You need to understand how “the Internet” proper actually works: you have local (often quite large) networks connected together by globally inter-connection services. Ever wondered why it was called Internet?
Your local network provider traditionally connected to at least one node on the such an interconnect service and pays them to carry their traffic to other networks.
Sorry if this is teaching you to suck eggs, but from a lot of the “disconnect” threads I see a lot of networking ignorance - which may not apply to you.
Some major internet companies have set up their own interconnect services to link their data centres with each other and the wider network environment. Basically they need faster performance than the “normal” interconnects for synchronization of services. They will often put a local interconnect node into an ISP.
They can then offer their higher speed connections for use to subscribing parties and to the services they host.
CDNs, Content Delivery Networks do something similar.
It kook like CCP is paying to use the AWS Global Accelerator service to handle it’s traffic - providing a lower latency service (and possibly static content caching) for the Eve Online servers than relying on whatever interconnect service your, possibly cheapskate, ISP is using. Remember they sell on the basis of Mbps to the house from their node not how good a connection they have to other networks.
The awsglobalaccelorator.com address is the reverse look up of the virtual point of presentation for the AWS service, the sub-address is probably an identifier for their customer.
The AWS compute address may relate to content or services they are off loading to AWS from the main Eve servers in London. It would be a sensible thing for them to do - though what components they would be I’m not sure. Eve is a database heavy system needing good performance for the single instance that makes Eve what it is.
Given they are using AWS Global Accelerator I suspect that disconnects are most likely in the local ISPs or due to the ISPs not having (for various reasons) an accelerator node in it (small ISP, not viable for AWS to put a node in, ISP not wanting to pay for multiple interconnects - many reasons, not normally technical). Not having a node forces traffic to do a hop on their internet connection service to an AWS node.