I feel like the actual interesting bit about anycast was 2 minutes long near the end and it didn't really explain the implementation details very well. The primer at the start is useful for setting up context but another 5 minutes on anycast implementation would've been more useful.
Anycast server IP and server IP will be different but all the dns servers will sit in same DNS anycast pool. Anycast IP is used by all the users instead of user configure dns servers located locally or at remote location. May be you could cover ospf also along with BGP in the session. Great session . Congratulations
Static contents which have to be present on this can have a push model from a central DC, or you have a P2P based approach where the PoPs sync with each other the latest version which they have on them. The centralised approach will be preferable in cases where you want to ramp a feature only to a specific set of user groups based on geographies(canary deployments).
HI Aditya ! Thank you so much for the video .. it's crystal clear. I have one doubt say if one of the POP is down. How does it manages to reroute the traffic the active one
Whenever a PoP goes down, it stops advertising in the BGP peering process and hence its address gets removed from the routes. This ensures that the traffic stops going to that PoP. This ofcourse can lead to some traffic going to that PoP during the new routes are broadcasted. This is avoided to a large extent by using proactive health checks.
Thanks, Madhur! Yes, I should have quickly differentiated between PoP and DC, the only problem is that there is a very thin line between them. You can have a PoP consist of just few routers and a couple of servers to hundreds of servers. The only abstract differentiation would be that a DC is usually much larger than a PoP, and a PoP is mostly used for serving cached content and then it is able to route the request to the DC if it does not have what is being requested for.
@@AdityaKamat That description definitely helps! Not sure if TH-cam comments are the right way to "discuss" but a follow up question - Is there any dedicated connection between a PoP and the DC? Or the traffic has to flow over the general Internet?
In most of the cases that I have come across, I have seen a dedicated connection from PoP to DC. This helps reduce latency in fetches for requests from servers in DC, plus the sync of static content between DC and PoP is also quicker.
How to prevent BGP hijacking? I can advertise like I own particular IP address of a well known service. And redirect users to my server. Like DNS poisoning. How to prevent this or what are the safety mechanism present to avoid this situation when we use anycast.
Quite an informative video. There were a few points I could not really get though more related to POP on what it is and why do we need it? Is it an instance of the base website at each location? If yes, then do we need anycast IP to access the POP as we already have the instance. I know it is a lot of questions, but any info would be great. Thanks!
Just having an instance serving the website is not enough. You also need an IP address so your requests can reach the server. Hope it is clear so far. Now if each PoP have their individual IP address then there needs to be some logic at your end to pick the correct instance and send your request there, which is a huge task for the client. This is where an anycast IP helps and the internet helps you make that decision about which instance to send your request to.
Applications usually are hosted on one or set of servers for providing redundancy and anycast helps in choosing the shortest path and route the traffic to closest authoritative dns server to answer the query. You need to remember only anycast IP and rest anycast will do it for you by routing the query to Closest DNS server. Routing on layer 3 switches or router will decide what is the best route
Thanks for the video..it's really very interesting.. I have one query.. In aneycast, end users will connect to pop and pop will route the traffic to the dc where the web server is hosted..in case of DDOS attack all traffic will reach to the web server through different POPs, and web server will be down.. Does a single web server can be protected by anycast ip or we need to place multiple web server in multiple regions?
Thanks for watching the Video and glad you liked it. Anycast IP is not for DDOS protection, you would need IP pattern filtering mechanisms and firewalls in place for those. It would also make sense to distribute your web servers in multiple regions if you can afford the cost of that.
It depends on what parameters are prioritized in the routing logic on these POPs. If the application is time critical, parameters based on latency will be compared between the pops. If its more like a video streaming/heavy data transfer service they would look for bandwidth related parameters between the two pop's. If you take a zero probability case where all parameters are the same between both the POPs, it would then choose randomly.
th-cam.com/video/Q_YpTh0Qq_k/w-d-xo.html I talk about BGP peering at 10:50, sorry I'm on mobile and not able to share the video link with time in it. And thanks for watching :)
So basically for 13 root servers, you are saying that we have other redundant servers over the globe and they have the same IPs as of root servers. is the ip just same or its a complete replica of the root servers.
Ip address of the anycast server will be different and servers IP will be different. Anycast will have a route to each of the server and have to put dns servers which are authoritative to answer the query in same anycast pool
Great video, this is exactly what I was looking for, don't stop here, keep posting.
Thanks, this helped me understand aws global accelerator
Very clearly described. It's a good way to take use case and explain the concept.
Great step by step approach of tackling the Do's and Doesn't s. . Thanks
Glad you liked it :)
I feel like the actual interesting bit about anycast was 2 minutes long near the end and it didn't really explain the implementation details very well. The primer at the start is useful for setting up context but another 5 minutes on anycast implementation would've been more useful.
I quite agree it was a bit of rush without much saying about how ipv6 - Anycast works
Thank you aditya.. Appreciate ur efforts for simplifying the concept.. I was searching from long time..
Thanks for watching Amit! Glad you liked it :)
very informative, now i get the big picture about anycast ip
Thanks Nelson! Glad you found it informative.
Anycast server IP and server IP will be different but all the dns servers will sit in same DNS anycast pool.
Anycast IP is used by all the users instead of user configure dns servers located locally or at remote location.
May be you could cover ospf also along with BGP in the session.
Great session . Congratulations
Very informative, end to end issues and solution discussed. Thanks for sharing.
Thanks Seshagiri :)
superb!!!!!
subscribed
Man, you have a good talent of making people understand a concept. Why are you running a scam like Heycoach these days?
