Files
spark/src/server/sparkd/cmds/Benchmark.cpp
T
Pedro Pinto 8e00da5a72 SPARK-1653, The initial crash that john reported turned out to be a monster.
Events were getting re-ordered in our event queue, possibly due to
this:

https://github.com/cameron314/concurrentqueue/issues/126

We abandoned the moody camel queue.
2018-08-30 21:49:01 +00:00

704 lines
19 KiB
C++

#include "Benchmark.h"
#include "lib/api/services/OrderService.server.pb.h"
#include "lib/utils/PBUtils.h"
#include "lib/utils/Puid.h"
#include "lib/utils/SpinLock.h"
#include "lib/utils/Uuid.h"
#include "server/sparkd/cmds/CmdRegistry.h"
#include "lib/jetfix/jetfix.h"
#include "lib/mercury/broker/CoreEventBroker.h"
#include "lib/mercury/mercury_utils/FatWrapper.h"
#include <benchmark/benchmark.h>
#include <boost/container/flat_map.hpp>
#include <boost/container/small_vector.hpp>
#include <boost/lockfree/queue.hpp>
#include <boost/lockfree/stack.hpp>
#include <boost/pool/object_pool.hpp>
#include <boost/pool/pool.hpp>
#include <boost/pool/pool_alloc.hpp>
#include <boost/pool/singleton_pool.hpp>
#include <cstdlib>
#include <iostream>
#include <moodycamel/concurrentqueue.h>
#include <mutex>
#include <tbb/concurrent_queue.h>
namespace bts {
namespace cmd {
auto sec_str = " id: 3\n"
" symbol: \"ES\"\n"
" exchange: VENUE_GLBX\n"
" contract_spec {\n"
" low_tick: 0.25\n"
" high_tick: 0.25\n"
" cash_per_point: 50\n"
" price_multiplier: 0.01\n"
" }\n"
" future {\n"
" expiration_year: 2018\n"
" expiration_month: 9\n"
" expiration_day: 1\n"
" }\n"
" glbx {\n"
" exchange_id: 57287\n"
" security_desc: \"ESU8\"\n"
" cfi_code: \"FFIXSX\"\n"
" cme_symbol: \"ES\"\n"
" }\n";
auto order_str = " order_key: 189285\n"
" timestamp {\n"
" seconds: 1509644176\n"
" nanos: 717250000\n"
" }\n"
" native {\n"
" account_name: \"BTS \"\n"
" operator_id: \"BTS \"\n"
" security_key: 3\n"
" side: SIDE_BUY\n"
" time_in_force: TIF_DAY\n"
" manual_order_indicator: MOI_YES\n"
" order_type: OT_LIMIT\n"
" price: 2500\n"
" order_quantity: 12\n"
" }\n"
" customer_or_firm: COF_FIRM\n"
" cti_code: CTI_CTI1\n"
" order_status: OS_UNACKNOWLEDGED\n";
auto order_state_str = " status: ORDER_STATUS_COMPLETED\n"
" price: 65.56\n"
" quantity: 5\n"
" filled_quantity: 5\n"
" last_price: 65.56\n"
" avg_price: 65.56\n"
" last_change {\n"
" seconds: 1532982945\n"
" nanos: 378466000\n"
" }\n"
" limit {\n"
" native_state {\n"
" exchange_order_id: \"12345678901234567\"\n"
" native_status: NATIVE_ORDER_STATUS_FILLED\n"
" last_msg_seq_num: 868\n"
" routing_record {\n"
" session: \"sim\"\n"
" account: \"ca_ACC A\"\n"
" operator_id: \"op_id_bts\"\n"
" cti_code: CTI_CODE_CTI_CODE1\n"
" customer_or_firm: CUSTOMER_OR_FIRM_FIRM\n"
" location: \"us,nc\"\n"
" }\n"
" native_type: NATIVE_ORDER_TYPE_LIMIT\n"
" }\n"
" }\n"
" created {\n"
" seconds: 1532962902\n"
" nanos: 989007000\n"
" }\n";
static void gen_id(benchmark::State &state) {
std::string scope = "benchmark";
for (auto _ : state) {
auto id = utils::reserve_ids(scope, 1);
benchmark::DoNotOptimize(id + 1 > 0);
}
}
BENCHMARK(gen_id);
static void mutex_lock_unlock(benchmark::State &state) {
int i = 0;
std::mutex lock;
for (auto _ : state) {
std::lock_guard<std::mutex> l(lock);
benchmark::DoNotOptimize(i++);
}
}
BENCHMARK(mutex_lock_unlock);
static void spin_lock_unlock(benchmark::State &state) {
int i = 0;
std::atomic_flag _mutex = ATOMIC_FLAG_INIT;
for (auto _ : state) {
