#101 2021-01-10 13:18:45

ab
Administrator
From: France
Registered: 2010-06-21
Posts: 13,268
Website

Re: Revision 2.x of the framework

Just to improve reading, and keep the lines small.
If you have property + read + write on the same line, it is sometimes either too long, or unreadable.

Offline

#102 2021-01-11 08:10:40

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

I understand, formatting the code is a matter of preference. Just adding my opinion I think, this is a good way for long properties, but for simple ones, and they are a lot, it makes the code somehow more complicated and scary. For example, TSynConnectionDefinition is a simple class, but with this new format, it looks like it has something tricky going on. Maybe, I am not used to it, but I thought it is good to share it with you.

Offline

#103 2021-01-11 10:25:33

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

What is the preferred version of Lazarus and FPC to work with version 2? Also can you please add Lazarus packages?

Offline

#104 2021-01-11 11:59:49

ab
Administrator
From: France
Registered: 2010-06-21
Posts: 13,268
Website

Re: Revision 2.x of the framework

About the formatting, we rather sticked to a single formatting option everywhere.
Changing depending on the context is what we did with previous mORMot code, and it ended up to be difficult to maintain and read.

We currently use the same version as with mORMot 1.18.
That is, FPC 3.2 and Lazarus 2.10.

Offline

#105 2021-01-11 12:05:36

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

Thanks for the clarification about formatting.
I've got some problem with Trunk so I was asking the exact version.
There is no Lazarus 2.10 branch that I could find, do you mean 2.0.10 or 2.1.0 (Trunk)? If it is the second one, as it is the trunk can you give me the exact revision so I can debug?
BTW I am using FPCUPdelux.

Offline

#106 2021-01-11 12:07:01

ab
Administrator
From: France
Registered: 2010-06-21
Posts: 13,268
Website

Re: Revision 2.x of the framework

I meant 2.0.10 sorry.

It is the current FPC + Lazarus stable version, in fact.
Trunk is too unstable for any serious work, unless you validate yourself a very specific SVN revision.

Offline

#107 2021-01-11 12:21:17

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

Just tested, it works fine, Thank you.
Now I can try my test package. Although it has some problems with mormot.core.text being recompiled and lead to checksum changes.

Offline

#108 2021-01-12 07:09:24

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

I was trying to compare them and find out that V2 is slower in some cases.
Simple and transactional inserts in the SQLite database are the same, but it is slower when arrays come to play when using batch or memory databases.  I tried a bigger number (1M) of records for the memory engine, and the results are the same.
Both tested with stable Lazarus on Win64, on release mode, and ran independently.
Can you please give them a check that if my results are valid?

V1 code:

program project1;

uses
  SysUtils,
  SynCommons,
  mORMot,
  SynSQLite3,
  mORMotSQLite3,
  SynSQLite3Static;

type

  { TTest }

  TTest = class(TSQLRecord)
  private
    FName: RawUtf8;
  published
    property Name: RawUtf8 read FName write FName;
  end;

var
  Server: TSQLRest;
  Test: TTest;

  procedure TestServer;
  var
    T: ILocalPrecisionTimer;
    I, C: Integer;
    Batch: TSQLRestBatch;
  begin
    T := TLocalPrecisionTimer.Create;
    C := 2000;
    //One
    T.Start;
    for I := 1 to C do
      Server.Add(Test, True);
    WriteLn('One: ', T.Stop, ', ', T.ByCount(C), ', ', T.PerSec(C));

    //Transaction
    T.Start;
    Server.TransactionBegin(TTest, 1);
    for I := 1 to C do
      Server.Add(Test, True);
    Server.Commit(1);
    WriteLn('Transaction: ', T.Stop, ', ', T.ByCount(C), ', ', T.PerSec(C));

    //Batch
    T.Start;
    Batch := TSQLRestBatch.Create(Server, TTest);
    for I := 1 to C do
      Batch.Add(Test, True);
    Server.BatchSend(Batch);
    Batch.Free;
    WriteLn('Batch: ', T.Stop, ', ', T.ByCount(C), ', ', T.PerSec(C));

