Page 2 of 2 FirstFirst 12
Results 26 to 42 of 42
  1. #26
    Join Date
    Feb 2008
    Location
    Houston, Texas, USA
    Posts
    3,262
    Hi,

    The way we do failover, and during dry runs, most customers have reported that they were able to access the new website, email, control panel, etc within 2 minutes of main node failure. Despite all the 3rdparty DNS and application caching.

    That's
    99.995% uptime SLA with TWO downtimes a month
    99.990% uptime SLA with FOUR downtimes a month
    99.950% uptime SLA with TEN downtimes a month
    That's still 99.9% even at the worst facility there is.

    Regards
    UNIXy - Fully Managed Servers and Clusters - Established in 2006
    Server Management - Unlimited Servers. Unlimited Requests. One Plan!
    cPanel Varnish Plugin -- Seamless SSL Caching (Let's Encrypt, AutoSSL, etc)
    Slow Site or Server? Unable to handle traffic? Same day performance fix: joe@unixy
      0 Not allowed!

  2. #27
    Join Date
    Oct 2007
    Posts
    455
    Quote Originally Posted by nerdie View Post
    DNS failover doesn't work as well as one might think, even if the TTL is 0 major ISPs will cache it so a lot of people won't be able to view your site when it fails over.
    Everytime this question comes up someone comes up with this response. So please, give me the ISPs that will ignore TTLs that are 180 seconds or above. We sell this solution to many of our clients and we have never had a problem.
      0 Not allowed!

  3. #28
    Join Date
    Oct 2007
    Posts
    455
    Quote Originally Posted by nerdie View Post
    We could go back and forth on this topic for days. I'm just explaining my experience with it, had an issue just last weekend where an AOL DNS server cached my entry for 3 days, even though the TTL was 3600 (1 day)
    The first problem you are having is that you have to know your TTLs and what to expect. 3600 is one hour, not one day.

    But I just tested your AOL issue and that is not true at all. Not even close. I set a TTL to 15 minutes (900 seconds). I did a ping on a AOL dial-up account, changed the IP, waited 15 minutes, did a ping again, and it change to the new IP the next time I pinged.

    I may not understand what you mean by your example, but please provide details on how you came up with these results.
      0 Not allowed!

  4. #29
    Join Date
    Oct 2007
    Posts
    239
    Quote Originally Posted by BuffaloBill View Post
    Everytime this question comes up someone comes up with this response. So please, give me the ISPs that will ignore TTLs that are 180 seconds or above. We sell this solution to many of our clients and we have never had a problem.
    Was that supposed to be 1800 seconds? 180 sounds a little low. Most will respect 600 seconds and upwards, though. Including AOL.
      0 Not allowed!

  5. #30
    Join Date
    Oct 2007
    Posts
    239
    Quote Originally Posted by BuffaloBill View Post
    The first problem you are having is that you have to know your TTLs and what to expect. 3600 is one hour, not one day. .
    He is probably just tired from "using DNS all day long".
      0 Not allowed!

  6. #31
    Join Date
    Jul 2005
    Posts
    3,784
    Quote Originally Posted by Shoel View Post
    He is probably just tired from "using DNS all day long".
    It seems this thread has gotten a bit childish. The OP asked about DNS failover, I gave him my experiences with it from our DNS hosting company and why we don't offer it.

    Hopefully the OP has better luck then we did. If he can let us know how it works by coming back to this thread once it's setup, it would help us re-evaluate it and perhaps start offering it again on our own service.
      0 Not allowed!

  7. #32
    Thanks everyone for the replies!! I will discuss with our team. Once we have a solution I will post it here to share with you all.
      0 Not allowed!

  8. #33
    Join Date
    Oct 2007
    Posts
    239
    Quote Originally Posted by nerdie View Post
    It seems this thread has gotten a bit childish.
    Oh come on, lighten up. You should have some self-irony for making such a statement in the first place.

    And with all due respect, your advice was formed as a factual statement based on incorrect information - so it is only fair that others correct you. You did not say "our experience with DNS was..." - you said "DNS failover doesn't work (...) a lot of people won't be able to view your site when it fails over." - which is incorrect providing it is done properly.
      0 Not allowed!

