It's not even a network problem, it's an application design problem. Even if all you wanted to do was load balance a user session (user ID & some other text-based info) across data centers, it's not that trivial, particularly if you didn't plan for that up front. At the bare minimum all you need is a session ID, but the easiest way to do that is with a session server specifically for this purpose. Even with 1 LB, 2 Web Heads and 1 DB server in the same subnet in the cloud, it's the same problem - user uploads something to X but their next request is served by Y.
The content belongs on a CDN. It sounds to me like you are describing one of the most common use cases for one. You don't need to know the user went to server X or Y since their session is on server S and the content is on CDN and S & CDN are both accessible to X & Y. This architecture is commonplace (now) but I wouldn't say it's `drop dead easy`. Just my $0.02 - On Mon, Apr 4, 2011 at 8:41 PM, Brian Moon <br...@moonspot.net> wrote: > That is not a bad design. That is drop dead easy. > > You are asking to this list and memcached to magically solve a problem that > is not realistically solvable with the current architecture of the Internet > at a scale you are likely to be running on. > > Now, if you would like to invest in private OC3's that run from data center > to data center to ensure sub millisecond latency from one to the other, be > my guest that would solve your problem. But, this is not a memcached > problem. It is a network problem. > > Brian. > http://brian.moonspot.net > > On 4/4/11 7:13 PM, Mohit Anchlia wrote: >> >> Bad design. Besides not that easy :) If it was I wouldn't have posted >> here. >> >> On Mon, Apr 4, 2011 at 5:04 PM, Brian Moon<br...@moonspot.net> wrote: >>> >>> You have full control over what resources your internal servers use. Just >>> assign them a datacenter and go. >>> >>> Brian. >>> http://brian.moonspot.net >>> >>> On 4/4/11 6:59 PM, Mohit Anchlia wrote: >>>> >>>> Problem here is that lot of traffic is generated internally by server >>>> hosted by other projects within same co. now this need to be load >>>> balanced. If we used geo then 70% of our traffic will be stuck on one >>>> site. If all our clients were browser based then it would have been >>>> easier. >>>>> >>>>> 2011/4/4 Brian Moon<br...@moonspot.net>: >>>>>> >>>>>> We are active/active as well. But, we use geo dns so that people only >>>>>> get >>>>>> DNS for one data center. Having someone be able to hit any datacenter >>>>>> in >>>>>> the >>>>>> world at any time without any temporary loss of service is not >>>>>> reasonable. I >>>>>> don't care who you are. Even Google sticks you to a geo-regional based >>>>>> datacenter. >>>>>> >>>>>> Brian. >>>>>> http://brian.moonspot.net >>>>>> >>>>>> On 4/4/11 5:39 PM, Mohit Anchlia wrote: >>>>>>> >>>>>>> That is already in place but business requirement is to do >>>>>>> active/active hence need for more complicated solution. >>>>>>> >>>>>>> >>>>>>> >>>>>>> On Mon, Apr 4, 2011 at 3:32 PM, Brian Moon<br...@moonspot.net> >>>>>>> wrote: >>>>>>>> >>>>>>>> Use geo dns instead to stick users to a single datacenter and only >>>>>>>> fail >>>>>>>> over >>>>>>>> to the other data center when there is an issue. This will be much >>>>>>>> less >>>>>>>> of a >>>>>>>> headache than trying to move cache data back and forth over the net. >>>>>>>> >>>>>>>> Brian. >>>>>>>> http://brian.moonspot.net >>>>>>>> >>>>>>>> On 4/4/11 3:03 PM, Mo wrote: >>>>>>>>> >>>>>>>>> We have multiple data centers and are now planning to make this >>>>>>>>> application active/active. Which means user can be load balanced. >>>>>>>>> User >>>>>>>>> generally uploads a file and it should be accessible on both sites. >>>>>>>>> >>>>>>>>> We expect it will take upto 1 hr to replicate files in worst case >>>>>>>>> scenario and we are not able to come up with good solution since >>>>>>>>> cookies wouldn't work for us. >>>>>>>>> >>>>>>>>> What we really need is someway of storing User and Site eg: User A >>>>>>>>> visited site X. Based on that information we can then redirect user >>>>>>>>> to >>>>>>>>> correct site. After one hour this info will expire and generate new >>>>>>>>> info. >>>>>>>>> >>>>>>>>> I am planning to use memcached on httpd apache server accross 2 >>>>>>>>> data >>>>>>>>> centers to keep cache in sync. >>>>>>>>> >>>>>>>>> I understand latency will be a factor but I am assuming we can also >>>>>>>>> do >>>>>>>>> async and it shouldn't be that slow since we are only talking about >>>>>>>>> small set of data. >>>>>>>>> >>>>>>>>> Need help from experienced users if they have any good suggestions >>>>>>>>> on >>>>>>>>> how to do this. >>>>>>>> >>>>>> >>>>> >>>>> >>>>> >>>>> -- >>>>> Roberto Spadim >>>>> Spadim Technology / SPAEmpresarial >>>>> >>> >
