://www.GuyMacon.com/>
Guy Macon <http://www.GuyMacon.com/Guy Macon <http://www.GuyMacon.com/>

[Guy Macon]

Guy Macon <http://www.GuyMacon.com/wrote:

>
How to increase web server uptime with DNS failover:

For illustration, let's start with some really bad hosting...

Find four free web hosts that are each 90% reliable
-- in other words they each are down on average one
day out of ten. Use a failover DNS system with monitoring
to switch servers if one fails. On average:

One server will be down one day out of ten

Two servers will be down one day out of a hundred

Three servers will be down one day out of a thousand

All four servers -- and thus your website -- will be
down at the same time one day out of ten thousand.

Two servers with 99% uptime will also be down at the same
time one day out of ten thousand. Three at 99% will be
down at the same time one day out of a million.

In addition, there will be a delay for some users before
the new server is switched in. One website measured this
as follows:

After 5 minutes, 3% of visitors see the new server.
After 10 minutes, 20% of visitors see the new server.
After 15 minutes, 37% of visitors see the new server.
After 20 minutes, 59% of visitors see the new server.
After 25 minutes, 69% of visitors see the new server.
After 30 minutes, 76% of visitors see the new server.
After 35 minutes, 80% of visitors see the new server.
After 45 minutes, 90% of visitors see the new server.
After 90 minutes, 95% of visitors see the new server.
After 18 hours, 99% of visitors see the new server.

With any number of servers at 90% uptime, this will
happen on average every ten days. With any number of
servers at 99% uptime, this will happen on average
every hundred days.

Needless to say, this works even better if you start
with high-uptime paid servers.

References:
http://blog.pyromod.com/2007/09/viab...-failover.html
http://www.webhostingtalk.com/showthread.php?t=524788
http://www.dnsmadeeasy.com/s0306/prod/dnsfosm.html
http://www.georedundanthosting.com/p...t-hosting.aspx
http://edgedirector.com/app/back.htm
http://www.webstrikesolutions.com/Page_ID.aspx?IDTXT=36

More References:
http://www.simplefailover.com/outbox/dns-caching.pdf
http://ausweb.scu.edu.au/aw03/papers...oth/paper.html
http://support.easydns.com/Failoverfaq.php
http://edgedirector.com/htm/note.htm


--
Guy Macon <http://www.GuyMacon.com/Guy Macon <http://www.GuyMacon.com/>
Guy Macon <http://www.GuyMacon.com/Guy Macon <http://www.GuyMacon.com/>
Guy Macon <http://www.GuyMacon.com/Guy Macon <http://www.GuyMacon.com/>
Guy Macon <http://www.GuyMacon.com/Guy Macon <http://www.GuyMacon.com/>

[viza]

On Sun, 26 Oct 2008 17:11:35 +0000, Guy Macon wrote:

How to increase web server uptime with DNS failover:

Two servers with 99% uptime will also be down at the same time one day
out of ten thousand. Three at 99% will be down at the same time one day
out of a million.

....as long as they run different operating systems and different servers
and host different websites, and are on different internets and are
independent in every other conceivable way.

[Guy Macon]

viza wrote:

>
Guy Macon <http://www.GuyMacon.com/wrote:

>How to increase web server uptime with DNS failover:

>Two servers with 99% uptime will also be down at the same time one day
out of ten thousand. Three at 99% will be down at the same time one day
out of a million.

...as long as they run different operating systems

Nope. They can both be running on linux. If every linux system
goes down at the same time, having one of your copies on, say,
solarus won't help you: the entire internet will be down.

>and different servers

I already specified that. Did you bother to read the post you are
responding to?

>and host different websites,

Nope. two servers in different places won't go down at the
same time just because the are serving the same HTML.

>and are on different internets

Now you are just being silly.

>and are independent in every other conceivable way.

That's not how the math behind redundancy works. the two
redundant systems have to not share a single point of
failure. Being "independent in every other conceivable
way." is not a requirement. My cellphone and my laser
printer both share the attribute of having copper
conductors and silicon microprocessors, (and thus are
not " independent in every conceivable way" but those
are not common points of failure. A copper wire may
fail, but all copper wires on earth are not going to
fail at the same time.