while (_mutex.test_and_set(std::memory_order_acquire))
;
benchmark::DoNotOptimize(i++);
_mutex.clear(std::memory_order_release);
}
}
BENCHMARK(spin_lock_unlock);
static void bts_spinlock_lock_unlock(benchmark::State &state) {
int i = 0;
utils::SpinLock l;
for (auto _ : state) {
std::lock_guard<utils::SpinLock> g(l);
benchmark::DoNotOptimize(i++);
}
}
BENCHMARK(bts_spinlock_lock_unlock);
// static void serialize_order(benchmark::State &state) {
// auto order_state =
// utils::pb::from_string<api::orderman::OrderState>(order_state_str);
// auto
// auto order_state_cpy = order_state;
// for (auto _ : state) {
// order_state_cpy = order_state;
// }
// }
// BENCHMARK(copy_order);
static void serialize_update(benchmark::State &state) {
api::orderman::Update update;
update.set_order_key(1);
update.set_price(1.2);
update.set_quantity(2);
std::string buffer;
buffer.resize(update.ByteSize());
bts_assert(update.SerializeToArray(&buffer[0], buffer.size()));
for (auto _ : state) {
update.SerializeWithCachedSizesToArray((unsigned char*) &buffer[0]);
}
}
BENCHMARK(serialize_update);
static void serialize_order(benchmark::State &state) {
auto order_state =
utils::pb::from_string<api::orderman::OrderState>(order_state_str);
std::string buffer;
buffer.resize(order_state.ByteSize());
bts_assert(order_state.SerializeToArray(&buffer[0], buffer.size()));
for (auto _ : state) {
order_state.SerializeWithCachedSizesToArray((unsigned char*) &buffer[0]);
}
}
BENCHMARK(serialize_order);
struct IntId {
IntId() { value = std::rand(); }
bool operator<(IntId const &v) const { return value < v.value; }
uint32_t value;
};
static void binary_search_of_guid(benchmark::State &state) {
using Id = utils::Uuid;
using Map = boost::container::flat_map<Id, int>;
auto sz = state.range(0);
Map m;
for (int i = 0; i < sz; i++)
m[Id()] = i;
auto target = Id();
int found_count = 0;
for (auto _ : state) {
auto it = m.find(target);
if (it != m.end())
found_count++;
benchmark::DoNotOptimize(found_count);
}
}
BENCHMARK(binary_search_of_guid)->Range(8, 8 << 10);
static void binary_search_of_int64(benchmark::State &state) {
using Id = IntId;
using Map = boost::container::flat_map<Id, int>;
auto sz = state.range(0);
Map m;
for (int i = 0; i < sz; i++)
m[Id()] = i;
auto target = Id();
int found_count = 0;
for (auto _ : state) {
auto it = m.find(target);
if (it != m.end())
found_count++;
benchmark::DoNotOptimize(found_count);
}
}
BENCHMARK(binary_search_of_int64)->Range(8, 8 << 10);
static void binary_search_of_int32(benchmark::State &state) {
using Id = IntId;
using Map = boost::container::flat_map<Id, int>;
auto sz = state.range(0);
Map m;
for (int i = 0; i < sz; i++)
m[Id()] = i;
auto target = Id();
int found_count = 0;
for (auto _ : state) {
auto it = m.find(target);
if (it != m.end())
found_count++;
benchmark::DoNotOptimize(found_count);
}
}
BENCHMARK(binary_search_of_int32)->Range(8, 8 << 10);
static void guid_gen(benchmark::State &state) {
for (auto _ : state) {
benchmark::DoNotOptimize(utils::Uuid());
}
}
BENCHMARK(guid_gen);
struct A {
int x;
};
static void shared_ptr_cpy(benchmark::State &state) {
auto a = std::make_shared<A>();
for (auto _ : state) {
std::shared_ptr<A> b;
benchmark::DoNotOptimize(b = a);
}
}
BENCHMARK(shared_ptr_cpy);
static void alloc_free(benchmark::State &state) {
for (auto _ : state) {
auto a = std::make_shared<A>();
benchmark::DoNotOptimize(a.get());
}