    //Transaction & Batch
    T.Start;
    Server.TransactionBegin(TTest, 1);
    Batch := TSQLRestBatch.Create(Server, TTest);
    for I := 1 to C do
      Batch.Add(Test, True);
    Server.BatchSend(Batch);
    Batch.Free;
    Server.Commit(1);
    WriteLn('Transaction & Batch: ', T.Stop, ', ', T.ByCount(C), ', ', T.PerSec(C));

  end;

var
  Servers: array of TSQLRest;
  LM: TSQLLockingMode;
  Sy: TSQLSynchronousMode;
begin
  Servers := nil;
  Insert(TSQLRestServerDB.CreateWithOwnModel([TTest], ChangeFileExt(ParamStr(0), '.db')), Servers, Length(Servers));
  Insert(TSQLRestServerDB.CreateWithOwnModel([TTest], SQLITE_MEMORY_DATABASE_NAME), Servers, Length(Servers));
  Insert(TSQLRestStorageInMemory.Create(TTest, nil), Servers, Length(Servers));
  Test := TTest.Create;
  Test.Name := 'Name';

  for Server in Servers do
  begin
    WriteLn(Server.ClassName);
    if Server is TSQLRestServerDB then
      with TSQLRestServerDB(Server) do
      begin
        CreateMissingTables;
        for LM in [lmNormal, lmExclusive] do
          for Sy in [smOff, smNormal, smFull] do
          begin
            DB.LockingMode := LM;
            DB.Synchronous := Sy;
            WriteLn(GetEnumName(TypeInfo(TSQLLockingMode), Integer(LM))^, ', ', GetEnumName(TypeInfo(TSQLSynchronousMode), Integer(Sy))^);
            TestServer;
            WriteLn;
          end;
      end
    else
      TestServer;
    Server.Free;
  end;

  Test.Free;
  ReadLn;
end.

V2 Code:

program project1;

uses
  SysUtils,
  mormot.core.base,
  mormot.core.perf,
  mormot.core.rtti,
  mormot.orm.core,
  mormot.orm.sql,
  mormot.orm.sqlite3,
  mormot.rest.core,
  mormot.rest.memserver,
  mormot.rest.sqlite3,
  mormot.db.raw.sqlite3,
  mormot.db.raw.sqlite3.static;

type

  { TTest }

  TTest = class(TSQLRecord)
  private
    FName: RawUtf8;
  published
    property Name: RawUtf8 read FName write FName;
  end;

var
  Server: TSQLRest;
  Test: TTest;

  procedure TestServer;
  var
    T: ILocalPrecisionTimer;
    I, C: Integer;
    Batch: TSQLRestBatch;
  begin
    T := TLocalPrecisionTimer.Create;
    C := 2000;
    //One
    T.Start;
    for I := 1 to C do
      Server.Add(Test, True);
    WriteLn('One: ', T.Stop, ', ', T.ByCount(C), ', ', T.PerSec(C));

    //Transaction
    T.Start;
    Server.TransactionBegin(TTest, 1);
    for I := 1 to C do
      Server.Add(Test, True);
    Server.Commit(1);
    WriteLn('Transaction: ', T.Stop, ', ', T.ByCount(C), ', ', T.PerSec(C));

    //Batch
    T.Start;
    Batch := TSQLRestBatch.Create(Server.Orm, TTest);
    for I := 1 to C do
      Batch.Add(Test, True);
    Server.BatchSend(Batch);
    Batch.Free;
    WriteLn('Batch: ', T.Stop, ', ', T.ByCount(C), ', ', T.PerSec(C));

    //Transaction & Batch
    T.Start;
    Server.TransactionBegin(TTest, 1);
    Batch := TSQLRestBatch.Create(Server.Orm, TTest);
    for I := 1 to C do
      Batch.Add(Test, True);
    Server.BatchSend(Batch);
    Batch.Free;
    Server.Commit(1);
    WriteLn('Transaction & Batch: ', T.Stop, ', ', T.ByCount(C), ', ', T.PerSec(C));

  end;

var
  Servers: array of TSQLRest;
  LM: TSQLLockingMode;
  Sy: TSQLSynchronousMode;
begin
  Servers := nil;
  Insert(TSQLRestServerDB.CreateWithOwnModel([TTest], ChangeFileExt(ParamStr(0), '.db')), Servers, Length(Servers));
  Insert(TSQLRestServerDB.CreateWithOwnModel([TTest], SQLITE_MEMORY_DATABASE_NAME), Servers, Length(Servers));
  Insert(TRestServerFullMemory.CreateWithOwnModel([TTest]), Servers, Length(Servers));
  Test := TTest.Create;
  Test.Name := 'Name';

