g*******y
发帖数: 1930 | 1 来自主题: JobHunting版 - 问一道老题 将原来7台机子上存的keys拿来rehashing一次，按照新的hash值移动到新的server。
考虑到data比较大，移动会有很大cost，可以在新server相应的储存该key的地方，设一个forward的标签，把fetch data的request forward到旧server上 |
|
H*M
发帖数: 1268 | 2 来自主题: JobHunting版 - 问一道老题 第一个方案肯定不行的，如果hash function没规律的话，移动data复杂度太大，最坏的
N太机器全要移动。
forward idea不错。不过可能会要forward好多层。每变一次hashfunction就要forward
。
设一个forward的标签，把fetch data的request forward到旧server上 |
|
G**********s
发帖数: 70 | 3 An array A[1...n] contains all the integers from 0 to n except for one
number which is
missing. In this problem, we cannot access an entire integer in A with a
single operation.
The elements of A are represented in binary, and the only operation we can
use to access
them is “fetch the jth bit of A[i]”, which takes constant time. Write code
to find the
missing integer. Can you do it in O(n) time? |
|
g*l
发帖数: 385 | 4 感觉有错误. 你算 costPerKm 用到了 numTrips, 这是包括了最后一趟的, 也就是说全
运走了, 没
有 remingingNuts. 但你在计算 traveled 是假设有 remainingNuts, 可是同时用的
costPerKm是以没有 remainingNuts 来算的. Circular logic?
// # trips you would travel back and forth
int numTrips = 2*(ceil(N/C) - 1) + 1;
// how many nuts you consume per km
double costPerKm = numTrips * F;
// remaining weight of nuts after consumption
double remainingNuts = C*(ceil(N/C) - 1.0);
// this is the distance you are able to travel before you
// reac... 阅读全帖 |
|
|
g***o
发帖数: 4 | 6 Follow the solution from gloomyturkey. Build the skip list of 1~n. Every
time, simply fetch the number at the position of count array[i] and delete
it from the list, which is lgn time. |
|
n******a
发帖数: 114 | 7 刚面完，累。基本的3人，后来又加上另一个组的一个，完了等了40多分钟，跟这4个人
的boss的boss谈了不到半小时，基本闲扯。
第一个很nice，上来就是聊背景和简历上的东西，聊了都1个小时了，才发现时间快到
了，随便问了个design and test vending machine的问题。
第2个也不错，很nice，吸取上一个教训，留够了时间谈算法，2个sorted linked list
，怎么combine，给了2个算法，recursive和non的，讨论好坏，最后写test plan，这
次提前了5分钟完工。
第3个比较话少，lunch interview，基本都是我在说，他吃完了，我才一半。上楼继续
算法，给了tree，但是每个node还包括父节点的指针，要求找LCA，注意，不是bst。这
个因为一开始没理解清楚，一直当是bst在想解法，耽误不少时间，最后他给了提示的
情况下，解出来了，是不是最好的也不知道，谈test plan时，开始急着总结了，说下
一个人等着在。
第4个南非的，带点英国腔。考虑一个application which is constantly fet... 阅读全帖 |
|
b*****n
发帖数: 760 | 8 1. You have a class that supports to input sample records and to compute the
average of the samples. The class has two members: total and count. How
would you make the class thread-safe? If 99% of the time average() is called
, how to optimize for that?
2. Talk about your recent interesting project/bug.
3. You have 100 files, each containing 10G sorted integers. How to merge all
integers into one sorted file?
4. Write a function to reverse digits of an integer. E.g. 123 --> 321, -890
--> -98.
5.... 阅读全帖 |
|
d********w
发帖数: 363 | 9 1. smart pointer implementation
2. thread safe hashmap, how to improve concurrency, I said to read/write
lock, use many locks to reduce 粒度, each lock cover one part of range of
buckets.
3. stack implementation, use array
4. insert node into linkedlist
5. LRU cache, getItem if not hit cache, should visit database to fetch that.
HashTable + DoubleLinkedlist
6. fib
7. implement spin lock, test and set 原语,
8. 32/64 alignment pading,
struct s{
char a;
int b;
void* c
long long d;
}
sizeof(s) in 32b... 阅读全帖 |
|
d********w
发帖数: 363 | 10 1. smart pointer implementation
2. thread safe hashmap, how to improve concurrency, I said to read/write
lock, use many locks to reduce 粒度, each lock cover one part of range of
buckets.