  9. #34
    Join Date
    Apr 2004
    Location
    Pacific Palisades, CA
    Posts
    3,641
    Quote Originally Posted by Shoel View Post
    That assumes that the user capable of doing such a change is available at that time and able to act instantly.
    That's essentially what I said. You are only saving an amount of time equal to the time doing it manually and the time it takes for a script to do it automatically.


    The majority will be swapped over reasonably fast, the vast majority within a couple of hours (97-98%) using a TTL of 30 minutes -
    and for many types of websites that is completely acceptable
    and

    I think you are being a little narrow minded. not everyone have business/enterprise demands. Please try to look beyond your own specific use case.
    Again, browsers don't use the RR TTL. Under the best of conditions, and absent any user action, IE users will experience 32 minutes of downtime (30 minutes default cache + 2 minutes recovery time) and FireFox users will experience 4 minutes of downtime ( 1 minute default cache + 3 minutes recovery time)

    This is unacceptable if you are concerned about current visitors. If its future visitors that is your major concern then TTL fiddling is probably ok. But it is you who are narrow minded for considering only the future visitors.

    Don't know whay you are so adamant about not presenting the full case to OP. You don't know his exact situation either
    Collabora Hosting - Unlimited Windows and Linux Hosting
    Web Security - VPS - Dedicated Servers
    Cloud and Managed WordPress Hosting
    Read how we do Unlimited Hosting at the Unlimited FAQ
      0 Not allowed!

  10. #35
    Join Date
    Oct 2007
    Posts
    239
    Quote Originally Posted by Collabora View Post
    That's essentially what I said. You are only saving an amount of time equal to the time doing it manually and the time it takes for a script to do it automatically.
    Which can be an entire day or even an entire weekend for many communities and non-commercial websites out there as an example. Thus an automatic solution that does the job for them within minutes can save a lot of downtime for the majority of their users.

    Quote Originally Posted by Collabora View Post
    Again, browsers don't use the RR TTL. Under the best of conditions, and absent any user action, IE users will experience 32 minutes of downtime (30 minutes default cache + 2 minutes recovery time) and FireFox users will experience 4 minutes of downtime ( 1 minute default cache + 3 minutes recovery time)
    From experience the average is much lower than 30+ minutes with a lower TTL, and even that would be an acceptable compromise between longer downtime or a more costly solution for a lot of people out there. Those requiring absolute zero downtime and having the budget for it is a relatively small group among all the websites on the internet.

    Quote Originally Posted by Collabora View Post
    This is unacceptable if you are concerned about current visitors. If its future visitors that is your major concern then TTL fiddling is probably ok. But it is you who are narrow minded for considering only the future visitors.

    Don't know whay you are so adamant about not presenting the full case to OP. You don't know his exact situation either
    Are you being serious? Have you even read my posts? Again, and again I have *clearly* underlined that those who are truly concerned with having an instant failover for all users should obviously not go for a DNS based solution. Let me quote for you:

    Quote Originally Posted by Shoel View Post
    DNS failover works perfectly well as long as it does not have to be absolutely instant
    Quote Originally Posted by Shoel View Post
    If you need a fully reliable and instant business/enterprise failover solution with a 100% instant swap for all users, then it should be quite obvious that a DNS based system will not do the job.
    Quote Originally Posted by Shoel View Post
    If you are going to sell people plans for high availability hosting, you would obviously not choose a DNS based failover solution.
    Shall I say it a 4th time for good measure, perhaps?

    If anything, *I* have in fact presented both sides of the stoy - while nerdie has completely dismissed the DNS based solution as a usable one based on his own narrow experience with the special case of providing it as a commercial service, and applied that experience as fact for all use cases.

    I do indeed not know the OP's exact situation, and I have never pretended to do so either. I have merely argued against some people's mindset in that a DNS failover is not a solution that can work for many. What you guys are doing is assuming that the OP needs a failover for a business with HA requirements, or for selling such features to others. What I am saying over and over again is that far from everyone has such a requirement, and can therefore be perfectly happy with a DNS based solution. In other words, you are clearly the ones being narrow minded regarding the areas of use for a failover. Trying to point that finger at me is quite ridiculous.
      0 Not allowed!

