summaryrefslogtreecommitdiff
path: root/technical
diff options
context:
space:
mode:
authorKévin Le Gouguec <kevin.legouguec@gmail.com>2018-06-07 18:25:23 +0200
committerKévin Le Gouguec <kevin.legouguec@gmail.com>2018-06-07 18:25:42 +0200
commit01f75c314e19d0ab671acdc7a46f993653dde0d2 (patch)
treead81d07a0ca2de6c2f61d54c7c6593f0bb94a260 /technical
parentbc858629703748e4ace70c9dd08128e55148f970 (diff)
downloadmemory-leaks-01f75c314e19d0ab671acdc7a46f993653dde0d2.tar.xz
Start adding notes on linux.conf.au 2018
More to come.
Diffstat (limited to 'technical')
-rw-r--r--technical/reviews/linux.conf.au-2018.md67
1 files changed, 67 insertions, 0 deletions
diff --git a/technical/reviews/linux.conf.au-2018.md b/technical/reviews/linux.conf.au-2018.md
new file mode 100644
index 0000000..3fa1d66
--- /dev/null
+++ b/technical/reviews/linux.conf.au-2018.md
@@ -0,0 +1,67 @@
+# linux.conf.au 2018
+
+## Making Technology More Inclusive Through Papercraft and Sound
+
+By Andrew Huang.
+
+I like how the talk goes over a range of cross-domain topics:
+
+- high-level motivations
+
+Improving inclusiveness is necessary to make open-source actually
+empower people; right now a very small subset of the population is
+computer-savvy enough to take advantage of it. If the situation does
+not improve, a handful of developers will hold a lot of power over
+lots of alienated users, and lawmakers may resort to "preposterous"
+solutions to attempt to regain control, e.g. license bonds for
+software developments.
+
+- Kickstarter campaign management
+- design choices & rationale
+ - "China-ready"
+ - "patience of a child" constraint
+- gory hardware details
+- the end result
+
+## QUIC: Replacing TCP for the Web
+
+By Jana Iyengar.
+
+Starts by introducing impressive application performance improvements,
+although where were those measured? E.g. rural areas?
+
+Advantage that can already be inferred from the layer view: QUIC needs
+fewer handshakes than TCP+TLS.
+
+Achieves 0-RTT when the server's cryptographic credentials are known.
+
+Supports "stream multiplexing": the upper layer (e.g. HTTP) can
+transfer multiple objects independently in a single connection.
+Losing part of one object does not block the others: retransmission is
+managed at the stream level, not at the connection level.
+
+On top of UDP: allows userspace (Chrome) implementation.
+
+> If you think of layers as a set of functions, things that you want
+> done, UDP is not a transport protocol.
+
+I.e. UDP does not provide reliability, same-order delivery…
+
+Jana was "in the SCTP bandwagon".
+
+They actually have *better performance improvements* for *bigger
+latencies*? Nice.
+
+> § QUIC improvements by country
+
+👏
+
+(Of course the end goal is probably to make sure regions with poor
+connections do not miss out on the adfest; still, these remain welcome
+technical improvements)
+
+Transport headers are encrypted to prevent "middlebox ossification".
+They left a *single* byte unencrypted (the flags byte): this allowed
+middleboxes to observe that it kinda had the same value on most
+connections, assume that this was a "nominal" value, and block traffic
+when this value differed.