Obtaining 100% Uptime

  • rDolay
  • Expert
  • Expert
  • rDolay
  • Posts: 541

Post 3+ Months Ago

Multiple DNS's wont able to give better uptime but probably make your uptime worse.
It is too long to tell it here but if your Server that hosts your primary DNS zone's DOWN then it does not matter that second server have the same DNS zones (teorically it must have the same accounts with the primary server too) because second DNS zones will not recieve the primary server DNS zones signals since its down ... Its meaningless - It also could not be done via Registerar Child Nameservers because web sites are not static but updated everytime and the updates that applied to the primary server wont be the same with the second-child server accounts data... At the end if you try to do this you will only crash.

There is only one option that applied to the real very important datacenters that they bind 2-6 servers via LAN and they made one of them primary and others are ns2-ns6... but any of the web hosting services could not serve it because of its incredible costs (dublicate data transfer - dublicate connection port - dublicate server costs... i claim it will cost 10-15 times more expensive then ordinary hosting).

100% uptime is nearly impossible since the every servers have to be rebooted once a 2-3 days plus every web servers auto up2date's by its-selfs once a 2-3days and those properties make the server down/unreachable at least 0.01%/Monthly.

P.S We are trying to 99.9% uptime and and we are successful to serve it.
  • Anonymous
  • Bot
  • No Avatar
  • Posts: ?
  • Loc: Ozzuland
  • Status: Online

Post 3+ Months Ago

  • Daemonguy
  • Moderator
  • Web Master
  • User avatar
  • Posts: 2700
  • Loc: Somewhere outside the box in Sarasota, FL.

Post 3+ Months Ago

dolay wrote:
Multiple DNS's wont able to give better uptime but probably make your uptime worse.
It is too long to tell it here but if your Server that hosts your primary DNS zone's DOWN then it does not matter that second server have the same DNS zones (teorically it must have the same accounts with the primary server too) because second DNS zones will not recieve the primary server DNS zones signals since its down ... Its meaningless - It also could not be done via Registerar Child Nameservers because web sites are not static but updated everytime and the updates that applied to the primary server wont be the same with the second-child server accounts data... At the end if you try to do this you will only crash.

There is only one option that applied to the real very important datacenters that they bind 2-6 servers via LAN and they made one of them primary and others are ns2-ns6... but any of the web hosting services could not serve it because of its incredible costs (dublicate data transfer - dublicate connection port - dublicate server costs... i claim it will cost 10-15 times more expensive then ordinary hosting).

100% uptime is nearly impossible since the every servers have to be rebooted once a 2-3 days plus every web servers auto up2date's by its-selfs once a 2-3days and those properties make the server down/unreachable at least 0.01%/Monthly.

P.S We are trying to 99.9% uptime and and we are successful to serve it.


Multiple nameservers make "uptime" worse? Again, I must not understand the context, since that is most assuredly not the case. NS's are generally set up with a weighted primary, secondary, tertiary, etc. If a primary is unavailable, the secondary is queried. these are generally slaves, to the master which handles zone transfers. However, data which pre-exists on the slaves will still respond as authoritative; you just are unable to process zt's. (Though, that may be solved with auto-promoting master/slaves).

Again, I fail to see the correlation to the web services. A DNS resolver will answer a name query with an IP, a VIP or round robin IP's. Whether that machine is a slave or a master is not relevant.

I am probably missing some important point though, so if I am speaking out of turn, please excuse me.

I know we have a combination of all of the above; IP's, VIP's and round robin.

Also, we have maintained 100% uptime for the last 62 months running. It's not impossible, it merely requires deep pockets. :) We guarantee 100% uptime to all of our customers, with NO maintenance windows.

Cheers.
  • rDolay
  • Expert
  • Expert
  • rDolay
  • Posts: 541

Post 3+ Months Ago

Daemonguy wrote:
dolay wrote:
Multiple DNS's wont able to give better uptime but probably make your uptime worse.
It is too long to tell it here but if your Server that hosts your primary DNS zone's DOWN then it does not matter that second server have the same DNS zones (teorically it must have the same accounts with the primary server too) because second DNS zones will not recieve the primary server DNS zones signals since its down ... Its meaningless - It also could not be done via Registerar Child Nameservers because web sites are not static but updated everytime and the updates that applied to the primary server wont be the same with the second-child server accounts data... At the end if you try to do this you will only crash.

There is only one option that applied to the real very important datacenters that they bind 2-6 servers via LAN and they made one of them primary and others are ns2-ns6... but any of the web hosting services could not serve it because of its incredible costs (dublicate data transfer - dublicate connection port - dublicate server costs... i claim it will cost 10-15 times more expensive then ordinary hosting).

100% uptime is nearly impossible since the every servers have to be rebooted once a 2-3 days plus every web servers auto up2date's by its-selfs once a 2-3days and those properties make the server down/unreachable at least 0.01%/Monthly.

P.S We are trying to 99.9% uptime and and we are successful to serve it.


Multiple nameservers make "uptime" worse? Again, I must not understand the context, since that is most assuredly not the case. NS's are generally set up with a weighted primary, secondary, tertiary, etc. If a primary is unavailable, the secondary is queried. these are generally slaves, to the master which handles zone transfers. However, data which pre-exists on the slaves will still respond as authoritative; you just are unable to process zt's. (Though, that may be solved with auto-promoting master/slaves).