DNS failover can increase web server uptime. That is a
basig engineering calulation that has been tested many
times of many systems.

"Arguing with anonymous strangers on the Internet is a sucker's
game because they almost always turn out to be -- or to be
indistinguishable from -- self-righteous sixteen-year-olds
possessing infinite amounts of free time."
-Neil Stephenson, _Cryptonomicon_

--
Guy Macon
<http://www.GuyMacon.com/>

[Tim Greer]

viza wrote:

On Sun, 26 Oct 2008 17:11:35 +0000, Guy Macon wrote:

>How to increase web server uptime with DNS failover:

>Two servers with 99% uptime will also be down at the same time one
day
out of ten thousand. Three at 99% will be down at the same time one
day out of a million.

...as long as they run different operating systems and different
servers and host different websites, and are on different internets
and are independent in every other conceivable way.

The OS they run is irrelevant. Different systems on different networks
is the only thing that really plays a role, not the web site, type of
web site, platform/OS, or anything else. None of that would help
ensure better uptime, and only the network used between the two would
matter.

Ideally, you want to host DNS on different networks so if a DNS server
fails, the requestor's system or network doesn't see it as nonexistent
(just non responsive). There's a difference between non responsive,
down and non existent and those responses could be cached.

Beyond DNS, is actual DNS round-robin balancing, so if one system or
service goes down (not the DNS server/service) the other will be used
(one is down, too slow, overloaded, etc., the other one is used), but
you'd want to just do more than a round-robin solution for failover
services, if you have the option to.
--
Tim Greer, CEO/Founder/CTO, BurlyHost.com, Inc.
Shared Hosting, Reseller Hosting, Dedicated & Semi-Dedicated servers
and Custom Hosting. 24/7 support, 30 day guarantee, secure servers.
Industry's most experienced staff! -- Web Hosting With Muscle!

[Jerry Stuckle]

Guy Macon wrote:

viza wrote:

>Guy Macon <http://www.GuyMacon.com/wrote:

>>How to increase web server uptime with DNS failover:
Two servers with 99% uptime will also be down at the same time one day
out of ten thousand. Three at 99% will be down at the same time one day
out of a million.

...as long as they run different operating systems

Nope. They can both be running on linux. If every linux system
goes down at the same time, having one of your copies on, say,
solarus won't help you: the entire internet will be down.

>and different servers

I already specified that. Did you bother to read the post you are
responding to?

>and host different websites,

Nope. two servers in different places won't go down at the
same time just because the are serving the same HTML.

>and are on different internets

Now you are just being silly.

>and are independent in every other conceivable way.

That's not how the math behind redundancy works. the two
redundant systems have to not share a single point of
failure. Being "independent in every other conceivable
way." is not a requirement. My cellphone and my laser
printer both share the attribute of having copper
conductors and silicon microprocessors, (and thus are
not " independent in every conceivable way" but those
are not common points of failure. A copper wire may
fail, but all copper wires on earth are not going to
fail at the same time.

But he's right - both systems share the same internet. Completely
different than your cell phone and laser printer. This includes, among
other things, the DNS system, which isn't necessarily independent of
other parts of the internet.

They are a common point of failure.

DNS failover can increase web server uptime. That is a
basig engineering calulation that has been tested many
times of many systems.

"Arguing with anonymous strangers on the Internet is a sucker's
game because they almost always turn out to be -- or to be
indistinguishable from -- self-righteous sixteen-year-olds
possessing infinite amounts of free time."
-Neil Stephenson, _Cryptonomicon_

--
==================
Remove the "x" from my email address
Jerry Stuckle
JDS Computer Training Corp.
js*******@attglobal.net
==================

[viza]

Hi

On Mon, 27 Oct 2008 23:09:08 +0000, Guy Macon wrote:

viza wrote:

>>Guy Macon <http://www.GuyMacon.com/wrote:

>>How to increase web server uptime with DNS failover:

>>Two servers with 99% uptime will also be down at the same time one day
out of ten thousand. Three at 99% will be down at the same time one
day out of a million.

...as long as they run different operating systems

>>and different servers

>>and host different websites,

Nope. two servers in different places won't go down at the same time
just because the are serving the same HTML.

Just as an example: consider a site that goes down because of a
vulnerability in some bloated php cms. If your two machines are running
the same cgi scripts then they are more likely to both go down.

