03.03.2016 19:46, Dimitry Sibiryakov wrote:
> 03.03.2016 17:38, Dmitry Yemanov wrote:
>> Pretty no difference for short hash strings, thanks.
>
> Your conclusion is...?
I don't see much value, but it looks like it does not hurt either. Feel
free to commit.
Dmitry
03.03.2016 17:38, Dmitry Yemanov wrote:
> Pretty no difference for short hash strings, thanks.
Your conslusion is...?
--
WBR, SD.
--
Site24x7 APM Insight: Get Deep Visibility into Application Performance
APM +
03.03.2016 17:25, Vlad Khorsun wrote:
>> Tests shown no drawback for longer data.
> Does you test the code without that two if's ?
Yes, I specially compared performance with and without them. Function with
these ifs
won on short (1-2 bytes) data and shown the same performance
03.03.2016 19:33, Dimitry Sibiryakov wrote:
> Original:
>> Elapsed time= 1.243 sec
vs
> CRC32:
>> Elapsed time= 1.238 sec
Pretty no difference for short hash strings, thanks.
Dmitry
--
Site24x7 APM Insight: Get Deep
03.03.2016 17:13, Dmitry Yemanov wrote:
> Can you perform the same test but with INT join key?
I made number of records in table i2 four times bigger to get rid of timer's
jitter,
but still...
Original:
> SQL> select count(*) from i1 join i2 on i1.a=i2.a;
>
> PLAN HASH (I2 NATURAL, I1
03.03.2016 16:15, Dimitry Sibiryakov wrote:
> 03.03.2016 15:07, Vlad Khorsun wrote:
Are you sure that lock manager doesn't work with data of one or two
bytes long? At
these sizes the function had a weak spot without these ifs.
>> Perfect. All other 99% cases (with key >=
03.03.2016 18:36, Dimitry Sibiryakov wrote:
> varchar(2000) filled with rpad('', 2000, uuid_to_char(gen_uuid())).
Can you perform the same test but with INT join key?
Dmitry
--
Site24x7 APM Insight: Get Deep
03.03.2016 18:49, Dimitry Sibiryakov wrote:
>> Out of curiosity, why do we need a separate crc32c.cpp?
>
> To let engine work on processors without SSE4.2 command set.
Ah, I missed processor-specific compilation flags for that module.
Thanks for clarifying.
Dmitry
03.03.2016 16:31, Dmitry Yemanov wrote:
> I had the same doubts about the first two if's, but
> Vlad already asked about them.
And as I already said: they improve performance with short data without
visible affect
for long data. At least two months old tests showed me that. If anyone's tests
03.03.2016 16:20, Vlad Khorsun wrote:
>> will you believe my numbers?..
> Why not ?
Exactly because of
> special case, but very far from real life, i'm afraid.
> I guess you have join on long strings.
Not much long: varchar(2000) filled with rpad('', 2000,
03.03.2016 14:07, Dimitry Sibiryakov wrote:
> Ok, here you have it, attached.
Much better, thanks. I had the same doubts about the first two if's, but
Vlad already asked about them.
Out of curiosity, why do we need a separate crc32c.cpp? Cannot the
CRC32C function be defined inside Hash.cpp?
03.03.2016 16:51, Dimitry Sibiryakov wrote:
> 03.03.2016 15:07, Vlad Khorsun wrote:
>>> I wouldn't expect much from lock manager, but hash joins can get
>>> significant boost,
especially on long keys.
>> You don't need "server with dozen of cores" to check it, isn't is ?
>
>
03.03.2016 15:07, Vlad Khorsun wrote:
>> I wouldn't expect much from lock manager, but hash joins can get
>> significant boost,
>> >especially on long keys.
> You don't need "server with dozen of cores" to check it, isn't is ?
You are right, this part I can test even with my
03.03.2016 15:07, Vlad Khorsun wrote:
>> > Are you sure that lock manager doesn't work with data of one or two
>> > bytes long? At
>> >these sizes the function had a weak spot without these ifs.
> Perfect. All other 99% cases (with key >= 4 bytes) must pay for nothing.
Tests shown no
03.03.2016 15:32, Dimitry Sibiryakov wrote:
> 03.03.2016 12:56, Vlad Khorsun wrote:
>>> If you mean overall profit under multiuser load, then - no, I have no
>>> numbers for that.
>>
>> Perfect
>
> If you have at hand server with dozen of cores and test suite you are
> welcome to
03.03.2016 6:28, Dmitry Yemanov wrote:
I've asked for a particular change patch, you've provided 220KB of the
total tree diff to scan for the needed changes. If you want your patches
reviewed sooner, better start to respect other's time.
Ok, here you have it, attached.
--
WBR, SD.
03.03.2016 00:41, Dimitry Sibiryakov wrote:
> I did. Look into ICQ logs.
> http://www.ibphoenix.com/ibpr_devel/Avalerion.63058.7z
> contain every changed source file in obedience to IDPL.
I've asked for a particular change patch, you've provided 220KB of the
total tree diff to scan for the
02.03.2016 19:22, Dmitry Yemanov wrote:
> When did the project politics change to allow not approved patches?
When participation in development of the project had been limited to chosen
ones only?..
--
WBR, SD.
Dimitry,
> Revision: 63072
>http://sourceforge.net/p/firebird/code/63072
> Author: aafemt
> Date: 2016-03-02 15:33:16 + (Wed, 02 Mar 2016)
> Log Message:
> ---
> Savepoints refactoring that fixes CORE-4701, CORE-4424 and CORE-4483
When did the project politics
19 matches
Mail list logo
