Обсуждение: Is OpenSSL AES-NI not available in pgcrypto?

Hi all,

A question, may I wrong.

I've a Rocky Linux 8 with OpenSSL 1.1.1 FIPS  and Intel cpu with aes 
support (cat /proc/cpuinfo | grep aes)

Test made with openssl gives me a huge performance with aes enabled vs not:

"openssl speed -elapsed -evp aes-128-cbc" is about 5 time faster than 
"openssl speed -elapsed aes-128-cbc" or another "software calculated 
test", eg. "openssl speed -elapsed bf-cbc"

So OpenSSL is ok.

Postgresql 15 is compiled with openssl:

select name, setting from pg_settings where name = 'ssl_library';
     name     | setting
-------------+---------
  ssl_library | OpenSSL
(1 row)

So, a test with pgcrypto:

select pgp_sym_encrypt(data::text, 'pwd') --default to aes128
from generate_series('2022-01-01'::timestamp, '2022-12-31'::timestamp, 
'1 hour'::interval) data

vs

select pgp_sym_encrypt(data::text, 'pwd','cipher-algo=bf') -- blowfish
from generate_series('2022-01-01'::timestamp, '2022-12-31'::timestamp, 
'1 hour'::interval) data

In my test both queries execution is similar....aes-128 was expected 
about  5 time faster.

So, why?

Pgcrypto use OpenSSL as backend, so, does it explicit force software aes 
calculation instead of AES-NI cpu ones?

Thanksfor support.

Best regards,

Agharta

On 02.01.23 17:57, agharta82@gmail.com wrote:
> select pgp_sym_encrypt(data::text, 'pwd') --default to aes128
> from generate_series('2022-01-01'::timestamp, '2022-12-31'::timestamp, 
> '1 hour'::interval) data
> 
> vs
> 
> select pgp_sym_encrypt(data::text, 'pwd','cipher-algo=bf') -- blowfish
> from generate_series('2022-01-01'::timestamp, '2022-12-31'::timestamp, 
> '1 hour'::interval) data
> 
> In my test both queries execution is similar....aes-128 was expected 
> about  5 time faster.
> 
> So, why?
> 
> Pgcrypto use OpenSSL as backend, so, does it explicit force software aes 
> calculation instead of AES-NI cpu ones?

I suspect it is actually using AES hardware support, but all the other 
overhead of pgcrypto makes the difference not noticeable.

Hi,

I see, I was hoping that wasn't the case.

Thanks a lot for your support.

My best regards,

Agharta


Il 03/01/23 16:54, Peter Eisentraut ha scritto:
> On 02.01.23 17:57, agharta82@gmail.com wrote:
>> select pgp_sym_encrypt(data::text, 'pwd') --default to aes128
>> from generate_series('2022-01-01'::timestamp, 
>> '2022-12-31'::timestamp, '1 hour'::interval) data
>>
>> vs
>>
>> select pgp_sym_encrypt(data::text, 'pwd','cipher-algo=bf') -- blowfish
>> from generate_series('2022-01-01'::timestamp, 
>> '2022-12-31'::timestamp, '1 hour'::interval) data
>>
>> In my test both queries execution is similar....aes-128 was expected 
>> about  5 time faster.
>>
>> So, why?
>>
>> Pgcrypto use OpenSSL as backend, so, does it explicit force software 
>> aes calculation instead of AES-NI cpu ones?
>
> I suspect it is actually using AES hardware support, but all the other 
> overhead of pgcrypto makes the difference not noticeable.
>

On Mon, Jan  2, 2023 at 05:57:38PM +0100, agharta82@gmail.com wrote:
> So, a test with pgcrypto:
> 
> select pgp_sym_encrypt(data::text, 'pwd') --default to aes128
> from generate_series('2022-01-01'::timestamp, '2022-12-31'::timestamp, '1
> hour'::interval) data
> 
> vs
> 
> select pgp_sym_encrypt(data::text, 'pwd','cipher-algo=bf') -- blowfish
> from generate_series('2022-01-01'::timestamp, '2022-12-31'::timestamp, '1
> hour'::interval) data

To see the difference, I think you need to construct a single large
query that calls many pgcrypto functions, with a small return result, so
the network, parsing, and optimizer overhead are minimal compared to the
OpenSSL overhread.

-- 
  Bruce Momjian  <bruce@momjian.us>        https://momjian.us
  EDB                                      https://enterprisedb.com

Embrace your flaws.  They make you human, rather than perfect,
which you will never be.