  for Server in Servers do
  begin
    WriteLn(Server.ClassName);
    if Server is TSQLRestServerDB then
      with TSQLRestServerDB(Server) do
      begin
        CreateMissingTables;
        for LM in [lmNormal, lmExclusive] do
          for Sy in [smOff, smNormal, smFull] do
          begin
            DB.LockingMode := LM;
            DB.Synchronous := Sy;
            WriteLn(GetEnumName(TypeInfo(TSQLLockingMode), Integer(LM))^, ', ', GetEnumName(TypeInfo(TSQLSynchronousMode), Integer(Sy))^);
            TestServer;
            WriteLn;
          end;
      end
    else
      TestServer;
    Server.Free;
  end;

  Test.Free;
  ReadLn;
end.

V1 Results:

TSQLRestServerDB
lmNormal, smOff
One: 590.84ms, 295us, 3384
Transaction: 6.98ms, 3us, 286450
Batch: 5.82ms, 2us, 343288
Transaction & Batch: 3.44ms, 1us, 580551

lmNormal, smNormal
One: 5.77s, 2.88ms, 346
Transaction: 9.57ms, 4us, 208986
Batch: 26.32ms, 13us, 75973
Transaction & Batch: 5.72ms, 2us, 349101

lmNormal, smFull
One: 8.86s, 4.43ms, 225
Transaction: 10.74ms, 5us, 186098
Batch: 31.87ms, 15us, 62749
Transaction & Batch: 7.08ms, 3us, 282366

lmExclusive, smOff
One: 35.09ms, 17us, 56991
Transaction: 6.74ms, 3us, 296559
Batch: 3.24ms, 1us, 616712
Transaction & Batch: 3.07ms, 1us, 650829

lmExclusive, smNormal
One: 6.42s, 3.21ms, 311
Transaction: 10.60ms, 5us, 188643
Batch: 28.30ms, 14us, 70671
Transaction & Batch: 6.10ms, 3us, 327385

lmExclusive, smFull
One: 8.36s, 4.18ms, 239
Transaction: 11.20ms, 5us, 178491
Batch: 37.02ms, 18us, 54014
Transaction & Batch: 7.88ms, 3us, 253742

TSQLRestServerDB
lmNormal, smOff
One: 9.11ms, 4us, 219466
Transaction: 6.57ms, 3us, 304043
Batch: 3.12ms, 1us, 639181
Transaction & Batch: 2.96ms, 1us, 673854

lmNormal, smNormal
One: 9.41ms, 4us, 212336
Transaction: 6.66ms, 3us, 300165
Batch: 2.99ms, 1us, 668002
Transaction & Batch: 2.98ms, 1us, 670690

lmNormal, smFull
One: 9.55ms, 4us, 209314
Transaction: 6.68ms, 3us, 299132
Batch: 3.01ms, 1us, 664451
Transaction & Batch: 3.01ms, 1us, 663790

lmExclusive, smOff
One: 9.57ms, 4us, 208877
Transaction: 6.70ms, 3us, 298107
Batch: 3.02ms, 1us, 661157
Transaction & Batch: 3ms, 1us, 666000

lmExclusive, smNormal
One: 9.49ms, 4us, 210548
Transaction: 6.69ms, 3us, 298641
Batch: 3.01ms, 1us, 663349
Transaction & Batch: 3ms, 1us, 664893

lmExclusive, smFull
One: 9.47ms, 4us, 211104
Transaction: 6.56ms, 3us, 304878
Batch: 2.93ms, 1us, 682128
Transaction & Batch: 3.05ms, 1us, 655093

TSQLRestStorageInMemory
One: 1.54ms, 0us, 1296176
Transaction: 1.51ms, 0us, 1321877
Batch: 145us, 0us, 13793103
Transaction & Batch: 101us, 0us, 19801980

V2 Results:

TSqlRestServerDB
lmNormal, smOff
One: 592.18ms, 296us, 3377
Transaction: 6.87ms, 3us, 291078
Batch: 6.65ms, 3us, 300480
Transaction & Batch: 4.06ms, 2us, 492004

lmNormal, smNormal
One: 5.93s, 2.96ms, 336
Transaction: 9.13ms, 4us, 218842
Batch: 23.47ms, 11us, 85215
Transaction & Batch: 6.18ms, 3us, 323153

lmNormal, smFull
One: 8.67s, 4.33ms, 230
Transaction: 10.70ms, 5us, 186776
Batch: 33.59ms, 16us, 59530
Transaction & Batch: 7.96ms, 3us, 251067

lmExclusive, smOff
One: 35.45ms, 17us, 56404
Transaction: 6.71ms, 3us, 297885
Batch: 3.92ms, 1us, 509813
Transaction & Batch: 3.73ms, 1us, 536049

lmExclusive, smNormal
One: 6.67s, 3.33ms, 299
Transaction: 9.88ms, 4us, 202347
Batch: 26.87ms, 13us, 74410
Transaction & Batch: 6.62ms, 3us, 301841

lmExclusive, smFull
One: 8.46s, 4.23ms, 236
Transaction: 10.68ms, 5us, 187143
Batch: 34.14ms, 17us, 58575
Transaction & Batch: 7.91ms, 3us, 252684

TSqlRestServerDB
lmNormal, smOff
One: 9.20ms, 4us, 217296
Transaction: 6.42ms, 3us, 311332
Batch: 3.79ms, 1us, 526592
Transaction & Batch: 3.58ms, 1us, 558035

lmNormal, smNormal
One: 9.15ms, 4us, 218483
Transaction: 6.48ms, 3us, 308641
Batch: 3.61ms, 1us, 553556
Transaction & Batch: 3.59ms, 1us, 556173

lmNormal, smFull
One: 9.35ms, 4us, 213766
Transaction: 6.48ms, 3us, 308261
Batch: 3.65ms, 1us, 547495
Transaction & Batch: 3.62ms, 1us, 551724

lmExclusive, smOff
One: 9.29ms, 4us, 215215
Transaction: 6.52ms, 3us, 306560
Batch: 3.64ms, 1us, 548546
Transaction & Batch: 3.61ms, 1us, 552638

lmExclusive, smNormal
One: 9.35ms, 4us, 213857
Transaction: 6.53ms, 3us, 306278
Batch: 3.62ms, 1us, 551419
Transaction & Batch: 3.61ms, 1us, 553709

lmExclusive, smFull
One: 9.37ms, 4us, 213265
Transaction: 6.49ms, 3us, 307787
Batch: 3.62ms, 1us, 551571
Transaction & Batch: 3.63ms, 1us, 550964

TRestServerFullMemory
One: 1.46ms, 0us, 1363326
Transaction: 1.51ms, 0us, 1319261
Batch: 1.78ms, 0us, 1119194
Transaction & Batch: 1.79ms, 0us, 1115448

Last edited by okoba (2021-01-12 08:01:50)

Offline

#109 2021-01-12 13:54:37

ab
Administrator
From: France
Registered: 2010-06-21
Posts: 13,268
Website

Re: Revision 2.x of the framework

I guess your timing difference is in the error range of your computer.
If you use a laptop CPU for instance, then you should be aware that timing is not to be trusted due to energy saving and turbo mode.

SQLite3 process is exactly the same, so performance should be identical.
TSQLRestStorageInMemory numbers are pretty weird for mORMot 1. I guess there is something wrong and nothing is actually inserted.

Here are some numbers about mORMot 1.18: https://gist.github.com/synopse/005068d … dae1c3da96
And the the tests on mORMot 2: https://gist.github.com/synopse/670ee98 … 002e877d3c
Some of the tests are not the same - but if the assertions count is similar, the tests are likely to be similar.

I run the mORMot 2 tests twice, and you can see that the numbers differ on the very same machine.
We may expect 10% difference from run to run. Especially if you have a browser and some VM running in the background (as I do).

Offline

#110 2021-01-12 14:16:29

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

I am trying to discover V2 in attempting to match your samples in the blog and documents. That is the reason I can not make sure my code is correct. Maybe I am doing something wrong. I am running that exact code I shared before, maybe you can run them.
Also, testing in a PC with an i9 with configurated on full performance and trying the test at least five times before sharing with you. Both projects are in Release (-o3 and no debug) mode, and the V1 package is on -o3 optimization and no debug info. And of course, I am using the latest version pulled again now to be sure.
Here is the result for running only the Memory servers with much more (2M) inserts.