Good one....now I understood the concept of Anycast IP in GCP load balancing...Thank you Aditya for this information..!! :)
That's good to hear. Thanks for watching :)
Thank you. Very informative.
Emmm
12:17
How does the POP(point of present)s sync with each other?
Static contents which have to be present on this can have a push model from a central DC, or you have a P2P based approach where the PoPs sync with each other the latest version which they have on them. The centralised approach will be preferable in cases where you want to ramp a feature only to a specific set of user groups based on geographies(canary deployments).
@@AdityaKamat Thanks for the indepth answer.
HI Aditya ! Thank you so much for the video .. it's crystal clear.
I have one doubt say if one of the POP is down. How does it manages to reroute the traffic the active one
Whenever a PoP goes down, it stops advertising in the BGP peering process and hence its address gets removed from the routes. This ensures that the traffic stops going to that PoP.
This ofcourse can lead to some traffic going to that PoP during the new routes are broadcasted. This is avoided to a large extent by using proactive health checks.
Very informative... keep it up the good work 👍
Thanks husain!
Nice explanation Aditya.
Thanks Rohit :)
Great video! Keep creating more. Would've been helpful if you could have quickly explained the difference between a PoP v/s a DC.
Thanks, Madhur!
Yes, I should have quickly differentiated between PoP and DC, the only problem is that there is a very thin line between them. You can have a PoP consist of just few routers and a couple of servers to hundreds of servers. The only abstract differentiation would be that a DC is usually much larger than a PoP, and a PoP is mostly used for serving cached content and then it is able to route the request to the DC if it does not have what is being requested for.
@@AdityaKamat That description definitely helps! Not sure if TH-cam comments are the right way to "discuss" but a follow up question - Is there any dedicated connection between a PoP and the DC? Or the traffic has to flow over the general Internet?
In most of the cases that I have come across, I have seen a dedicated connection from PoP to DC. This helps reduce latency in fetches for requests from servers in DC, plus the sync of static content between DC and PoP is also quicker.
How to prevent BGP hijacking?
I can advertise like I own particular IP address of a well known service. And redirect users to my server. Like DNS poisoning. How to prevent this or what are the safety mechanism present to avoid this situation when we use anycast.
BGP hijacking is much easier said than done. I've personally never seen ways to prevent it as yet.
11:10 Let's look at how anycast works
18:52 Let's consider a DDoS attack
Quite an informative video. There were a few points I could not really get though more related to POP on what it is and why do we need it? Is it an instance of the base website at each location? If yes, then do we need anycast IP to access the POP as we already have the instance. I know it is a lot of questions, but any info would be great. Thanks!
Just having an instance serving the website is not enough. You also need an IP address so your requests can reach the server. Hope it is clear so far.
Now if each PoP have their individual IP address then there needs to be some logic at your end to pick the correct instance and send your request there, which is a huge task for the client. This is where an anycast IP helps and the internet helps you make that decision about which instance to send your request to.
Applications usually are hosted on one or set of servers for providing redundancy and anycast helps in choosing the shortest path and route the traffic to closest authoritative dns server to answer the query.
You need to remember only anycast IP and rest anycast will do it for you by routing the query to Closest DNS server.
Routing on layer 3 switches or router will decide what is the best route
Very Good Informative Video...
Thanks Tushar :)
great post to explain anycast. well done!
Thanks :)
Thanks for the video..it's really very interesting..
I have one query..
In aneycast, end users will connect to pop and pop will route the traffic to the dc where the web server is hosted..in case of DDOS attack all traffic will reach to the web server through different POPs, and web server will be down..
Does a single web server can be protected by anycast ip or we need to place multiple web server in multiple regions?
Thanks for watching the Video and glad you liked it. Anycast IP is not for DDOS protection, you would need IP pattern filtering mechanisms and firewalls in place for those. It would also make sense to distribute your web servers in multiple regions if you can afford the cost of that.
what will be the answer for if two pop having same metric ?? is it loadbalancing
Only one of the POPs will usually be chosen for a client. I haven't seen load balancing of a client happening between POPs
Aditya Kamat can you tell on which basis it will preferred
It depends on what parameters are prioritized in the routing logic on these POPs. If the application is time critical, parameters based on latency will be compared between the pops. If its more like a video streaming/heavy data transfer service they would look for bandwidth related parameters between the two pop's. If you take a zero probability case where all parameters are the same between both the POPs, it would then choose randomly.
Great explanation of Anycast!
Thank you Stephen!
Almost 500 subscribers now!!
Hey thanks for the great video. You mentioned about BGP peering video. But there is no link. Do you mind sharing the link?
th-cam.com/video/Q_YpTh0Qq_k/w-d-xo.html
I talk about BGP peering at 10:50, sorry I'm on mobile and not able to share the video link with time in it.
And thanks for watching :)
@@AdityaKamat Thanks for the quick response
I will definitely subscribe ur channel and suggest my friends too
Thanks a lot Aman, means a lot to me :)
Good job. Thanks!
So basically for 13 root servers, you are saying that we have other redundant servers over the globe and they have the same IPs as of root servers. is the ip just same or its a complete replica of the root servers.
Ip address of the anycast server will be different and servers IP will be different.
Anycast will have a route to each of the server and have to put dns servers which are authoritative to answer the query in same anycast pool
Best video! Thanks a lot
Glad you liked it :)
Great stuff
Thanks Somashekhar. Glad you liked it!
It's sounds more familiar 😂,u don't have to worry about subscribers , India me hi boht hai log
Amazing
Video is very gud .bro ur English is gud but still I will recommend u use Hinglish
Sure, thanks for the suggestion!
Bsnl launch anycast dns .. good luck