Right. We have that issue in Colossus. I assume your storage system is on bare metal, that you don't have network-attached storage, and that you want to allow more than one node to be upgraded at a time?
-
This Tweet is unavailable.
-
This Tweet is unavailable.
-
This Tweet is unavailable.
-
This Tweet is unavailable.
-
This Tweet is unavailable.
-
This Tweet is unavailable.
-
Seeing as "rook" is > 86,000 loc, incredibly WIP to a certain degree, and subjectively very difficult to reason about, I would guess that this is overly optimistic, and forgets all the negative aspects of this giant code base. One bug and all your data goes *poof*, right?
What level of availability and durability are you expecting? This is an example of why Google builds higher-level storage systems on top of lower-level ones -- a machine can vanish, and even a whole zone can go away. One such system was Megastore:https://ai.google/research/pubs/pub36971 …
-
-
I agree! One disadvantage of the layered approach is that as the number of layers increase, so can the complexity. Assuming I can optimize one of the layers significantly (complexity and safety) I think this could have some great benefits, eventually obsoleting a layer.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.