V1:

One: 2.03s, 1us, 982155
Transaction: 2.17s, 1us, 918703
Batch: 2.68s, 1us, 744059
Transaction & Batch: 2.58s, 1us, 773406

V2:

One: 2.01s, 1us, 993000
Transaction: 2.14s, 1us, 930859
Batch: 5.12s, 2us, 390090
Transaction & Batch: 2.86s, 1us, 698332

As you can see the big difference is in the Batch mode.
I checked the ram usage too, the V1 uses 550MB but V2 uses 1100MB.

Also please not that in my previous tests, even SQLite tests in V2 and Batch are slower, it is less than Memory engine difference as I am inserting 2K records.

Last edited by okoba (2021-01-12 14:19:56)

Offline

#111 2021-01-12 14:24:14

ab
Administrator
From: France
Registered: 2010-06-21
Posts: 13,268
Website

Re: Revision 2.x of the framework

OK.
I looked at the code.

First problem.
You should not call TransactionBegin manually before the batch process. Transactions are done within the batch process itself automatically.
Note that transactions are not implemented with the memory servers.

Second problem.
You compare TSQLRestStorageInMemory and TRestServerFullMemory which are not the same.
TRestServerFullMemory has a full REST process included.
So you should try to test with TSQLRestServerFullMemory instead.

Offline

#112 2021-01-12 15:37:55

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

Thanks for the note about problem one. I did not know that and thought that because of posts Like https://blog.synopse.info/?post/2013/06 … cking-mode ("Batch Trans" column)
If I should not call TransactionBegin manually, why it speeds up the inserts in the "Transaction & Batch" test?

For the second problem, yes, that was the problem, and now both speeds are the same as TRestServerFullMemory is 2X slower.
Is TSQLRestStorageInMemory removed form V2? I could not find any note about it in the V2 code.

One question is still unresolved, why "Batch" test is slower in V2 even in TSQLRestServerDB?

Thanks.

Offline

#113 2021-01-12 16:29:38

ab
Administrator
From: France
Registered: 2010-06-21
Posts: 13,268
Website

Re: Revision 2.x of the framework

It is called TRestStorageInMemory and implemented in mormot.orm.storage.pas.

Batch should not be slower in V2. This part of the code has been untouched.
Have you an idea where it may come from?

Offline

#114 2021-01-12 21:28:45

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

I think you may like to add a TSQLRestStorageInMemory as a compatibility for V1 like the other classes.

Maybe you missed my question so I will ask again:
If I should not call TransactionBegin manually, why it speeds up the inserts in the "Transaction & Batch" test?

For the slowness, all the codes were the same, so it made me to look into the Batch code.
Everything look the same, although I recognizes changes in the JSON handling but anything I found had a sign of improvement in performance not slowness (Great!).
After a while I found out that this line looks weird:

MethodTable := PosChar(Method, '@');

I double checked it, and it seems disabling inline for PosChar, will fix the issue. At least in the test I sent because MethodTable will always be nil.
As of the reason, I do not know why exactly but for V1, FPC will warn that the function will not be inlined, but it will not warn for the V2, and it seems it tried to inline it but can not or made it worse. I couldn't investigate Asm code as the EngineBatchSend function is huge.

Offline

#115 2021-01-13 09:26:35

ab
Administrator
From: France
Registered: 2010-06-21
Posts: 13,268
Website

Re: Revision 2.x of the framework

PosChar() is exactly the same with v1 and v2.

Did you try with -O2 or -O1 ?

Offline

#116 2021-01-13 09:28:15

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

-O3 (Slow optimization), This is the default for the Release mode of Lazarus/FPC.

Offline

#117 2021-01-13 09:28:50

ab
Administrator
From: France
Registered: 2010-06-21
Posts: 13,268
Website

Re: Revision 2.x of the framework

So try with -O1 and -O2.