3. stack implementation, use array
4. insert node into linkedlist
5. LRU cache, getItem if not hit cache, should visit database to fetch that.
HashTable + DoubleLinkedlist
6. fib
7. implement spin lock, test and set 原语,
8. 32/64 alignment pading,
struct s{
char a;
int b;
void* c
long long d;
}
sizeof(s) in 32b... 阅读全帖 |
|
f*****7
发帖数: 92 | 11 DP的定义是递归的
我们要得到原问题的最优解，就得先算出若干个子问题的最优解，然后extend到原问题
。我们不断地把大问题归结为若干个小问题，最后就是解决base case。这种思维方式
by nature就是递归的思想。----最优子结构
对于多个大问题，要解决它们所用到的子问题可能有重复。所以我们需要用cache记录
已经计算过的子问题，如果该子问题被解决过了，直接从cache中fetch子问题的解。如
果该子问题没有被解决，那么就解决这个子问题，并且将solution存在cache对应的
entry里。----重复子问题
这两个是DP的重要性质。
CLRS对于DP的算法有两种
1. Top-down recursion with memoization
这种写法就是递归，用数组保存子问题的solution。
好处在于解决某些大问题，并不需要tabulate所有的子问题的时候，我们可以节约计算
时间，类似lazy evaluation。子问题只有在需要的时候才会被计算。第二个好处是直
接从定义出发，递归结构清晰，易于调试。
坏处是递归函数需要OS维护stack frame，如果问... 阅读全帖 |
|
d*****y
发帖数: 205 | 12 我胡说几句，可能说得不对，请别介意。
个人觉得，最后一个large scale题目其实答得挺好的，这题不会太失分
如果是我答的话，第一个问题可能不会直接问面试官“DHT,还是cluster”，
这一步可能跳的有点快了，
第一个问题可能会问，key 是什么样的key，value包含哪些（正常的话，面试官会说
key是string，或者什么都可以，value也是什么都可以，之类的，这问题虽然简单，包
含很多信息，问一下这个问题，面试官很轻松的回答你，就表明往这个方向走是对的）。
我的猜测是，这个题不是要考你BigTable是怎么design的，那个估计很多人面试前都准
备过，
其实估计他就是想问最简单的DHT原理，这个题面的设计，甚至于不expect你懂得DHT，
否则他就直接问你tell me something about DHT
(G倾向问比较基本的概念和问题，就算背景是large system，别被现有成熟技术束缚，
想一下不是每个人都有large scale system的研究和工作背景，但是每个人都会被问到
这样的题目，如果有一个人是学EE的，来了也会被问这道题的，他压根就没听... 阅读全帖 |
|
y**********u
发帖数: 6366 | 13 工作2年
我已经tlb啊，cache啊，都还给老师了
只知道cache miss的如果是指令，就要stall cycle,如果是数据读miss，就要fetch一
个line (多个字)，如果是写miss，好像可以buffer write，同时invalidate cache
line?
要做
所以 |
|
b*********3
发帖数: 748 | 14 来自主题: JobHunting版 - SEO题目 一个很简单的hyperlink指向另一个网页，但是这个网页要call 3rd party web
service，需要3到5秒才能拿到response然后display，问怎样做到search engine
friendly？
googlebot会把速度考虑在内，太慢会影响排名。其他方面的考虑是这个web service很
慢，所以实现的时候是交给background thread pool去fetch response, set the
response in cache。前台polling从cache得到数据。 |
|
Z*****Z
发帖数: 723 | 15 小小的补充一下
x == 0的时候也成立~~~~~~~~~~~~~~~~~~
3.1 Fetch the right most 1-bit
Ans: x & -x
30th) |
|
a*******y
发帖数: 1040 | 16 how will you implement dictionary for client . suppose data is stored in
database .you don't need to implement dictionary but explain entire
interaction and systems involved .