}
BENCHMARK(alloc_free);
static void new_delete(benchmark::State &state) {
for (auto _ : state) {
auto a = new A();
delete a;
benchmark::DoNotOptimize(a);
}
}
BENCHMARK(new_delete);
static void boost_pool(benchmark::State &state) {
boost::pool<> p(sizeof(A));
for (auto _ : state) {
auto a = p.malloc();
p.free(a);
benchmark::DoNotOptimize(a);
}
}
BENCHMARK(boost_pool);
static void boost_object_pool(benchmark::State &state) {
boost::object_pool<A> p;
for (auto _ : state) {
auto a = p.malloc();
p.free(a);
benchmark::DoNotOptimize(a);
}
}
BENCHMARK(boost_object_pool);
static void protected_boost_pool(benchmark::State &state) {
boost::pool<> p(sizeof(A));
std::atomic_flag _mutex = ATOMIC_FLAG_INIT;
for (auto _ : state) {
while (_mutex.test_and_set(std::memory_order_acquire))
;
auto a = p.malloc();
_mutex.clear(std::memory_order_release);
while (_mutex.test_and_set(std::memory_order_acquire))
;
p.free(a);
_mutex.clear(std::memory_order_release);
benchmark::DoNotOptimize(a);
}
}
BENCHMARK(protected_boost_pool);
static void boost_singleton_pool(benchmark::State &state) {
using Pool = boost::singleton_pool<struct Single, sizeof(A)>;
for (auto _ : state) {
auto a = Pool::malloc();
Pool::free(a);
benchmark::DoNotOptimize(a);
}
}
BENCHMARK(boost_singleton_pool);
static void atomic_inc(benchmark::State &state) {
std::atomic<size_t> i(0);
for (auto _ : state) {
benchmark::DoNotOptimize(i++);
}
}
BENCHMARK(atomic_inc);
static void mem_copy(benchmark::State &state) {
char *src = new char[state.range(0)];
char *dst = new char[state.range(0)];
memset(src, 'x', state.range(0));
for (auto _ : state)
memcpy(dst, src, state.range(0));
state.SetBytesProcessed(int64_t(state.iterations()) *
int64_t(state.range(0)));
delete[] src;
delete[] dst;
}
BENCHMARK(mem_copy)->Arg(8)->Arg(64)->Arg(128)->Arg(512)->Arg(1 << 10)->Arg(
8 << 10);
static void alloc_and_copy_from_pool(benchmark::State &state) {
boost::pool<> p(sizeof(A));
for (auto _ : state) {
auto a = new (p.malloc()) A(); // NOLINT
a->~A(); // NOLINT
p.free(a);
benchmark::DoNotOptimize(a);
}
}
BENCHMARK(alloc_and_copy_from_pool);
static void lambda_copy(benchmark::State &state) {
auto p = std::make_shared<A>();
std::function<int()> f = [p]() { return 1; };
std::array<std::function<int()>, 1> buffer;
for (auto _ : state) {
benchmark::DoNotOptimize(buffer[0] = f);
}
}
BENCHMARK(lambda_copy);
static void boost_pool_allocator(benchmark::State &state) {
boost::fast_pool_allocator<A> p;
for (auto _ : state) {
auto ptr = std::allocate_shared<A>(p);
benchmark::DoNotOptimize(ptr.get());
}
}
BENCHMARK(boost_pool_allocator);
int Benchmark::execute(std::vector<std::string> const &args) {
bts_assert(args.size() < 99);
int argc = args.size() + 1;
char *argv[100];
argv[0] = (char *)"spark";
for (size_t i = 0; i < args.size(); i++)
argv[i + 1] = (char *)args[i].c_str();
benchmark::Initialize(&argc, argv);
benchmark::RunSpecifiedBenchmarks();
return 0;
}
struct Notify {
mercury::EventBrokerID id;
mercury::SubID suid;
mercury::StreamID sid;
utils::FatWrapper<80> payload;
};
static void queue_notification(benchmark::State &state) {
std::array<utils::FatWrapper<128>, 1> queue;
for (auto _ : state) {
auto &n = queue[0].wrap_in_place<Notify>();
n.id = mercury::EventBrokerID(1);
n.suid = mercury::SubID(1);
n.sid = mercury::StreamID(1);
auto &a = n.payload.wrap_in_place<A>();
a.x = 1;
benchmark::DoNotOptimize(queue[0].empty());
}
}
BENCHMARK(queue_notification);
static void queue_notification2(benchmark::State &state) {
std::array<utils::FatWrapper<128>, 2> queue;
auto &n = queue[0].wrap_in_place<Notify>();
n.id = mercury::EventBrokerID(1);
n.suid = mercury::SubID(1);
n.sid = mercury::StreamID(1);
auto &a = n.payload.wrap_in_place<A>();
a.x = 1;
for (auto _ : state) {
queue[1] = queue[0];
benchmark::DoNotOptimize(queue[1].empty());
}
}
BENCHMARK(queue_notification2);
static void boost_lockless_queue(benchmark::State &state) {
using FreeList = boost::lockfree::queue<void *>;
FreeList queue(1000000);
for (auto _ : state) {
void *result;
queue.push(nullptr);
benchmark::DoNotOptimize(queue.pop(result));
}
}
BENCHMARK(boost_lockless_queue);
static void tbb_concurrent_queue(benchmark::State &state) {
using FreeList = tbb::concurrent_queue<void *>;
FreeList queue;
for (auto _ : state) {
void *result;
queue.push(nullptr);
benchmark::DoNotOptimize(queue.try_pop(result));
}
}
BENCHMARK(tbb_concurrent_queue);
static void moody_camel_queue(benchmark::State &state) {
using Q = moodycamel::ConcurrentQueue<void *>;
Q queue(1000000);
for (auto _ : state) {
void *result;
void *arg = &queue;
queue.enqueue(arg); // NOLINT
benchmark::DoNotOptimize(queue.try_dequeue(result));
}
}
BENCHMARK(moody_camel_queue);
static void moody_camel_fat_push(benchmark::State &state) {
using Msg = utils::FatWrapper<80>;
using Q = moodycamel::ConcurrentQueue<Msg>;
Q queue;
Msg fw;
auto &n = fw.wrap_in_place<Notify>();
n.id = mercury::EventBrokerID(1);
n.suid = mercury::SubID(1);
n.sid = mercury::StreamID(1);
auto &a = n.payload.wrap_in_place<A>();
a.x = 1;
for (auto _ : state) {
queue.enqueue(fw);
Msg fw2;
benchmark::DoNotOptimize(queue.try_dequeue(fw2));
}
}
BENCHMARK(moody_camel_fat_push);
static void boost_lockless_stack(benchmark::State &state) {
using FreeList = boost::lockfree::stack<void *>;
FreeList stack;
for (auto _ : state) {
void *result;
stack.push(nullptr);
benchmark::DoNotOptimize(stack.pop(result));
}
}
BENCHMARK(boost_lockless_stack);
using Msg = utils::FatWrapper<80>;
struct PPQueue {
std::atomic_flag _mutex = ATOMIC_FLAG_INIT;
std::vector<Msg *> v;
void push(Msg *e) {
while (_mutex.test_and_set(std::memory_order_acquire))
;
v.push_back(e);
_mutex.clear(std::memory_order_release);
}
bool pop(Msg **e) {
auto result = false;
while (_mutex.test_and_set(std::memory_order_acquire))
;
if (v.size()) {
*e = v.back();
v.pop_back();
result = true;
}
_mutex.clear(std::memory_order_release);
return result;
}
};
static void moody_camel_slim_push(benchmark::State &state) {
using Q = moodycamel::ConcurrentQueue<Msg *>;
Q queue;
PPQueue store;
store.v.reserve(10);
Msg fw;
auto &n = fw.wrap_in_place<Notify>();
n.id = mercury::EventBrokerID(1);
n.suid = mercury::SubID(1);
n.sid = mercury::StreamID(1);
auto &a = n.payload.wrap_in_place<A>();
a.x = 1;
store.push(&fw);
for (auto _ : state) {
Msg *fw_ptr = nullptr;
store.pop(&fw_ptr);
auto &n = fw_ptr->wrap_in_place<Notify>();
n.id = mercury::EventBrokerID(1);
n.suid = mercury::SubID(1);
n.sid = mercury::StreamID(1);
auto &a = n.payload.wrap_in_place<A>();
a.x = 1;
queue.enqueue(fw_ptr);
queue.try_dequeue(fw_ptr);
store.push(fw_ptr);
benchmark::DoNotOptimize(fw_ptr);
}
}
BENCHMARK(moody_camel_slim_push);
using EventBrokerID = uint32_t;
using StreamID = uint32_t;
using SubID = uint32_t;
using TopicPtr = std::shared_ptr<int>;
