spoiler alert 9pfs does read only mmap by default color me sad https://twitter.com/b0rk/status/984431593954590728 …
-
This Tweet is unavailable.
-
Replying to @jessfraz
For shared; for private maps write is always possible. There's really no good way to implement shared writable mmap of remote file though...
1 reply 0 retweets 1 like -
Replying to @RichFelker @jessfraz
You'd need cooperation between both kernels (client & server) such that only one keeps the mapped pages writable at any given time (at mmu level), and faults change ownership.
2 replies 0 retweets 1 like -
Replying to @RichFelker @jessfraz
Couldn't you use a mechanism similar to RDMA? Client kernel wouldn't really be involved; cache coherence gets handled by the server kernel.
1 reply 0 retweets 0 likes -
There is no way to do it without client kernel involvement, because both the client and server (and other clients, through the server) need to exclude writes by the other in order to impose a global order on atomics in the shared memory.
1 reply 0 retweets 0 likes -
The problem is roughly equivalent to memory synchronization in NUMA nodes.
1 reply 0 retweets 0 likes -
Replying to @RichFelker @jessfraz
If the problem is roughly equivalent, what is wrong with using similar cache coherency protocols as used for ccNUMA? Obviously, performance would be terrible.
1 reply 0 retweets 0 likes
It could only be done with special clients and servers that have kernel hooks to do that. Normally 9p server does not involve kernel at all.
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.