Change Details

This patch improves block caching algorithm for remote media on low-BW/high-delay links by only triggering read-ahead if two subsequent blocks have been requested by the upper-level code. This cuts boot time over IP KVMs noticeably, since the initialization stage reads bunch of small .4th (and now .lua) files that are not in the same cache stripe. The original regression came in 2017 with revision 87ed2b7f5a. Even doing something like ls is painful on KVMs ever since.

This patch improves block caching algorithm to work better for remote media on low-BW/high-delay links by only triggering read-ahead if two subsequent blocks have been requested by the upper-level code. This cuts boot time over IP KVMs noticeably (like 15-20x from less than a minute to 10-15 minutes till it even start loading kernel), since the initialization stage reads bunch of small .4th (and now .lua) files that are not in the same cache stripe. The regression came in 2017 with revision 87ed2b7f5a. Even doing something like ls is painful on a remote KVMs ever since. Here is a real-world deployment back in 2018 taking 10 minutes just to start loading kernel: https://youtu.be/g1iWzExHFJE

This patch improves block caching algorithm to work better for remote media on low-BW/high-delay links by only triggering read-ahead if two subsequent blocks have been requested by the upper-level code. This cuts boot time over IP KVMs noticeably (like 15-20x from less than a minute to 10-15 minutes till it even start loading kernel), since the initialization stage reads bunch of small .4th (and now .lua) files that are not in the same cache stripe. The original regression came in 2017 with revision 87ed2b7f5a. Even doing something like ls is painful on KVMs ever since.a remote KVMs ever since. Here is a real-world deployment back in 2018 taking 10 minutes just to start loading kernel: https://youtu.be/g1iWzExHFJE