The interviewer was not interested in knowing how dictionary is internally
implemented . he was more interested in understanding bigger picture ie.
high level , eg.
i explained you will have ajax call from client page to fetch matching words
, explained arguments that i will pass in ajax , explained high level server
side ... 阅读全帖 |
|
L******9
发帖数: 3 | 17 面试前也在板上看了不少面经。新人也贡献一下自己的经历。
在career fair上提交简历。之后校园面试。面试前不分SDE,SDET或PM
面试共30分钟。一开始问了几个非技术类得题。比如想做三个工作中的哪一个，为什么
？ 说说自己做过的比较challenge的project。由于楼主是学生，就随便扯了个class
project。
然后coding部分。第一题很简单。给一个n*m个小方块拼成的长方形。从左上角出发到
右下角。每次只能向右或向下移动一步。求总共有多少种可能得路径。 先给出coding
算法后，又让我给出数学解法。
第二题说是high level的问题。比如写一个facebook birthday的app。希望手指点到相
应人的时候会出现人的profile。为了加快速度，只能在开启app的时候提前fetch所有
朋友的profile。但是同时app也需要从网络上提取其他data（有可能有更加高的
priority）为了不让这些job之间冲突，问如何处理这种情况？
由于不是cs专业 lz当时很懵。完全不懂app实现那些相关的东西。后来瞎扯了几句优先
队列。于是问题就变成如何... 阅读全帖 |
|
f*****7
发帖数: 92 | 18 using a min-heap, a max-heap, and a median variable to implement a data
structure to fetch median of a integer stream in O(1) time. of course, it
requires additional storage. |
|
v***o
发帖数: 287 | 19 15. He: how would you design a distributed key-value store
Me: DHT or just using clusters
He: details?
Me: we have a large number of machines. first we use a hash function to
retrieve machine ID from the key. Then we connect to the machine and use
another hash function to retrieve the address from the key. Then fetch data
from that address.
两个hash function对一样的key, 而且，第二个hash(key) 要address， 不太可能吧。
decided
interviews |
|
v***o
发帖数: 287 | 20 15. He: how would you design a distributed key-value store
Me: DHT or just using clusters
He: details?
Me: we have a large number of machines. first we use a hash function to
retrieve machine ID from the key. Then we connect to the machine and use
another hash function to retrieve the address from the key. Then fetch data
from that address.
两个hash function对一样的key, 而且，第二个hash(key) 要address， 不太可能吧。
decided
interviews |
|
j******s
发帖数: 48 | 21 好吧，自己回答一下思路，虚心求各路大神拍，给点意见
第一题应该是建立一个binary tree,然后recursive求每一个节点的左右子树的最高高
度,他们和就是在这一位上的largest distance，可以online做，但是需要在每个节点
保存左右子树的高度并且动态更新。
o(n) time + O(n) space
第二题,
Assumption:
1 Log file is typically very small, operations are append, delete and read.
2 Cluster is built on Hadoop, which has a chunk size of 64MB, roughly, and
it might be too large and inefficient if we use this chunk size for the log
file.
3 Log information is typically not very important, less effort is needed for
redundan... 阅读全帖 |
|
z*******3
发帖数: 13709 | 22 给你发的邮件里面应该会有具体的描述，你没看过？
Software Development Engineers - Machine Learning and Big Data Analysis
In this role, you will design and build systems to analyze huge amounts of
anonymized data to provide innovative features to our customers. Some of
these will be directly visible to customers, such as Trending Pages. Others
will empower browser optimizations, like predicting and fetching the next
page that the customer is going to load in order to provide a zero latency
experience. These systems will ... 阅读全帖 |
|
f********d
发帖数: 51 | 23 if you read that section carefully, it's not defining what byte code
instruments they are. it's defining how you lookup your field and method.
and of course, a speicial javac can refuse to do that. then fundamentally it
's not java, maybe javaX. :) it's easier for scala, because it's called
scalac, it doesn't need to worry about the name and dont have to claim
himself as java. it only needs to say i'm compatible with java AT RUNTIME.
