-
Notifications
You must be signed in to change notification settings - Fork 24
/
Copy pathlearnrecord.txt
146 lines (100 loc) · 4.72 KB
/
learnrecord.txt
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
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
lecture 1
------------
听了Robert Morris的6.824第一堂课，不用ppt，没有教材，拿着两页纸开讲，强调实践。佩服！
lecture2
------------
先讲了RPC，以及at-least-once的缺点，at-most-once的好处，然后讲thread，以golang为例，介绍了golang中的goroutine, channel, sync.Mutex的情况。期间学生提了不少问题，课题气氛比较活跃。课程中花了近 40 mins的时间用在golang上，结合看golang的代码，确实比clang简洁不少。
2/5/2013~2/9/2013
lab1 hand out in Tue
Tue~Fri: learn golang
Fri~Sat: finish lab1
$ git clone http://am.lcs.mit.edu/6.824-2013/golabs-class.git 6.824
$ cd 6.824
$ ls
src
$
The design idea is important, also should understand the goloang program. At first I made a mistake that let every lock contain a request seq num, and seq num is increased . In fact, the client should contian a request seq num, and the seq num could be rand (more easy to implementation)
key:
0 notice at-most-once mechanism in RPC process. and only kill Pri or Bak srv (only one , but not two all), and the killed pri srv can not handle the request. Dying pri srv could handle rqueest but not reply info to client.
1 define xid(unique id) for client's request's seq number
2 server should store the used(or seen) request(with unique id) and related reply result, then it can decide if the request(xid) has been handled before?
3 Pri srv does the same thing as that of Bak srv. But when Pri srv finish the handling, it have to diliver a rpc call to Bak srv to ask Bak srv handle the same request from client.
I write a design.txt for lab1
lecture3
-------------
talk lab2 info, and I missed this part. :( (~60 mins)
then talk the paper (http://pdos.csail.mit.edu/6.824/papers/bressoud-hypervisor.pdf), should read paper first, then I can better understand what RTM said (~35 mins)
student have several questions.
Lab 2 Part A Due: Feb 15, 5:00pm :(
Tutor
----------------
feb 15
TUT 2: Design project 1 (run by TAs)
Preparation: Book section 2.5: "Case study: UNIX File System Layering and Naming"
小班制，10人左右。
王曦用近50分钟，讲解了fs的设计与实现（基于unix fs），觉得他已经算讲得很好了，但感到对没有学过操作系统的学生来说难度较大。
lecture4
--------------
6.024::lec4::Robert Morris:: talk about lab2 design (he said simple, but I spend a lot of time), then paxos, very difficult to understand, even I read some materials before class. Robert talk about the implement framework of paxos and give some example. then ask to finish LAB3!
lecture5
--------------
holiday
lecture6
--------------
Robert Morris::lec6::2/26/13:: talk about paper: Replication in the Harp File System, key part: GLB | < | LB < | AP < | CP < |FP , then give some examples with the explain of key parts. then Robert or students ask How /What, etc. Hard for me to understand this paper. :(
lecture7
--------------
||6.824::Lec7::02/28/13::Spanner: Google’s Globally-Distributed Database|| I'm lost. Even I read the translated Chinese paper of span and some cn webpage, I also didn't understand the internal of spanner. Hope when I finish lab4, I can know more. DS is harder than I envisage.
lecture8
--------------
||6.824::Lec8::3/5/13:SUNDR paper (OSDI04) First: continue to talk about spanner. Robert think this paper is interesting. talk baout TrueTime, the comparing of different times now()-->[early, late] talk 55 mins on spanner. I need to understand. Now sundr paper, and it seems sample.
lecture9
--------------
||6.824::Lec9::03/07/13
consistency
meaning of read/write
perf v.s. easy for prog
intro:
--strict consistency
--sequential consistency
Then discuss the paper of Prof. Kai Li (Memory Coherence in Shared Virtual Memory Systems )
lecture10
--------------
||6.824::Lec10::3/12/13
consistency model
treadmarks solvers
false sharing -- two writes on the same page
write application - one byte write
write diffs
Next goal: allow multiple writable copies
release consistency (RC)
acquire(l)
x=s'
release(l) -->broadcast write diff on release to all copies of page(S)
lazy release consistency
the improtant part is tell the implementation and the demo/examplify the process
Rules
-----
1. every var protected
2. must lock around write
3. must lock around read(if you want to see the fresh data)
lecture11
----------------
||6.824::Lec11::3/14/13
causal order consistency
vector timestamps
file synchronizers
-----------------------
Do it first, fix later
Replicas
causal consistancy
If x causally precedes y
all see x than y
vector timestamps -- version vectors
VT op
a<b -- Vi a[:] =< b[:] and Sj a[j]=<B[j]
highest seq from each host
------------
File Sync
No lost Updates
next time Mid-term quiz in 2-190