题目介绍
https://pdos.csail.mit.edu/6.828/2016/homework/lock.html
比较有价值的一个作业。重点考察了两个知识:
- 哈希表的实现
- 多线程与互斥锁
题目已给出基本的框架,大意是多核处理器下,使用多线程对共享数据(哈希表)进行操作,会带来效率的提升,然而如果不加锁就会出现不正确的结果。所以需要你在适当地方使用互斥锁。
直接上代码吧。为了方便理解做了一些变量名的更改与注释。
#include
#include
#include
#include
#include
#include
#define NBUCKET 5
#define NKEYS 100000
/*
* 实现了一个哈希表,对 NBUCKET 取余进行散列
*
*/
struct entry {
int key;
int value;
struct entry *next;
};
// table 中存放的是 5 个地址,链表头
struct entry *table[NBUCKET];
int keys[NKEYS];
// 线程数量默认为1
int nthread = 1;
// done 用于统计有多少个线程已经完成
volatile int done;
double now() {
struct timeval tv;
gettimeofday(&tv, 0);
return tv.tv_sec + tv.tv_usec / 1000000.0;
}
static void print(void) {
int i;
struct entry *e;
for (i=0; inext) {
printf("%d ", e->key);
}
printf("\n");
}
}
static void insert(int key, int val, struct entry **p, struct entry *n) {
// 在链表头插入
struct entry *e = (struct entry *)malloc(sizeof(struct entry));
e->key = key;
e->value = val;
e->next = n;
*p = e;
}
static void put(int key, int val) {
int i = key % NBUCKET;
insert(key, val, &table[i], table[i]);
}
static struct entry* get(int key) {
struct entry *e = 0;
for (e = table[key % NBUCKET]; e != 0; e = e->next) {
if (e->key == key) break;
}
return e;
}
static void *task(void *xa){
long n = (long) xa;
int b = NKEYS / nthread;
int no_found_count = 0;
double t0, t1;
t0 = now();
// 第 n 个线程put [n*b, (n+1)*b) 范围内的 entry
for (int i = 0; i < b; i++) {
put(keys[b*n + i], n);
}
t1 = now();
printf("thread %ld: put time = %f\n", n, t1-t0);
// 等待所有线程完成 put 操作
__sync_fetch_and_add(&done, 1);
while (done < nthread);
t0 = now();
for (int i=0; i
这个程序运行会出现问题,症状如下:
- 单线程
$ ./a.out 1
thread 0: put time = 0.004676
thread 0: get time = 11.021167
thread 0: 0 keys missing
completion time = 11.0260
- 多线程
$ ./a.out 2
thread 0: put time = 0.025475
thread 1: put time = 0.028295
thread 1: get time = 12.859959
thread 1: 15880 keys missing
thread 0: get time = 12.930442
thread 0: 15880 keys missing
completion time = 12.9592
显然,在几乎同样的时间里,多线程获得了效率提升(因为是多核cpu)。对整个哈希表查询了2轮,符合预期。然而,却出现了条目为空的情况。
分析
- 哪些是共享数据?
个人认为,本题中的共享数据,包括全局变量,以及在堆上的数据。 - 仅有写数据需要加锁,读数据并不需要。
在整个过程中,仅有 put 函数对全局变量做了更改。所以要加锁肯定加在 put 函数前。 - 为什么会丢失 key?
关键的 insert 函数中,如果线程 A 执行到e->next = n;
切换到线程B,线程B执行完成 insert 函数,返回线程 A 执行,则会出现连续改动两次链表头,相当于线程 B 的链表头直接被覆盖,造成丢失。
static void insert(int key, int val, struct entry **p, struct entry *n) {
// 在链表头插入
struct entry *e = (struct entry *)malloc(sizeof(struct entry));
e->key = key;
e->value = val;
e->next = n;
*p = e;
}
修改代码,加入全局变量:
pthread_mutex_t lock;
在 main 函数中初始化:
pthread_mutex_init(&lock, NULL);
将 task 函数修改为:
static void *task(void *xa){
long n = (long) xa;
int b = NKEYS / nthread;
int no_found_count = 0;
double t0, t1;
t0 = now();
// put 时加锁
pthread_mutex_lock(&lock);
for (int i = 0; i < b; i++) {
put(keys[b*n + i], n);
}
// put 结束时解锁
pthread_mutex_unlock(&lock);
t1 = now();
printf("thread %ld: put time = %f\n", n, t1-t0);
__sync_fetch_and_add(&done, 1);
while (done < nthread);
t0 = now();
for (int i=0; i
现在单线程执行结果为:
$ ./a.out 1
thread 0: put time = 0.004574
thread 0: get time = 10.766191
thread 0: 0 keys missing
completion time = 10.7710
多线程执行结果为:
./a.out 2
thread 0: put time = 0.002329
thread 1: put time = 0.004662
thread 1: get time = 10.282535
thread 1: 0 keys missing
thread 0: get time = 10.282641
thread 0: 0 keys missing
completion time = 10.2877