blob: 3fa1d665e9c687f1842c05c8733013f5aec1dd5d (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
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.
|