you have to follow jvm spec when implement, you can give warni... 阅读全帖 |
|
c*********m
发帖数: 43 | 24 最后一题写了个，大家看看有啥bug没
#define MAX_BUFFER_SIZE 10 * 1024
extern int fetch10k(char *buffer);
char temp[MAX_BUFFER_SIZE + 1];
int fetch(int size, char *buffer)
{
assert(size > 0 && buffer != NULL);
static int currentPos = 0;
string s;
int currSize = 0;
static int len = 0;
while (1)
{
if (currentPos != len)
{
int i;
for (i = currentPos; i < len; i++)
{
... 阅读全帖 |
|
r**h
发帖数: 1288 | 25 Yelp这家第一轮HR Screen的时候，HR会问你一些CS方面的基础问题
个人根据自己被问到的经历和玻璃门上的面经总结了一下（实际上没有这么多题不过基
本上都在里面），希望对申请他家的各位有所帮助
1. size of unsigned integer
2. http port no.?
3. ssl full form? (Secure Socket Layer, Encrypt in Transportation layer)
4. use of grep and kill
5. the runtime of adding something to a linked list? O(N)
6. SSL和TLS（Transport Layer Security）的区别：TLS是SSL的升级版（TLS3.0 =
SSL1.0）TLS先handshake再secure
7. hashmaps, DNS(Domain Name System),
8. python native datatypes Boolean, Number, String, Byte, List, ... 阅读全帖 |
|
r**h
发帖数: 1288 | 26 Yelp这家第一轮HR Screen的时候，HR会问你一些CS方面的基础问题
个人根据自己被问到的经历和玻璃门上的面经总结了一下（实际上没有这么多题不过基
本上都在里面），希望对申请他家的各位有所帮助
1. size of unsigned integer
2. http port no.?
3. ssl full form? (Secure Socket Layer, Encrypt in Transportation layer)
4. use of grep and kill
5. the runtime of adding something to a linked list? O(N)
6. SSL和TLS（Transport Layer Security）的区别：TLS是SSL的升级版（TLS3.0 =
SSL1.0）TLS先handshake再secure
7. hashmaps, DNS(Domain Name System),
8. python native datatypes Boolean, Number, String, Byte, List, ... 阅读全帖 |
|
h**o
发帖数: 548 | 27 书上用对半排除法为什么就是time = O（n), 是否有问题那？
fetch(a, i, col) 需 O(1), 所以 把一个a[i]排除掉就需lg(n), 把所有奇/偶a[i]排
除掉就需n * lg(n),这本身就需 time = O（n lgn). 所以这道题time complexity 应
该是 nlgn, 还不如直接用bitmap（求a[i]，然后把a[i]bitmap) 来做呢。
请问我的理解是否对， partition然后排除法好在那儿那 |
|
b*******e
发帖数: 123 | 28 An array A[1...n] contains all the integers from 0 to n except for one
number which is
missing. In this problem, we cannot access an entire integer in A with a
single opera-
tion. The elements of A are represented in binary, and the only operation we
can use
to access them is “fetch the jth bit of A[i]”, which takes constant time.
Write code to
find the missing integer. Can you do it in O(n) time
Inspired by leetcode question, can we figure out 0^1^2^...^n first then do
an xor for each element i... 阅读全帖 |
|
f******n
发帖数: 53 | 29 map[7][6]
000000
012110
000010
011110
010010
021120
000000
//Assume 2 is product, 1 is blank, 0 is block. 2 is also block before
fetching
use BFS
result start at map[4][4]
order map[1][2] map[5][4] map[5][1] total 20 steps |
|
n********e
发帖数: 41 | 30 楼主 看到 你 isDiffByOne 函数就知道你必然要超时
Leetcode免费给你用 就别挑剔了。凡事要先从自身找问题
你那个测试不是大数据
真正的大数据在这里：
start = "nanny";
end = "aloud";
String[] d = {"ricky","grind","cubic","panic","lover","farce","gofer
","sales","flint","omens","lipid","briny","cloth","anted","slime","oaten","
harsh","touts","stoop","cabal","lazed","elton","skunk","nicer","pesky","
kusch","bused","kinda","tunis","enjoy","aches","prowl","babar","rooms","
burst","slush","pines","urine","pinky","bayed","mania","light","flare","
wares","wom... 阅读全帖 |
|
p****a
发帖数: 447 | 31 没load进来会cache miss, cache miss 和 hit当然不一样了
估计他的意思是最后都进cache了?
