4 ms·
"Cheating" is doing all of the work here. They are sending transactions to their "home location" which is worst case all the way on the other side of the plane
by jumpingmice 7y ago
"Cheating" is doing all of the work here. They are sending transactions to their "home location" which is worst case all the way on the other side of the planet and then they are only counting the time to apply the transaction at that location, instead of including the latency between the client and the homed frontend. So it's low latency if you draw the system boundaries in such a way but from the perspective of the client it's still high latency.
- AnthonBerg 7y ago... unless you preload the cheat setup, right? IOW you start to suspect that the client may execute a multi-region transaction. Then you can prepare by syncing data across regions.
- abadid 7y agoLatency is measured from the client. In the example in the post (and more details in the paper), you see the latency tail from when clients access data that is far from them. The challenge is to make multi-home transactions no worse than regular Paxos latency. In previous systems, this required multiple rounds of communication across the homes that are involved in that transaction. In systems like PNUTS, they would disable such transactions altogether. SLOG's ability to handle such transactions with latency no worse than Paxos is a big step forward.