  11. #36
    Join Date
    Apr 2004
    Location
    Pacific Palisades, CA
    Posts
    3,641
    Quote Originally Posted by Collabora View Post

    Again, browsers don't use the RR TTL. Under the best of conditions, and absent any user action, IE users will experience 32 minutes of downtime (30 minutes default cache + 2 minutes recovery time) and FireFox users will experience 4 minutes of downtime ( 1 minute default cache + 3 minutes recovery time)
    and your response is...

    Quote Originally Posted by Shoel View Post

    From experience the average is much lower than 30+ minutes with a lower TTL, and even that would be an acceptable compromise between longer downtime or a more costly solution for a lot of people out there. Those requiring absolute zero downtime and having the budget for it is a relatively small group among all the websites on the internet.
    You are still not getting it. The TTL is not passed to browsers. They maintain a cache of dns results according to the built-in browser programming code -- has nothing to do with RR TTLs. Thus lowering TTL will have no affect on IE or Firefox.
    Collabora Hosting - Unlimited Windows and Linux Hosting
    Web Security - VPS - Dedicated Servers
    Cloud and Managed WordPress Hosting
    Read how we do Unlimited Hosting at the Unlimited FAQ
      0 Not allowed!

  12. #37
    Join Date
    Oct 2007
    Posts
    239
    Quote Originally Posted by Collabora View Post
    You are still not getting it. The TTL is not passed to browsers. They maintain a cache of dns results according to the built-in browser programming code -- has nothing to do with RR TTLs. Thus lowering TTL will have no affect on IE or Firefox.
    No, you are interpeting what I said incorrectly. I never said DNS TTL is passed down to the browser, that is a subject you have managed to come up with on your own.

    I quoted your statement regarding downtime and answered that - so the natural assumption would be that my "average" was refering to downtime / failover time for the end user. I said that from experience, the average [typical timeframe] a user experiences from the failure of the primary site to the failover site being available is lower than 30+ minutes using a TTL which is below that - e.g. 600-900.

    Furthermore I pointed out that even with your example of 32 minutes downtime, this would be an acceptable compromise between longer downtime or a more costly solution for a lot of people out there.

    Those requiring absolute zero downtime and having the budget for it is a relatively small group among all the websites on the internet. Those users, along with people interested in providing commerical HA services, should of course - for the 4th (or is it the 5th?) time - use a non-DNS based failover. That does however not mean that a DNS-failover will not work well for those with more down to earth requirements.
      0 Not allowed!

  13. #38
    Join Date
    Apr 2004
    Location
    Pacific Palisades, CA
    Posts
    3,641
    Quote Originally Posted by Shoel View Post
    Furthermore I pointed out that even with your example of 32 minutes downtime, this would be an acceptable compromise between longer downtime or a more costly solution for a lot of people out there.
    Both solutions require two servers. Its not just a question of cost, its also a question of value. One can also ask, why purchase a second server in order to implement a flawed failover strategy? My method, i.e., the proper method, requires an additional device, namely a load balancer. But this additional cost is balanced by the added value it gives to the 2nd server.

    If one's situation justifies the purchase of a second server, its not much of a leap to add aload balancer and implementing best practices. If the load balancer is not justified then its likely that the 2nd server isn't either
    Collabora Hosting - Unlimited Windows and Linux Hosting
    Web Security - VPS - Dedicated Servers
    Cloud and Managed WordPress Hosting
    Read how we do Unlimited Hosting at the Unlimited FAQ
      0 Not allowed!

  14. #39
    Join Date
    Oct 2007
    Posts
    239
    Quote Originally Posted by Collabora View Post
    Both solutions require two servers. Its not just a question of cost, its also a question of value. One can also ask, why purchase a second server in order to implement a flawed failover strategy? My method, i.e., the proper method, requires an additional device, namely a load balancer. But this additional cost is balanced by the added value it gives to the 2nd server.

