Score:0

How multiple load balancers work

bh flag

Assume we face x requests per seconds, where x is magic number of requests which single load balancer cannot handle.

So we have to use multiple load balancers to handle all requests, yes? But how? I need to load balance the load balancers farm? Poor idea, at the end it also occurs that single load balancer must handle all requests.

So how to use multiple load balancers to handle enormous number of requests?

Zac67 avatar
ru flag
There are other load balancing solutions than dedicated load balancers which cannot be cascaded. However your question is quite unclear and needs further details.
jp flag
You can utilize Equal-cost multi-path routing (ECMP) to send traffic to multiple load balancers. See for example [booking.com load balancing architecture](https://www.haproxy.com/user-spotlight-series/scaling-the-edge-how-booking-com-powers-a-global-application-delivery-network-with-haproxy) and [github architecture](https://www.haproxy.com/user-spotlight-series/inside-the-github-load-balancer)
HBruijn avatar
in flag
Your false assumption seems to be that all traffic, all requests must first reach a single destination before requests can be distributed evenly (by a loadbalancer). That isn't the case in properly designed solution. You can leverage for example the DNS layer to ensure that the visitors to `www.example.com` will see different IP-address based on their geographic location. And a smart DNS will ensure that vistors from a specific will will be directed to the `www.example.com` instance hosted behind nearby IP-address X while other vistors will see IP-address Y for `www.example.com`.
HBruijn avatar
in flag
Many other techniques, based on for example routing will do the same: ensure that all traffic does not end up in a single destination. Of course that takes a properly decentralised application design.
Score:4
jp flag

Instead of passing all the traffic through a single load balancer you can load balance on the network level utilizing Equal-cost multi-path routing (ECMP).

There are two presentations from HAProxy Conf 2019 which describe load balancing architecture of Booking.com and GitHub

Booking.com ECMP architecture

GitHub GLB

ws flag
or just use rrDNS
jp flag
@symcbean round-robin DNS isn't very good for HA.
ws flag
Perhaps, but installing an BigIp f5 on the local network of everyone who might visit your site is a little expensive. Also ECMP becomes very difficult to implement across multiple sites/multiple proviers.
ws flag
....and I wouldn't expect someone operating on the scale of booking.com or github to base their planning of a new infrastructure solely on discussion like this.
I sit in a Tesla and translated this thread with Ai:

mangohost

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.