Offline

#118 2021-01-13 09:31:26

ab
Administrator
From: France
Registered: 2010-06-21
Posts: 13,268
Website

Re: Revision 2.x of the framework

Also try the following version:

function PosChar(Str: PUtf8Char; Chr: AnsiChar): PUtf8Char;
begin
  result := nil;
  if Str <> nil then
  begin
    repeat
      if Str^ = #0 then
        exit
      else if Str^ = Chr then
        break;
      inc(Str);
    until false;
    result := Str;
  end;
end;

Offline

#119 2021-01-13 09:47:18

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

Insert PerSec results for 100K, Batch, TSQLRestServerDB and SQLITE_MEMORY_DATABASE_NAME, lmNormal, smOff:
V1:
Old Pos(Inline):
-O3: 662405

Old Pos(Not Inline):
-O3: 663574


New Pos(Inline):
-O3: 541996

New Pos(Not Inline):
-O3: 670591


V2:
Old Pos(Inline):
-O1: 517357
-O2: 533991
-O3: 535920

Old Pos(Not Inline):
-O1: 632255
-O2: 651177
-O3: 652622


New Pos(Inline):
-O1: 520540
-O2: 531053
-O3: 532257

New Pos(Not Inline):
-O1: 631524
-O2: 646981
-O3: 650474

Last edited by okoba (2021-01-13 09:47:25)

Offline

#120 2021-01-13 16:06:45

ab
Administrator
From: France
Registered: 2010-06-21
Posts: 13,268
Website

Re: Revision 2.x of the framework

It is pretty weird that PosChar() is the bottleneck.

I have just reimplemented PosChar() using branchless SSE2 asm on x86_64.
Check https://github.com/synopse/mORMot2/comm … 7404552a1a
Hope it helps.

Offline

#121 2021-01-13 16:54:22

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

Checked it and I can confirm that it is working fast now. On more than 10 tests, it was 95% speed of the V1 in the Batch test for TSQLRestServerDB and 120% for TSQLRestServerFullMemory.
I tried the two Pascal versions and they will be slower in the context of V2.
Thanks for the update.

Offline

#122 2021-01-13 17:12:14

ab
Administrator
From: France
Registered: 2010-06-21
Posts: 13,268
Website

Re: Revision 2.x of the framework

It was very weird, but I am happy we are in similar speed now.

If you have any more feedback, don't hesitate!
smile

Offline

#123 2021-01-13 17:18:58

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

With pleasure.

Offline

#124 2021-01-13 19:34:48

macfly
Member
From: Brasil
Registered: 2016-08-20
Posts: 374

Re: Revision 2.x of the framework

Thanks @okoba and @ab for these tests and improvements.

@okoba any particular reason for using the "aliases" for classes in V2?
Example TSQLRest instead of TRest.

Offline

#125 2021-01-14 05:16:36

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

I wanted to both code look alike so I can make sure that I'm testing the same code. For future work I will use V2.

Offline

#126 2021-01-14 10:41:09

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

I was a little bothered about why V2 is not faster, as it seems you optimized many parts, including JSON parts.
Investigating more, and find out that the new use of overloaded IdemPCharArray(Method, 'POPUDESI') will slow down the code.
I replaced it with the old IdemPCharArray(Method,['POST','PUT','DELETE','SIMPLE']), and it speeded up the "Batch" test more than 10% for TSQLRestServerDB and 25% for TSQLRestServerFullMemory.
For more info, doing old IdemPCharArray for 10M time takes 60ms, compared to 3000ms for V2.
If I wanted to go for a more aggressive path, I would use PosExChar(Method[1],'OUEI') or PosChar('OUEI',Method[1])^ as they are 2X faster than old IdemPCharArray, but in the context of "Batch" test, it will only improve near 5%.

Offline

#127 2021-01-14 14:30:20

ab
Administrator
From: France
Registered: 2010-06-21
Posts: 13,268
Website

Re: Revision 2.x of the framework

There is something wrong with your tests.
IdemPCharArray(Method, 'POPUDESI') can't possibly slow down the code: it compiles inline into a few lines of asm with no loop.

Offline

#128 2021-01-14 14:47:26

ab
Administrator
From: France
Registered: 2010-06-21
Posts: 13,268
Website

Re: Revision 2.x of the framework

Quick test:
https://gist.github.com/synopse/fd2265c … 83bdb49b3d

1.54ms 6493506
383us 26109660

The new variant with IdemPCharArray('DELETE', 'POPUDEUP') is more than 40 times faster...
So I don't understand where your speed difference comes from.