    If one's situation justifies the purchase of a second server, its not much of a leap to add aload balancer and implementing best practices. If the load balancer is not justified then its likely that the 2nd server isn't either
    No, we were discussing 3rd party hosted DNS failover solutions - such as DNS Made Easy, which is a relatively cheap solution (around $20-25/site per year with fully hosted/managed DNS) not requiring any additional hardware. This can be used by everyone from John Doe having two cheap shared hosting accounts (one for his main community site, another with a different host for a failover) to someone having for an example a dedicated server for their main site and a 3rd party shared hosting account for a failover backup (as well as a number of other use cases).

    A local hardware failover would require at a minimum one appliance/server to act as a LB/FO device and two end targets for a primary and secondary service. If you want such a failover solution to be fully location/network independant (if you lose network and/or power at the DC for an example), it quickly gets more complicated and costly. While this would be the way to go for providing a commercial service or sites that are highly dependant on 100% availability - setting this up yourself for sites such as a smaller non-commerial community or similar would be rather unrealistic. A solution provided by a 3rd party will typically cost a whole lot more than a hosted DNS-failover service - as you'd actually have to pass all your traffic through it, and the infrastructure cost for them would me much greater.

    Yet again you seem to only think of it from a large scale perspective of a host wanting to offer a service, or for an example a major online store being dependant on not missing a single minute of uptime. There's a ton of websites on the internet, and believe me.. the majority of these have much more moderate uptime demands when it comes down to a cost/benefit ratio - and can accept 30 minutes downtime once in a blue moon over the cost of a better failover solution any day. Just a single dedicated webserver is far fetched for many of these, let alone two + a load balancer.
      0 Not allowed!

  15. #40
    Join Date
    Apr 2004
    Location
    Pacific Palisades, CA
    Posts
    3,641
    Having no failover system is better than paying for a flawed one, no matter how big or small you are. If going from 99.9% uptime to 99.999% uptime has no value for your org then why bother with the el-cheapo flawed solution either?
    Collabora Hosting - Unlimited Windows and Linux Hosting
    Web Security - VPS - Dedicated Servers
    Cloud and Managed WordPress Hosting
    Read how we do Unlimited Hosting at the Unlimited FAQ
      0 Not allowed!

  16. #41
    Join Date
    Oct 2007
    Posts
    239
    Quote Originally Posted by Collabora View Post
    Having no failover system is better than paying for a flawed one, no matter how big or small you are. If going from 99.9% uptime to 99.999% uptime has no value for your org then why bother with the el-cheapo flawed solution either?
    That is not true. A lot of these sites will have a single person responsible for the site administration, or a limited few. They may not be commercial, far from every site out there is backed by any form of serious organization, and as such do not have any form of shift rotation to make sure someone's available at all times. While they do not necessarily have a 99.999% uptime requirement, they will of course want their site to have reasonable uptime or at least provide some form of static information in case of trouble. A DNS failover, which is not a flawed solution - just a slower one, will provide this at an affordable pricetag. This means that even if the sysadmin is without internet access for the weekend, away for the day, or sound asleep in his bed - the site will switch to the backup in reasonable time (from their perspective) and thus let the majority of their users continue to use the site - or be presented with basic/essential information. This comes down to convenience more than any ecconomically driven need for many of them. 30 minutes or less downtime for the majority of users, or perhaps one or more days for everyone.. the choice will be simple.
      0 Not allowed!

  17. #42
    Join Date
    Jul 2002
    Location
    Victoria, Australia
    Posts
    36,941
    Quote Originally Posted by vecwebdev View Post
    Thanks everyone for the replies!! I will discuss with our team. Once we have a solution I will post it here to share with you all.
    Let us know if you need the thread reopened
      0 Not allowed!

Page 2 of 2 FirstFirst 12

Similar Threads

  1. Replies: 0
    Last Post: 05-27-2008, 10:41 PM
  2. Replies: 0
    Last Post: 03-07-2008, 10:37 PM
  3. Replies: 0
    Last Post: 02-28-2008, 09:44 AM
  4. Replies: 0
    Last Post: 02-20-2008, 12:25 AM
  5. Replies: 0
    Last Post: 02-09-2008, 05:24 PM

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •