hi

i have read PDR 7111

they talk about CFG (Control Flow Guard) but i'm not sure about it.
i have made a small App to put 1000 Controls on a Dialog.

it take about 2.50 Sec on my FX-8350 (4 GHz) with GT1030 on Windows 10 64/32 
Bit
as i have 4 x SSD with diferent OS i try Windows 8.1 64 Bit and it use 1.01 
Sec for same Code.
last Test was on Windows 7 32 Bit which take 0.66 Sec.

so Windows 10 ist 250% slower that 8.1 and 400% to Windows 7 !

i have try to disable CFG Exploit but that make no Diffence in Windows 10 64 
Bit while 32 Bit Version was a bit quicker 2.30 Sec.
than i try to disable ALL Exploit in Offline-Mode but that help not much 
still about 2 Sec.

so what else make Windows 10 GDI32 so slow ?




---
Diese E-Mail wurde von AVG auf Viren gepruft.
http://www.avg.com

more Test

have compiled Source with harbour using GTWVG and it need 0.41Sec. under
Windows 10 32/64Bit

on Windows v8.1 it need 0.22 Sec which is about same as VFP with 1000
Textbox need 0.14 Sec.

so what is the real Problem with XbParts under different Windows OS ?





---
Diese E-Mail wurde von AVG auf Viren geprüft.
http://www.avg.com

On 3/14/2019 9:21 AM, Jim Lee wrote:
> more Test
> 
> have compiled Source with harbour using GTWVG and it need 0.41Sec. under
> Windows 10 32/64Bit
> 
> on Windows v8.1 it need 0.22 Sec which is about same as VFP with 1000
> Textbox need 0.14 Sec.
> 
> so what is the real Problem with XbParts under different Windows OS ?
> 
> 
> 
> 
> 
> ---
> Diese E-Mail wurde von AVG auf Viren geprüft.
> http://www.avg.com
> 
Hi,

For what it's worth I compiler and ran your test app with Xbase++ 2.0 
version 1079 on my laptop with an Intel Core i7-6600U @ 2.80GHz and 
Intel(R) HD Graphics 520 Display adapter running Win 10 64 ver 1607.

Timing was between .93 - .94 seconds.

Regards... Jonathan

jonathan.leeming@the-family-centre.com
Edmonton, Alberta, Canada

hi,


> For what it's worth I compiler and ran your test app with Xbase++ 2.0
> version 1079 on my laptop with an Intel Core i7-6600U @ 2.80GHz and
> Intel(R) HD Graphics 520 Display adapter running Win 10 64 ver 1607.
>
> Timing was between .93 - .94 seconds.

as i know Problem begin with 1703 so if you upgrade to 1809 (or 1904 in some
week) same Test will slow down.
this is for ALL Apps using GDI32 so also my harbour App is slow but still
400% quicker than XbParts.

see also next Msg about XbParts when using > 4GB "for all Apps and OS"



---
Diese E-Mail wurde von AVG auf Viren geprüft.
http://www.avg.com

On 3/14/2019 10:44 PM, Jim Lee wrote:
> hi,
> 
> 
>> For what it's worth I compiler and ran your test app with Xbase++ 2.0
>> version 1079 on my laptop with an Intel Core i7-6600U @ 2.80GHz and
>> Intel(R) HD Graphics 520 Display adapter running Win 10 64 ver 1607.
>>
>> Timing was between .93 - .94 seconds.
> 
> as i know Problem begin with 1703 so if you upgrade to 1809 (or 1904 in some
> week) same Test will slow down.
> this is for ALL Apps using GDI32 so also my harbour App is slow but still
> 400% quicker than XbParts.
> 
> see also next Msg about XbParts when using > 4GB "for all Apps and OS"
> 
> 
> 
> ---
> Diese E-Mail wurde von AVG auf Viren geprüft.
> http://www.avg.com
> 
Hi Jim,

I plan to take a spare computer with just Windows and your SLE1000 
installed and try it with progressively newer versions of windows 10 
starting with 1511, then 1607, then 1703 and finally 1809 to compare 
timings on the exact same machine.

If you have any additional suggestions please let me know.

Regards... Jonathan

jonathan.leeming@the-family-centre.com
Edmonton, Alberta, Canada

On 3/15/2019 12:04 PM, Jonathan Leeming wrote:
> On 3/14/2019 10:44 PM, Jim Lee wrote:
>> hi,
>>
>>
>>> For what it's worth I compiler and ran your test app with Xbase++ 2.0
>>> version 1079 on my laptop with an Intel Core i7-6600U @ 2.80GHz and
>>> Intel(R) HD Graphics 520 Display adapter running Win 10 64 ver 1607.
>>>
>>> Timing was between .93 - .94 seconds.
>>
>> as i know Problem begin with 1703 so if you upgrade to 1809 (or 1904 
>> in some
>> week) same Test will slow down.
>> this is for ALL Apps using GDI32 so also my harbour App is slow but still
>> 400% quicker than XbParts.
>>
>> see also next Msg about XbParts when using > 4GB "for all Apps and OS"
>>
>>
>>
>> ---
>> Diese E-Mail wurde von AVG auf Viren geprüft.
>> http://www.avg.com
>>
> Hi Jim,
> 
> I plan to take a spare computer with just Windows and your SLE1000 
> installed and try it with progressively newer versions of windows 10 
> starting with 1511, then 1607, then 1703 and finally 1809 to compare 
> timings on the exact same machine.
> 
> If you have any additional suggestions please let me know.
> 
> Regards... Jonathan
> 
OK...

  I took a Windows 10 64 bit machine and started off with version 1511 
and did multiple runs of the SLE1000 test utility Jim Lee posted and 
found the following timings (in seconds):

WinVer		Build		SLE1000	Time	Xbase version
Version 1511	OS 10586.1540	.90 - 1.0	2.0.1079
Version 1607 	OS 14393.0	.81-.94		2.0.1079
Version 1703	OS 15063.0	.86-.91		2.0.1079
Version 1709	OS 16299.15	.76-.84		2.0.1079
Version 1803	OS 17134.1	1.01-1.08	2.0.1079
Version 1803	OS 17134.1	1.02-1.08	2.0.1087
Version 1809	OS 17763.107	1.76-1.85	2.0.1087

Showing that something has changed on the Windows side that appears to 
be causing the test program to take at least twice as long in Windows 
1803 & 1809 compared to earlier versions.

The machine I used for all tests was...

AsusTek motherboard with bus clock @ 100 MHz
Processor: Intel Core i5-6400 (4 core) at 2.7 GHz w/6M cache
RAM: 8192 MB
Drive: Kingston 120 GB SSD (SATA Controller)

For the version 1809 test I disabled Windows Defender.

Aside from the standard windows "stuff" nothing else was installed on 
the workstation... just the latest drivers.

The main code from Jim's test program (SLE1000) is...

PROCEDURE MAIN
LOCAL nEvent, mp1, mp2, oXbp
LOCAL oDlg1, i, nStart, nStop

   oDlg1         := XbpDialog():new(AppDesktop() ,, {0,0}, {1024,768} )
   oDlg1:title   := "1000 SLE Test"
   oDlg1:taskList := .T.
   oDlg1:create()
   oDlg1:Hide()
   CenterControl(oDlg1)

   nStart := SECONDS()
   FOR i := 1 TO 1000
      oXbp := XbpSLE():new(oDlg1:DrawingArea, ,{i,i/4*3}, {100,30})
      oXbp:create()
   NEXT
   nStop := SECONDS()
   oDlg1:Show()
   Msgbox("Zeit "+STR(nStop-nStart)+" Sec.")

   oDlg1:Destroy()
RETURN

Regards... Jonathan

jonathan.leeming@the-family-centre.com
Edmonton, Alberta, Canada

Jonathan,

>   I took a Windows 10 64 bit machine and started off with version 1511 
> and did multiple runs of the SLE1000 test utility Jim Lee posted and 
> found the following timings (in seconds):
> 
> WinVer		Build		SLE1000	Time	Xbase version
> Version 1511	OS 10586.1540	.90 - 1.0	2.0.1079
> Version 1607 	OS 14393.0	.81-.94		2.0.1079
> Version 1703	OS 15063.0	.86-.91		2.0.1079
> Version 1709	OS 16299.15	.76-.84		2.0.1079
> Version 1803	OS 17134.1	1.01-1.08	2.0.1079
> Version 1803	OS 17134.1	1.02-1.08	2.0.1087
> Version 1809	OS 17763.107	1.76-1.85	2.0.1087
> 
> Showing that something has changed on the Windows side that appears to 
> be causing the test program to take at least twice as long in Windows 
> 1803 & 1809 compared to earlier versions.

Nice test.
How long did it take you to do this tests, installing a different version
of Windows on the machine each time?

Peter

Also you have to take into account all mitigations due to Meltdown and Spectre
exploits that are done on microcode and OS kernel level.

@Phoronix tests it was not or just a bit noticable in range of few percents,
and only for database (sql databases) tests and synthetic tests there was
measurable impact which can be seen at phoronix dot com.

If there was GDI redesign it should be found about it on MS technet ?



Peter Alderliesten wrote in message
news:1ts55kk4ivgi5$.1j40o1w04bton.dlg@40tude.net...
>Jonathan,
>
>>   I took a Windows 10 64 bit machine and started off with version 1511 
>> and did multiple runs of the SLE1000 test utility Jim Lee posted and 
>> found the following timings (in seconds):
>> 
>> WinVer		Build		SLE1000	Time	Xbase version
>> Version 1511	OS 10586.1540	.90 - 1.0	2.0.1079
>> Version 1607 	OS 14393.0	.81-.94		2.0.1079
>> Version 1703	OS 15063.0	.86-.91		2.0.1079
>> Version 1709	OS 16299.15	.76-.84		2.0.1079
>> Version 1803	OS 17134.1	1.01-1.08	2.0.1079
>> Version 1803	OS 17134.1	1.02-1.08	2.0.1087
>> Version 1809	OS 17763.107	1.76-1.85	2.0.1087
>> 
>> Showing that something has changed on the Windows side that appears to 
>> be causing the test program to take at least twice as long in Windows 
>> 1803 & 1809 compared to earlier versions.
>
>Nice test.
>How long did it take you to do this tests, installing a different version
>of Windows on the machine each time?
>
>Peter

On 4/12/2019 8:17 AM, Matej Jurac wrote:
> Also you have to take into account all mitigations due to Meltdown and Spectre
> exploits that are done on microcode and OS kernel level.
> 
> @Phoronix tests it was not or just a bit noticable in range of few percents,
> and only for database (sql databases) tests and synthetic tests there was
> measurable impact which can be seen at phoronix dot com.
> 
> If there was GDI redesign it should be found about it on MS technet ?
> 
> 
> 
> Peter Alderliesten wrote in message
> news:1ts55kk4ivgi5$.1j40o1w04bton.dlg@40tude.net...
>> Jonathan,
>>
>>>    I took a Windows 10 64 bit machine and started off with version 1511
>>> and did multiple runs of the SLE1000 test utility Jim Lee posted and
>>> found the following timings (in seconds):
>>>
>>> WinVer		Build		SLE1000	Time	Xbase version
>>> Version 1511	OS 10586.1540	.90 - 1.0	2.0.1079
>>> Version 1607 	OS 14393.0	.81-.94		2.0.1079
>>> Version 1703	OS 15063.0	.86-.91		2.0.1079
>>> Version 1709	OS 16299.15	.76-.84		2.0.1079
>>> Version 1803	OS 17134.1	1.01-1.08	2.0.1079
>>> Version 1803	OS 17134.1	1.02-1.08	2.0.1087
>>> Version 1809	OS 17763.107	1.76-1.85	2.0.1087
>>>
>>> Showing that something has changed on the Windows side that appears to
>>> be causing the test program to take at least twice as long in Windows
>>> 1803 & 1809 compared to earlier versions.
>>
>> Nice test.
>> How long did it take you to do this tests, installing a different version
>> of Windows on the machine each time?
>>
>> Peter

Hi Matej,

Following up on your comments I found that you are correct... but you 
probably knew that 

I disabled the Meltdown & Spectre protection on the test workstation 
running Win 10 Pro Version 1809 and found that the test results were in 
line with the older versions of Windows .79 - .85 seconds.

Regards... Jonathan

jonathan.leeming@the-family-centre.com
Edmonton, Alberta, Canada

Jonathan Leeming wrote in message
news:4d3e9a66$65465a33$34fce@news.alaska-software.com...
>On 4/12/2019 8:17 AM, Matej Jurac wrote:
>> Also you have to take into account all mitigations due to Meltdown and Spectre
>> exploits that are done on microcode and OS kernel level.
>> 
>> @Phoronix tests it was not or just a bit noticable in range of few percents,
>> and only for database (sql databases) tests and synthetic tests there was
>> measurable impact which can be seen at phoronix dot com.
>> 
>> If there was GDI redesign it should be found about it on MS technet ?
>> 
>> 
>> 
>> Peter Alderliesten wrote in message
>> news:1ts55kk4ivgi5$.1j40o1w04bton.dlg@40tude.net...
>>> Jonathan,
>>>
>>>>    I took a Windows 10 64 bit machine and started off with version 1511
>>>> and did multiple runs of the SLE1000 test utility Jim Lee posted and
>>>> found the following timings (in seconds):
>>>>
>>>> WinVer		Build		SLE1000	Time	Xbase version
>>>> Version 1511	OS 10586.1540	.90 - 1.0	2.0.1079
>>>> Version 1607 	OS 14393.0	.81-.94		2.0.1079
>>>> Version 1703	OS 15063.0	.86-.91		2.0.1079
>>>> Version 1709	OS 16299.15	.76-.84		2.0.1079
>>>> Version 1803	OS 17134.1	1.01-1.08	2.0.1079
>>>> Version 1803	OS 17134.1	1.02-1.08	2.0.1087
>>>> Version 1809	OS 17763.107	1.76-1.85	2.0.1087
>>>>
>>>> Showing that something has changed on the Windows side that appears to
>>>> be causing the test program to take at least twice as long in Windows
>>>> 1803 & 1809 compared to earlier versions.
>>>
>>> Nice test.
>>> How long did it take you to do this tests, installing a different version
>>> of Windows on the machine each time?
>>>
>>> Peter
>
>Hi Matej,
>
>Following up on your comments I found that you are correct... but you 
>probably knew that 
>
>I disabled the Meltdown & Spectre protection on the test workstation 
>running Win 10 Pro Version 1809 and found that the test results were in 
>line with the older versions of Windows .79 - .85 seconds.
>
>Regards... Jonathan

Hello Mr. Jonathan

You are not the first one that had same question (I think there was another
programmer here with same performance degradation between major releases). I
do some advising/consulting too and this performance degradation does come as
question.

On local workstation, behind router and firewall those mitigations are not
that mission critical, but are crucial on machines open to internet. 

Also there are differences between Intel and AMD (Ryzen) CPUs which have less
problems with those two vulneabilities and performance impact lesser.