using IEventBrokerPtr = TopicPtr;
using TopicFilter = std::function<void()>;
using InternalEventHandler = TopicFilter;
using InternalSubID = uint32_t;
struct NeighbourAddition {
NeighbourAddition() : ptr() {}
IEventBrokerPtr ptr;
};
struct NeighbourRemoval {
NeighbourRemoval() : id() {}
EventBrokerID id;
};
struct Advertisement {
Advertisement() : id(), sid(), topic() {}
EventBrokerID id;
StreamID sid;
TopicPtr topic;
};
struct Unadvertisement {
Unadvertisement() : id(), sid() {}
EventBrokerID id;
StreamID sid;
};
struct Subscription {
Subscription() : id(), suid(), sid() {}
EventBrokerID id;
SubID suid;
StreamID sid;
};
struct Unsubscription {
Unsubscription() : id(), suid(), sid() {}
EventBrokerID id;
SubID suid;
StreamID sid;
};
struct InternalSubscription {
InternalSubscription() : suid(), filter(), handler() {}
InternalSubID suid;
TopicFilter filter;
InternalEventHandler handler;
};
struct InternalUnsubscription {
InternalUnsubscription() : suid() {}
InternalSubID suid;
};
template <class T> static void lambda_alloc_copy(benchmark::State &state) {
utils::FatWrapper<128> fw;
using Action = std::function<int()>;
T t;
for (auto _ : state) {
benchmark::DoNotOptimize(
fw.wrap_in_place<Action>([]() { return sizeof(T); }));
}
}
BENCHMARK_TEMPLATE(lambda_alloc_copy, NeighbourAddition);
BENCHMARK_TEMPLATE(lambda_alloc_copy, NeighbourRemoval);
BENCHMARK_TEMPLATE(lambda_alloc_copy, Advertisement);
BENCHMARK_TEMPLATE(lambda_alloc_copy, Unadvertisement);
BENCHMARK_TEMPLATE(lambda_alloc_copy, Subscription);
BENCHMARK_TEMPLATE(lambda_alloc_copy, Unsubscription);
BENCHMARK_TEMPLATE(lambda_alloc_copy, InternalSubscription);
BENCHMARK_TEMPLATE(lambda_alloc_copy, InternalUnsubscription);
template <class T> static void simple_alloc_copy(benchmark::State &state) {
utils::FatWrapper<128> fw;
for (auto _ : state) {
benchmark::DoNotOptimize(fw.wrap_in_place<T>());
}
}
BENCHMARK_TEMPLATE(simple_alloc_copy, NeighbourAddition);
BENCHMARK_TEMPLATE(simple_alloc_copy, NeighbourRemoval);
BENCHMARK_TEMPLATE(simple_alloc_copy, Advertisement);
BENCHMARK_TEMPLATE(simple_alloc_copy, Unadvertisement);
BENCHMARK_TEMPLATE(simple_alloc_copy, Subscription);
BENCHMARK_TEMPLATE(simple_alloc_copy, Unsubscription);
BENCHMARK_TEMPLATE(simple_alloc_copy, InternalSubscription);
BENCHMARK_TEMPLATE(simple_alloc_copy, InternalUnsubscription);
static void get_thread_id(benchmark::State &state) {
for (auto _ : state) {
auto tid = std::this_thread::get_id();
benchmark::DoNotOptimize(tid == std::this_thread::get_id());
}
}
BENCHMARK(get_thread_id);
static void native_order_fix_data_create(benchmark::State &state) {
auto sz = state.range(0);
for (auto _ : state) {
auto p = new uint8_t[sz];
delete[] p;
benchmark::DoNotOptimize(p);
}
}
BENCHMARK(native_order_fix_data_create)->Range(8, 8 << 14);
static void native_order_fix_data_fragment(benchmark::State &state) {
for (auto _ : state) {
auto p = new jetfix::FIXFragment();
delete p;
benchmark::DoNotOptimize(p);
}
}
BENCHMARK(native_order_fix_data_fragment);
static void small_vec(benchmark::State &state) {
auto i = 0;
for (auto _ : state) {
boost::container::small_vector<size_t, 1> vec;
vec.emplace_back(i++);
benchmark::DoNotOptimize(vec[0]);
}
}
BENCHMARK(small_vec);
std::string Benchmark::help() {
return "Run micro-benchmarks. Use --benchmark_filter=regex to run only "
"specific benchmarks.";
}
} // namespace cmd
} // namespace bts