fetch |
|
j********2
发帖数: 82 | 32 我也不知道restore怎么写，贴一下我的flattern. 没用recursion, 用了一个stack:
void flatternAugListWithStack(AugListNode *&node)
{
if (!node) return;
stack st;
AugListNode *p = node; // current node
AugListNode *tail = node; // current tail
while (1)
{
if (!p && st.empty()) break;
// after we go over the children, fetch it from the stack
if (!p) {p = st.top(); st.pop(); tail->next = p; continue;}
if (!p->next) tail = p;
if (!p->child) p = p... 阅读全帖 |
|
z******g
发帖数: 271 | 33 barrier和volatile显然不是
barrier是用来prevent memory reordering的，而volatile只是强制compiler生成读取
memory的code。这俩可以说是primitive中的primitive
正确的应该是mutex和atomic operations（fetch-and-xxx，compare-and-swap） |
|
t********5
发帖数: 522 | 34 I was able to fetch the page with request in Node.js |
|
x****r
发帖数: 158 | 35 已跪。LOL。
============================
1. Coding
主要问了两个问题，一题是 Given two nodes in a tree, find their common
ancester。还有一题忘了，但也是 Leetcode 原题。
这一关我过得不错，很快代码写完，又被加问了一道，关于如何设计 hash table，how
would you design the structure? How would you handle collision and explain
in detail?
2. Coding
Given a list of points on a 2-D plane, return the max number of points pass
a single straight line - 同样是 Leetcode 原题。
这一关也还行。很快也写出代码，but the solution given wasn't optimal in terms
of space complexity, and missed a corner... 阅读全帖 |
|
x****r
发帖数: 158 | 36 已跪。LOL。
============================
1. Coding
主要问了两个问题，一题是 Given two nodes in a tree, find their common
ancester。还有一题忘了，但也是 Leetcode 原题。
这一关我过得不错，很快代码写完，又被加问了一道，关于如何设计 hash table，how
would you design the structure? How would you handle collision and explain
in detail?
2. Coding
Given a list of points on a 2-D plane, return the max number of points pass
a single straight line - 同样是 Leetcode 原题。
这一关也还行。很快也写出代码，but the solution given wasn't optimal in terms
of space complexity, and missed a corner... 阅读全帖 |
|
B********t
发帖数: 147 | 37 一个面试题：给一堆URL, 怎么实现同步fetch resources? |
|
|
e********3
发帖数: 229 | 39 设计一个itunes/google play购买物品的系统 with high availability. 比如某个游
戏是1刀,你点击之后server会跟银行talk要从你卡里扣除这一刀.还有这charge的一刀
有一部分是要给developer的.这整个过程如何设计
这个是不是应该上kafka.目前能想到的是:
1. front end发送购买event把request经过load balancer发送到application server
preprocess这个request.
2. 然后把提取出来的相关user id, item id等信息作为message发送到kafka.
3. consumer server从kafka里拿出kafka的message,去database 取出相关user的卡信
息,发送给另一个server去Process
4. 这个server跟银行talk,charge了这一刀.然后update database记录下该user相关购
买信息.
