CODEWITHSUNDEEP
.netweb-applicationsarchitecture
Satellite
Satellite

Reputation: 541

Application Architecture Advice (Scalability / Performance)

I’ve developed an entirely in-memory application using .NET. Data is loaded from persistent storage once at start-up, and from then only changes to objects are trickled back to persistent storage from the application tier through a work queue. It’s working great and benchmarking even better (100k api transactions per second). I realise this is a non-conventional and difficult to scale architecture. It was something of an experiment =)

Architecture

(Current architecture on the left, desired architecture on the right)

Now I’m starting to think about redundancy. I’d like to run two application servers side-by-side and load balance between them. This would mean keeping all in-memory objects synchronised - probably through persistent tcp connections shuttling binary serialized objects back and forth. Eventual consistency is ok. Conceptually I can see how this would work after both app servers are cold started to the same state from persistent storage, however I am having difficulty conceptualizing how I would instantiate and synchronize a new application server node while requests are streaming in. I guess this sounds like a snapshot + transaction log kind of thing?

Does this sound achievable? Is this kind of architecture in use anywhere of note?

Upvotes: 1

Views: 146

Answers (1)

Jeff Watkins
Jeff Watkins

Reputation: 6359

There are products out there which meet your needs. You haven't mentioned your nfrs so I don't know if you want to hand roll just for efficiency. Something like infinispan should work for you . What will be more challenging is transactionality of data given you're in the in memory space. It would be worth exploring how technical and destruction testing could be achieved.

Upvotes: 1

Related Questions