trying to fix bandwidth calculations

[oweals/gnunet.git] / TODO

diff --git a/TODO b/TODO
index 6b3af6d6b50a4b35c22aeb386b301e23e87c1970..c3b8aa051f382590243f89742b9300c1845d3e4e 100644 (file)
--- a/TODO
+++ b/TODO
@@ -13,26 +13,18 @@ away), in order in which they will likely be done:
 * ARM [Safey]
 
 Urgent items (before announcing ng.gnunet.org):
-* TRANSPORT:
-  - nondeterministic transport testcase failures [Nate]
-* CORE:
-  - test currently fails
-  - request connect not working [Christian, need transport first]
-* TOPOLOGY:
-  - needs testing [need transport first]
-* HOSTLIST:
-  - needs testing [need transport first]
 * FS (basic anonymous FS only)
-  - implement FS service (P2P operations)
-    + implement sending of queries (low-priority push!)
-    + need to bound queueing of replies for other peers    
-    + content migration support
-    + hot path routing    
   - test multi-peer search/download
+    + existing bandwidth-reservation code does not work with FS: don't get
+      DBLOCK size bandwidth for reply for a while => trouble for testcase
+      => need to more aggressively increase bandwidth per connection 
+      => need to check that bandwidth is assigned properly to begin with
+    + Even with bandwidth-reservation code restricted, message is not
+      delivered to other core/peer/fs.  Why?
+    + Same query is sent to the same peer in rather rapid succession (10x/minute?),
+      should tune code to reduce repetition...
 * new webpage
   - run peer => have a 0.9.x hostlist
-  - improve basic documentation (configure, dependencies, what works, etc.)
-  - write minimal welcome/introduction to 0.9.x-development/webpage/status
 => Deploy(able) development network
 
 
@@ -41,8 +33,14 @@ Urgent items (before announcing ng.gnunet.org):
   - load: need way to determine network load (up/down)
   - trust: need *fast* way to check/update trust in peers
            (async peerinfo would not be right)
-  - OS: existing waitpid call is not nice (not integratable with scheduler! fix this!)
-    => should 'scheduler' support SIGCHILD explicitly? Always?
+* TRANSPORT:
+  - disconnect notifications from TCP not always as nice as we might want
+    them to be (but how can we do this cleanly given that UDP is connectionless?)
+  - TCP not used bi-directionally (especially important also for PONG!)
+  - hostlist testcase shows that (randomly) we jump to an address that
+    seems to be a plugin function *after* the plugin has been unloaded
+    (directly from scheduler); check that *all* plugin tasks are cancelled
+    after plugin is unloaded! (need better testcases for that...)
 * STATISTICS:
   - synchronous/asynchronous API (& implementation) is not nice;
     => provide notification-based API
@@ -52,6 +50,10 @@ Urgent items (before announcing ng.gnunet.org):
     [=> eliminate for need to tell ARM about service starts most of the time!] [Safey]
 * HELLO:
   - need function to test "equivalency" of HELLOs; use in topology!
+* TOPOLOGY:
+  - needs more testing (especially F2F topology)
+* CORE:
+  - needs more testing
 * Module features to implement:
   - advanced FS API parts
     + namespaces: fundamental namespace API
@@ -68,7 +70,7 @@ Urgent items (before announcing ng.gnunet.org):
     + gnunet-download (many options)
     + gnunet-directory (man page, options)
     + gnunet-pseudonym (all of it)
-    + gnunet-service-fs (remove failing on-demand blocks, stats, load-based routing, nitpicks)
+    + gnunet-service-fs (remove failing on-demand blocks, stats, hot-path routing, load-based routing, nitpicks)
     + datastore: do active migration support here?   
   - implement adv. FS testcases 
     + getopt API