Again, I fail to see the correlation to the web services. A DNS resolver will answer a name query with an IP, a VIP or round robin IP's. Whether that machine is a slave or a master is not relevant.

I am probably missing some important point though, so if I am speaking out of turn, please excuse me.

I know we have a combination of all of the above; IP's, VIP's and round robin.

Also, we have maintained 100% uptime for the last 62 months running. It's not impossible, it merely requires deep pockets. :) We guarantee 100% uptime to all of our customers, with NO maintenance windows.

Cheers.

Yes you have missed a point at the below on my first post.
I tought that if the secondary nameserver ip is belong to the primary ip of the server then its meaningless to how many additional name servers(it is not nameservers then but it name is virtual nameservers ) you have. It only will help to regard the busy on the main ip but it wont help to stand your server if the primary ip fail , there is an important point here : How the primary ip will fail?
1- Becouse of an outgate
2- Server Crash
So how can you think that the secondary or so far nameservers ip that belongs to primary ip(virtual nameservers) will wakeup your server when it fails ?

As i tought at the above the only way is to assign additional servers to the secondary and so far namservers ip's and made them work coolarated in realtime with eachother (when a server has its site updated then secondary server have to apply those changes to its own built) and this means hell of cost - a deep pocket wont enough to serve this kind service but you will need a deep sack :D
  • Daemonguy
  • Moderator
  • Web Master
  • User avatar
  • Posts: 2700
  • Loc: Somewhere outside the box in Sarasota, FL.

Post 3+ Months Ago

OK, it seems that we're back on the application server -- in this case the web server. So let's try that one.

Dolay wrote:

Yes you have missed a point at the below on my first post.
I tought that if the secondary nameserver ip is belong to the primary ip of the server then its meaningless to how many additional name servers(it is not nameservers then but it name is virtual nameservers ) you have. It only will help to regard the busy on the main ip but it wont help to stand your server if the primary ip fail , there is an important point here


Huh? A secondary NS has, of course, it's own IP's. All of which are registered as authoritative for the domains in processes resolution for. All of your nameservers contain the same information regarding the domain.
The nameservers will round robin for each subsequent request. Using VIP's and load balancers for DNS, works the same as it does for the application serving. A request goes to a VIP, or a round-robin chain of VIPs, the LB sources the request to one of several NS's. If one fails, the LB marks it down, and does not forward requests to it, instead making use of the remaining NS's.

Dolay wrote:

: How the primary ip will fail?
1- Becouse of an outgate
2- Server Crash
So how can you think that the secondary or so far nameservers ip that belongs to primary ip(virtual nameservers) will wakeup your server when it fails ?


So, what we were talking about was DNS failure? I thought the original question was application server failure; which in this case was a web server, or more to the point, multiple web servers. Now I am very confused.
In any case, the same technology (whether it be software or hardware based, expensive or free) may be used to load balance (LB) the backend. With the caveat that the LB'er must be aware and accomplish health checks. (As I mentioned previously).

Dolay wrote:

As i tought at the above the only way is to assign additional servers to the secondary and so far namservers ip's and made them work coolarated in realtime with eachother (when a server has its site updated then secondary server have to apply those changes to its own built) and this means hell of cost - a deep pocket wont enough to serve this kind service but you will need a deep sack :D


Actually, running intelligent LB'ers is not that expensive. The Foundry Irons we have were some of the least expensive equipment we purchased -- comparatively speaking of course. Also, there are software packages which will accomplish this, on a much smaller scale.
As for DNS updates, it's trivial to have slaves accept and update zone files, though obviously one would not need to do this regularly... which is why we have tools that come with bind such as rndc. So I am again, unsure as to what you are alluding to.

Granted, most people do not go to this degree of redundancy. To make an entire architecture truly redundant *is* a costly venture, however, making a process more robust, especially web services, is again, trivial.


Cheers.
  • phpservers
  • Banned
  • Banned
  • phpservers
  • Posts: 114

Post 3+ Months Ago

100% uptime is definetely not impossible - out of the years that i have seen major websites, I have never soon google, msn, yahoo, etc. suffer downtime. then again they have budgets 500,000 times the majority of us.

imo the key to better uptime is high specification servers - lots of RAM, speed, and a good OS (like freeBSD)... the chances of downtime are reduced considerably because the server is capable of handling the most dubious of tasks.

bottom line - you get what you pay for with the majority of web hosts.

most data centers mention in their disclaimers that they arent liable for downtime due to major disasters that they cant prevent - e.g. hurricanes, floods etc. which have become a lot more common just in the past year.

You only need 100% uptime if you have a site that gets atleast 5000 uniques per day, otherwise it is pointless because 1% downtime will be ignored or not even noticed, at this point you would want collocation servers which will provide a lot more speed and reliability towards your target audience (world).

if you want 100% uptime you best buy a few servers from one datacenter; others off another; and more off another because the chances of them all being down are more like 1000000:1.

if you dont want to get dedicated servers; you have a much smaller chance because you will want/need to monitor uptime yourself.

i am just trying to conclude this discussion. the difference between 99% uptime and 100% uptime is a lot of money.

Post Information

  • Total Posts in this topic: 5 posts
  • Users browsing this forum: No registered users and 8 guests
  • You cannot post new topics in this forum
  • You cannot reply to topics in this forum
  • You cannot edit your posts in this forum
  • You cannot delete your posts in this forum
  • You cannot post attachments in this forum
 
cron
 

© 1998-2014. Ozzu® is a registered trademark of Unmelted, LLC.