Greetings, Matej

On 4/12/2019 3:37 AM, Peter Alderliesten wrote:
> Jonathan,
> 
>>    I took a Windows 10 64 bit machine and started off with version 1511
>> and did multiple runs of the SLE1000 test utility Jim Lee posted and
>> found the following timings (in seconds):
>>
>> WinVer		Build		SLE1000	Time	Xbase version
>> Version 1511	OS 10586.1540	.90 - 1.0	2.0.1079
>> Version 1607 	OS 14393.0	.81-.94		2.0.1079
>> Version 1703	OS 15063.0	.86-.91		2.0.1079
>> Version 1709	OS 16299.15	.76-.84		2.0.1079
>> Version 1803	OS 17134.1	1.01-1.08	2.0.1079
>> Version 1803	OS 17134.1	1.02-1.08	2.0.1087
>> Version 1809	OS 17763.107	1.76-1.85	2.0.1087
>>
>> Showing that something has changed on the Windows side that appears to
>> be causing the test program to take at least twice as long in Windows
>> 1803 & 1809 compared to earlier versions.
> 
> Nice test.
> How long did it take you to do this tests, installing a different version
> of Windows on the machine each time?
> 
> Peter
> 
Hi Peter,

Hope all is well with you!  Miss your SWXbase Sessions!!!

For the test each iteration took a little over 3 hours to setup...

Download Windows Image: 30 min
ISO Image creation on USB stick: 20 min
Windows Install: 20 min
Latest drivers down loads & install: 2 hrs

Then Xbase install < 2min
"Install" Jim's SLE1000 test and actual test runs: < 2 min

So all in, about 18 hours but most is unattended.

Each windows setup was done by an great IT support specialist we have on 
staff named Alfredo so all I had to do was the Xbase++ install and 
actual test runs.

Initially it took a bit of exploration to find a download source for 
each of the 4 GB windows downloads but Alfredo has a knack for finding 
these things... sometimes I just don't want to know!

Regards... Jonathan

jonathan.leeming@the-family-centre.com
Edmonton, Alberta, Canada

Jonathan,

>>>    I took a Windows 10 64 bit machine and started off with version 1511
>>> and did multiple runs of the SLE1000 test utility Jim Lee posted and
>>> found the following timings (in seconds):
>>>
>>> WinVer		Build		SLE1000	Time	Xbase version
>>> Version 1511	OS 10586.1540	.90 - 1.0	2.0.1079
>>> Version 1607 	OS 14393.0	.81-.94		2.0.1079
>>> Version 1703	OS 15063.0	.86-.91		2.0.1079
>>> Version 1709	OS 16299.15	.76-.84		2.0.1079
>>> Version 1803	OS 17134.1	1.01-1.08	2.0.1079
>>> Version 1803	OS 17134.1	1.02-1.08	2.0.1087
>>> Version 1809	OS 17763.107	1.76-1.85	2.0.1087
>>>
>>> Showing that something has changed on the Windows side that appears to
>>> be causing the test program to take at least twice as long in Windows
>>> 1803 & 1809 compared to earlier versions.
>> 
>> Nice test.
>> How long did it take you to do this tests, installing a different version
>> of Windows on the machine each time?
>> 
>> Peter
>> 
> Hi Peter,
> 
> Hope all is well with you!  Miss your SWXbase Sessions!!!

Actually I miss them too. Would have been nice to meet up again.
Also, giving a session is a good learning experience 
and there were enough sessions tot attend and learn from others as well.

> For the test each iteration took a little over 3 hours to setup...
> 
> Download Windows Image: 30 min
> ISO Image creation on USB stick: 20 min
> Windows Install: 20 min
> Latest drivers down loads & install: 2 hrs
> 
> Then Xbase install < 2min
> "Install" Jim's SLE1000 test and actual test runs: < 2 min
> 
> So all in, about 18 hours but most is unattended.

Wow, it's good to have such an inventive specialist around.
This would have been way to much work for me. 
And of course disruptive to my current projects.

Greetings,
Peter