Redundancy does increase uptime, but to claim to be able to calculate it
quantitatively without in depth analysis of the likelihood and
independence of every possible source of downtime is like picking numbers
out of a hat. In case you don't have a degree in statistic and several
years worth of logs for a few thousand servers then you are just making
numbers up out of nothing.

Your advice is worthwhile and may even be helpful to people, but your
numbers are just a fantasy.

[Guy Macon]

viza wrote:

>
Guy Macon <http://www.GuyMacon.com/wrote:

>viza wrote:

>>>Guy Macon <http://www.GuyMacon.com/wrote:

How to increase web server uptime with DNS failover:

Two servers with 99% uptime will also be down at the same time one day
out of ten thousand. Three at 99% will be down at the same time one
day out of a million.

...as long as they run different operating systems

>>>and different servers

>>>and host different websites,

Nope. two servers in different places won't go down at the same time
just because the are serving the same HTML.

Just as an example: consider a site that goes down because of a
vulnerability in some bloated php cms. If your two machines are running
the same cgi scripts then they are more likely to both go down.

Redundancy does increase uptime, but to claim to be able to calculate it
quantitatively without in depth analysis of the likelihood and
independence of every possible source of downtime is like picking numbers
out of a hat. In case you don't have a degree in statistic and several
years worth of logs for a few thousand servers then you are just making
numbers up out of nothing.

Your advice is worthwhile and may even be helpful to people, but your
numbers are just a fantasy.

The title of this thread is "How to increase web server uptime with
DNS failover" and the text of the starting post compares multiple
servers with DNS failover with single servers. Your example of a
CGI script bringing down multiple servers would also bring down
a single server, and thus does not enter into a calculation for
increasing web server uptime with DNS failover. There are a great
number of things that are equally likely to bring down both the
single server and the multiple servers with DNS failover. The
sun going supernova, for example. If we were trying to predict
absolute reliability, we would have to account for all such
factors. If, as is the case here, we are calculating reliability
increase due to using multiple servers with DNS failover, all such
factors can be ignored.

--
Guy Macon
<http://www.GuyMacon.com/>

[viza]

Guy Macon wrote:

viza wrote:

>>Guy Macon wrote:

>>viza wrote:
Guy Macon wrote:

How to increase web server uptime with DNS failover:

Two servers with 99% uptime will also be down at the same time one
day out of ten thousand. Three at 99% will be down at the same time
one day out of a million.

...as long as they run different operating systems
and different servers
and host different websites,

Nope. two servers in different places won't go down at the same time
just because the are serving the same HTML.

Just as an example: consider a site that goes down because of a
vulnerability in some bloated php cms. If your two machines are running
the same cgi scripts then they are more likely to both go down.

Redundancy does increase uptime, but to claim to be able to calculate it
quantitatively without in depth analysis of the likelihood and
independence of every possible source of downtime is like picking
numbers out of a hat. In case you don't have a degree in statistic and
several years worth of logs for a few thousand servers then you are just
making numbers up out of nothing.

Your advice is worthwhile and may even be helpful to people, but your
numbers are just a fantasy.

Your example of a CGI script bringing down multiple servers would also
bring down a single server,and thus does not enter into a calculation
for increasing web server uptime with DNS failover.

If we were trying to predict absolute reliability, we would have to
account for all such factors. If, as is the case here, we are
calculating reliability increase due to using multiple servers with DNS
failover, all such factors can be ignored.

You are either contradicting yourself or demonstrating a basic ignorance
of statistics. Have you considered working for the government?

Your original statement was "Two servers with 99% uptime will also be
down at the same time one day out of ten thousand"

This assumes that all causes of downtime are completely independent,
which they are not.

You cannot just say that non-independent sources of downtime are
"irrelevant", because they are included in the numbers that you have
already given.

Suppose you had said "two servers which independently go down for six
hours due to entirely internal hardware failure on average once a year,
and also at some unspecified interval for other reasons" then you could
calculate the uptime improvement by using redundant servers, but you
would then still have to add back onto that the downtime caused by non-
independent sources, and this would mask the majority of the miracle
effect that you are touting.

To reiterate, redundancy does increase uptime but your numbers are pure
fantasy.

viza