Revert "Ipv4.Fragments: use a mutable LRU cache (Lru.M.t instead of mutable cache : Lru.F.t):" #423
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…utable cache : Lru.F.t):" This reverts commit 2201224.
hannesm
force-pushed
the
fragments-lru-f
branch
from
fe49e94
to
7120595
Compare
|
build works, can not test functionality right now because it makes mirage-nat fail to build. |
hannesm
added a commit
to hannesm/opam-repository
that referenced
this pull request
Feb 8, 2020
CHANGES: * Revert "Ipv4.Fragments use a Lru.M.t instead of Lru.F.t" (mirage/mirage-tcpip#423 by @hannesm) A Lru.M.t allocates a Hashtbl.t of size = capacity (= 256 * 1024 in our case), this leads to excessive ~2MB memory consumption for each Fragment cache, reported by @xaki23 in mirage/qubes-mirage-firewall#93 * use SOCK_RAW for an ICMP socket in the unix sockets API (previously used SOCK_DGRAM which did not work) reported by @justinc1 in mirage/mirage-tcpip#358, fixed in mirage/mirage-tcpip#424 by @hannesm * tcp is now compatible with lwt >= 5.0.0 (where Lwt.async requires a function of (unit -> unit Lwt.t) (mirage/mirage-tcpip#370 mirage/mirage-tcpip#425 @cfcs @hannesm, issue mirage/mirage-tcpip#392 @emillon) * Add a dependency on dune-configurator to support dune 2.0.0 (mirage/mirage-tcpip#421 @avsm)
This was referenced Feb 8, 2020
mgree
pushed a commit
to mgree/opam-repository
that referenced
this pull request
Feb 19, 2020
CHANGES: * Revert "Ipv4.Fragments use a Lru.M.t instead of Lru.F.t" (mirage/mirage-tcpip#423 by @hannesm) A Lru.M.t allocates a Hashtbl.t of size = capacity (= 256 * 1024 in our case), this leads to excessive ~2MB memory consumption for each Fragment cache, reported by @xaki23 in mirage/qubes-mirage-firewall#93 * use SOCK_RAW for an ICMP socket in the unix sockets API (previously used SOCK_DGRAM which did not work) reported by @justinc1 in mirage/mirage-tcpip#358, fixed in mirage/mirage-tcpip#424 by @hannesm * tcp is now compatible with lwt >= 5.0.0 (where Lwt.async requires a function of (unit -> unit Lwt.t) (mirage/mirage-tcpip#370 mirage/mirage-tcpip#425 @cfcs @hannesm, issue mirage/mirage-tcpip#392 @emillon) * Add a dependency on dune-configurator to support dune 2.0.0 (mirage/mirage-tcpip#421 @avsm)
2 tasks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This reverts commit 2201224.
The motivation behind this change is that Lru.M.t uses a hashtable, pre-allocated in the size of the capacity (256 * 1024 ~> 2MB just for the keys (stored in an array)). This leads to quite some memory usage (as observed in mirage/qubes-mirage-firewall#93).
But the issue is slightly more complex: an OCaml hashtable has amortized time complexity of O(1), but in worst cases it may be much bigger (adding an element may lead to resize, which means allocation of a fresh array, re-hashing of all keys, and copying) -- leading to unpredictable behaviour, which is not what we want in real systems. Also, a hashtable is never shrinking its array.
I'm fine to revert this change once we find the performance bottleneck being in the IPv4 fragment cache.
//cc @talex5