5. consumer server commit.继续fetch message from kafka
... 阅读全帖 |
|
d*******s
发帖数: 65 | 40 非常同意。
被苹果的一个前微软principle面过，人让写spin_lock，我用load-link/store-
condition写，结果人像看见史前生物一样问我load-link/store-condition是干嘛的。
我又改成test-and-set，用TAS，结果人又是像看见史前生物一样问我TAS是啥意思，于
是我就彻底无语了。。。后来才知道人要写fetch-and-add的版本。然后就没有然后了 |
|
D***0
发帖数: 138 | 41 【 以下文字转载自 Java 讨论区 】
发信人: DK100 (dark knight), 信区: Java
标 题: java多线程问题请教
发信站: BBS 未名空间站 (Tue Jan 5 12:34:03 2016, 美东)
操作比较简单，就是有个cache，当cache miss时从数据库取值，同时更新cache。这里
的key是由两部分组成的，一个int和一个interface，函数输入是这两个。问题是在多
线程下如何synchronize。 我想是用concurrentHashMap当cache
Type是一个interface
一开始想用下面这个Key来做map的key，但是后来觉得不对，这也是想请教的一个地方
class Key {
public int a;
public Type t;
public Key(int a, Type t){...}
public int hashCode() {...}
public boolean equals(Key k) {... }
}
List f(int a, Type t) {
... 阅读全帖 |
|
d**********6
发帖数: 58 | 42 最近随便面了一下Yelp三番，没想到拿了个offer, 125K base ＋ 125K／4 RSU ＋
10k relocation，不算bonus是156k一年。目前推荐 advertisement backend组，可惜
manager是个三哥。绿卡说可以申请一下立即办。就这些，大神莫嘲笑，我的个人感觉
是湾区offer好大。纽约生活成本差不多，类似Yelp这种三线公司工资我应该拿不了这
么高。
本人Master＋1.5年 纽约某教育软件公司，base 9w8，其余几乎没啥了。 还没
negotiate，只有这一个offer，也谈不了几个钱，这对我个人已经是很大的raise了。
求教各位湾区的大神，在SF这个工资值得接么。
按板上惯例上面筋：
电面是group anagram和扯淡
Onsite：
1. match 俩address，给一个近似度的评分。非常简化及其简单，感觉主要是看coding
习惯
2. longest substring with all unique chars
3. Multiply two string
还问了git fetch干... 阅读全帖 |
|
G**O
发帖数: 147 | 43 我觉得还是能用cursor
cursor之前的timeline都被query过了，
cursor之后的timeline，全部fetch，然后按照rank 排序，返回rank高的几个并且在DB
里面mark成visited。。如果cursor恰好被mark了，就把cursor移动到下一个。
不过这样的问题在于，如果我是分布式的，我要把好几个replica都mark成visited，可
能会比较慢。
的 |
|
L****n
发帖数: 12932 | 44
it's hard to imagine a little townhome that size could fetch close to 300k.
i guess each local market is so very different. how much the rent would be
if you just rent it? |
|
m***y
发帖数: 14763 | 45 This is a little bit too far fetched. It's a damned house, no lives
endangered. And the house in on fire due to the owner's stupidity.
I think a more proper metaphor is a house under water. Say, due to the owner
's stupidity, his house is under water, should tax dollar be used to save
his ass? The popular answer in this county is, actually, YES!
Ideally, I'd beg to differ, we can only help those who help themselves. But
in reality, our tax dollars are squandered, anyway, so why should I prefer
t... 阅读全帖 |
|
c**i
发帖数: 6973 | 46 Fanny Liu, Flower farmer keeps roots in Taiwan. Central News Agency, Oct. 9,
2010.
http://focustaiwan.tw/ShowNews/WebNews_Detail.aspx?Type=aALL&ID=201010090008
Quote:
anthuriums are "not included in the 'early harvest' list 早期收穫清單 of
items that will enjoy preferential tariff treatment under the economic
cooperation framework agreement (ECFA) signed with China in late June
"a large white anthurium. At about 15 centimeters in diameter, the flowers
have sold in Japan for more than NT$100 apiece.
"H... 阅读全帖 |
|
m***y
发帖数: 14763 | 47 Right, and it's way too far-fetched if LZ wanna blame it on foundation. |
|
m***y
发帖数: 14763 | 48 OK, there is no account for tastes. So, I'll try not to judgmental.
Let me put it this way. LZ's current landscaping is popular, can fetch more
money on the market. If they go ahead with the plan, customizing it to their
"needs", well, it doesn't add values, but runs a higher risk to ruin it.
And your savings on grocery won't be enough to make up. |
|
l*****o
发帖数: 26631 | 49 我当然和她一起玩啦， 每天十五分钟fetch， 三十分钟出去遛。
说的玩具特指让她自己玩打发时间的，是小时候那些各种chew的玩具，装零食的益智玩
具，现在她都不怎么玩了。 |
|
k*******e
发帖数: 11147 | 50 我刚才还想着发帖子问你呢
假草坪需要吸尘么？怎么清理的？我考虑在泳池边上铺一道，PU玩fetch的必经之路。。 |
|
