[RocksDB perf] Cache speedup

Summary:
I have ran a get benchmark where all the data is in the cache and observed that most of the time is spent on waiting for lock in LRUCache.

This is an effort to optimize LRUCache.

Test Plan:
The data was loaded with fillseq. Then, I ran a benchmark:

    /db_bench --db=/tmp/rocksdb_stat_bench --num=1000000 --benchmarks=readrandom --statistics=1 --use_existing_db=1 --threads=16 --disable_seek_compaction=1 --cache_size=20000000000 --cache_numshardbits=8 --table_cache_numshardbits=8

I ran the benchmark three times. Here are the results:
AFTER THE PATCH: 798072, 803998, 811807
BEFORE THE PATCH: 782008, 815593, 763017

Reviewers: dhruba, haobo, kailiu

Reviewed By: haobo

CC: leveldb

Differential Revision: https://reviews.facebook.net/D14571
main
Igor Canadi 11 years ago
parent 5e4ab767cf
commit e8d40c31b3
  1. 24
      util/cache.cc

@ -10,7 +10,7 @@
#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include <list>
#include <vector>
#include "rocksdb/cache.h"
#include "port/port.h"
@ -111,8 +111,8 @@ class HandleTable {
}
void Resize() {
uint32_t new_length = 4;
while (new_length < elems_) {
uint32_t new_length = 16;
while (new_length < elems_ * 1.5) {
new_length *= 2;
}
LRUHandle** new_list = new LRUHandle*[new_length];
@ -255,18 +255,20 @@ Cache::Handle* LRUCache::Insert(
LRUHandle* e = reinterpret_cast<LRUHandle*>(
malloc(sizeof(LRUHandle)-1 + key.size()));
std::list<LRUHandle*> last_reference_list;
std::vector<LRUHandle*> last_reference_list;
last_reference_list.reserve(1);
e->value = value;
e->deleter = deleter;
e->charge = charge;
e->key_length = key.size();
e->hash = hash;
e->refs = 2; // One from LRUCache, one for the returned handle
memcpy(e->key_data, key.data(), key.size());
{
MutexLock l(&mutex_);
e->value = value;
e->deleter = deleter;
e->charge = charge;
e->key_length = key.size();
e->hash = hash;
e->refs = 2; // One from LRUCache, one for the returned handle
memcpy(e->key_data, key.data(), key.size());
LRU_Append(e);
LRUHandle* old = table_.Insert(e);

Loading…
Cancel
Save