2022-06-01 19:38:14

by Feng Zhou

Subject: Re: Re: [PATCH v4 2/2] selftest/bpf/benchs: Add bpf_map benchmark

在 2022/6/1 下午5:53, Alexei Starovoitov 写道:
> On Wed, Jun 1, 2022 at 10:42 AM Feng zhou <[email protected]> wrote:
>> +struct {
>> + __uint(type, BPF_MAP_TYPE_HASH);
>> + __type(key, u32);
>> + __type(value, u64);
>> + __uint(max_entries, MAX_ENTRIES);
>> +} hash_map_bench SEC(".maps");
>> +
>> +u64 __attribute__((__aligned__(256))) percpu_time[256];
> aligned 256 ?
> What is the point?

I didn't think too much about it here, just referenced it from
tools/testing/selftests/bpf/progs/bloom_filter_bench.c

>
>> +u64 nr_loops;
>> +
>> +static int loop_update_callback(__u32 index, u32 *key)
>> +{
>> + u64 init_val = 1;
>> +
>> + bpf_map_update_elem(&hash_map_bench, key, &init_val, BPF_ANY);
>> + return 0;
>> +}
>> +
>> +SEC("fentry/" SYS_PREFIX "sys_getpgid")
>> +int benchmark(void *ctx)
>> +{
>> + u32 key = bpf_get_prandom_u32() % MAX_ENTRIES + MAX_ENTRIES;
> What is the point of random ?
> just key = MAX_ENTRIES would be the same, no?
> or key = -1 ?

If all threads on different cpu trigger sys_getpgid and lookup the same
key, it will cause
"ret = htab_lock_bucket(htab, b, hash, &flags); "
the lock competition here is fierce, and unnecessary overhead is
introduced,
and I don't want it to interfere with the test.

>
>> + u32 cpu = bpf_get_smp_processor_id();
>> + u64 start_time = bpf_ktime_get_ns();
>> +
>> + bpf_loop(nr_loops, loop_update_callback, &key, 0);
>> + percpu_time[cpu & 255] = bpf_ktime_get_ns() - start_time;
>> + return 0;
>> +}
>> --
>> 2.20.1
>>

2022-06-01 20:13:47

by Alexei Starovoitov

Subject: Re: Re: [PATCH v4 2/2] selftest/bpf/benchs: Add bpf_map benchmark

On Wed, Jun 1, 2022 at 1:17 PM Feng Zhou <[email protected]> wrote:
>
> 在 2022/6/1 下午5:53, Alexei Starovoitov 写道:
> > On Wed, Jun 1, 2022 at 10:42 AM Feng zhou <[email protected]> wrote:
> >> +struct {
> >> + __uint(type, BPF_MAP_TYPE_HASH);
> >> + __type(key, u32);
> >> + __type(value, u64);
> >> + __uint(max_entries, MAX_ENTRIES);
> >> +} hash_map_bench SEC(".maps");
> >> +
> >> +u64 __attribute__((__aligned__(256))) percpu_time[256];
> > aligned 256 ?
> > What is the point?
>
> I didn't think too much about it here, just referenced it from
> tools/testing/selftests/bpf/progs/bloom_filter_bench.c
>
> >
> >> +u64 nr_loops;
> >> +
> >> +static int loop_update_callback(__u32 index, u32 *key)
> >> +{
> >> + u64 init_val = 1;
> >> +
> >> + bpf_map_update_elem(&hash_map_bench, key, &init_val, BPF_ANY);
> >> + return 0;
> >> +}
> >> +
> >> +SEC("fentry/" SYS_PREFIX "sys_getpgid")
> >> +int benchmark(void *ctx)
> >> +{
> >> + u32 key = bpf_get_prandom_u32() % MAX_ENTRIES + MAX_ENTRIES;
> > What is the point of random ?
> > just key = MAX_ENTRIES would be the same, no?
> > or key = -1 ?
>
> If all threads on different cpu trigger sys_getpgid and lookup the same
> key, it will cause
> "ret = htab_lock_bucket(htab, b, hash, &flags); "
> the lock competition here is fierce, and unnecessary overhead is
> introduced,
> and I don't want it to interfere with the test.

I see.
but using random leaves it to chance.
Use cpu+max_entries then?

2022-06-02 03:33:12

by Feng Zhou

Subject: Re: Re: Re: [PATCH v4 2/2] selftest/bpf/benchs: Add bpf_map benchmark

在 2022/6/1 下午7:37, Alexei Starovoitov 写道:
> On Wed, Jun 1, 2022 at 1:17 PM Feng Zhou <[email protected]> wrote:
>> 在 2022/6/1 下午5:53, Alexei Starovoitov 写道:
>>> On Wed, Jun 1, 2022 at 10:42 AM Feng zhou <[email protected]> wrote:
>>>> +struct {
>>>> + __uint(type, BPF_MAP_TYPE_HASH);
>>>> + __type(key, u32);
>>>> + __type(value, u64);
>>>> + __uint(max_entries, MAX_ENTRIES);
>>>> +} hash_map_bench SEC(".maps");
>>>> +
>>>> +u64 __attribute__((__aligned__(256))) percpu_time[256];
>>> aligned 256 ?
>>> What is the point?
>> I didn't think too much about it here, just referenced it from
>> tools/testing/selftests/bpf/progs/bloom_filter_bench.c
>>
>>>> +u64 nr_loops;
>>>> +
>>>> +static int loop_update_callback(__u32 index, u32 *key)
>>>> +{
>>>> + u64 init_val = 1;
>>>> +
>>>> + bpf_map_update_elem(&hash_map_bench, key, &init_val, BPF_ANY);
>>>> + return 0;
>>>> +}
>>>> +
>>>> +SEC("fentry/" SYS_PREFIX "sys_getpgid")
>>>> +int benchmark(void *ctx)
>>>> +{
>>>> + u32 key = bpf_get_prandom_u32() % MAX_ENTRIES + MAX_ENTRIES;
>>> What is the point of random ?
>>> just key = MAX_ENTRIES would be the same, no?
>>> or key = -1 ?
>> If all threads on different cpu trigger sys_getpgid and lookup the same
>> key, it will cause
>> "ret = htab_lock_bucket(htab, b, hash, &flags);"
>> the lock competition here is fierce, and unnecessary overhead is
>> introduced,
>> and I don't want it to interfere with the test.
> I see.
> but using random leaves it to chance.
> Use cpu+max_entries then?

Ok, will do. Thanks.