Offline

#129 2021-01-14 14:48:50

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

Yes, in an isolated test. But In the context of EngineBatchSend function it acts slower for me. I tried this code:

  //case IdemPCharArray(Method, 'POPUDESI') of
        case IdemPCharArray(Method,['POST','PUT','DELETE','SIMPLE']) of       

Just replacing this line will increase speed.
But in an isolated test, like below, new IdemPCharArray acts 3X faster as you say.

 program project1;

uses
  SysUtils,
  mormot.core.base,
  mormot.core.perf,
  mormot.core.unicode;

var
  T: ILocalPrecisionTimer;
  I, C: integer;
  MethodValue: RawUtf8;
  Method: PUtf8Char;
begin
  T := TLocalPrecisionTimer.Create;
  C := 1000000;
  MethodValue := 'SIMPLE';
  Method := Pointer(MethodValue);
  T.Start;
  for I := 1 to C do
    IdemPCharArray(Method, 'POPUDESI');
  WriteLn('1: ', T.Stop, ', ', T.ByCount(C), ', ', T.PerSec(C));
  T.Start;
  for I := 1 to C do
    IdemPCharArray(Method, ['POST', 'PUT', 'DELETE', 'SIMPLE']);
  WriteLn('2: ', T.Stop, ', ', T.ByCount(C), ', ', T.PerSec(C));
  for I := 1 to C do
    PosChar('OUEI', Method[1]);
  WriteLn('3: ', T.Stop, ', ', T.ByCount(C), ', ', T.PerSec(C));
  ReadLn;
end.                                

To be sure, I am using Lazarus Stable in a Win64 i9 machine.

Offline

#130 2021-01-14 14:51:18

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

My results for your exact test:

585us 17094017
192us 52083333

Offline

#131 2021-01-14 14:54:44

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

BTW I am confirming with another FPC too, fixes branch (3.2.1).

Offline

#132 2021-01-14 17:20:25

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

Because of our recent tests, I was thinking about how to have better comparison and benchmarking changes, and as D. Richard Hipp likes it for many things, including Fossil; I wanted to log the Tests in a DB. This way, I can recognize a leap/drop of speed and have a better way to query additional info about progress, and also, it is more fun to test and log code this way, and more future opportunities.
I started from your mormot2tests, first I wanted to inherit it, but it is tangled with the console log, so as a test, I did public some protected fields and made a test that stores a simple log.
Full implementation should use an improved TSynTest and more analytics, but as you said before, you like code more than talk, so here is sloppy code to showcase the idea.
Please let me know what you think.
https://gist.github.com/OkobaPatino/7db … a1d422572a

Last edited by okoba (2021-01-15 09:12:16)

Offline

#133 2021-01-15 07:44:23

ab
Administrator
From: France
Registered: 2010-06-21
Posts: 13,268
Website

Re: Revision 2.x of the framework

Please follow the forum rules, and don't post huge piece of code in the threads.
Use gist as I did.

About such testing, we usually rather use real application logs for performance enhancements, not tests.
Regression tests tend to be not realistic, whereas logs on actual production servers gives good information.
There is already extensive statistics in both the ORM and SOA part of mORMot - with database-accessible calls stats for interface-based services.

Offline

#134 2021-01-15 09:16:01

okoba
Member
Registered: 2019-09-29
Posts: 86

Re: Revision 2.x of the framework

Done.
What you explained is necessary for real life tasks, but difficult to find changes in small methods and tests.
The idea was, mORMot has a big test suit, maybe storing it in a DB gives a useful development analytics tool.
Thanks for the feedback.

Offline

#135 2021-01-18 19:22:36

PBa
Member
From: Austria
Registered: 2017-01-04
Posts: 18

Re: Revision 2.x of the framework

PBa wrote:

@ab:
in mORMot2 survey you also asked how to continue with source code management and there were many answers to also keep fossil as scm (at least in parallel to github). Just for curiosity: do you intend to set up also a fossil repository for mORMot2 in future?
Thanks a lot and have a nice evening!
Paul

Last edited by PBa (2021-01-18 19:23:21)

Offline

#136 2021-01-19 09:38:06

ab
Administrator
From: France
Registered: 2010-06-21
Posts: 13,268
Website

Re: Revision 2.x of the framework

Yes, I think I would make a fossil repository for mORMot2 in the future.

Offline

Board footer

Powered by FluxBB