SATA vs AHCI: why the massive performance difference?

[Dariusz Piatkowski]

So I needed to replicate my ThunderBird profile (separate thread on that topic elsewhere) to troubleshoot the Calendar/Google authentication issue and was taken aback by how long it took my backup profile to be copied over.

Did that a couple more times just to see that the results are consistent, and sure enough they are.

Take a look at the attached pics:

1) AHCI configuration (for SDD system) shows a nearly 29mins duration for a 9GB copy
2) SATA (DANIS506) (for the same SDD system) shows a 3 mins duration for the same 9GB copy

This is super strange, and even stranger so given that I do not recall seeing this issue before, albeit I may have updated the AHCI drivers since (haven't tried rolling them back to the previus version and re-testing this).

Is anyone else seeing this kind of an outcome???

System Config
==========
1) PROFILE - 9GB of content, about 5500 files that make up Thunderbird profile.
2) FS - JFS, with 1Gig cache
3) IFS => IFS=G:\OS2\JFS.IFS /CACHE:1048576 /LW:32,128,8 /AUTOCHECK:*
4) CAHEJFS => CALL=G:\OS2\CMD.EXE /Q /C G:\OS2\CACHEJFS.EXE /LW:32,128,8 /MINBUFFER:8000 /MAXBUFFER:24000 >NUL
5) DANIS506 - 1.08.16
6) AHCI - 2.09
7) MSI 990FXA-GD80 => details below

Code: [Select]

Bus 0 (PCI Express), Device Number 17, Device Function 0
Vendor 1002h Advanced Micro Devices, Inc. [AMD/ATI]
Device 4390h SB7x0/SB8x0/SB9x0 SATA Controller [IDE mode]
Command 0007h (I/O Access, Memory Access, BusMaster)
Status 0230h (Has Capabilities List, Supports 66MHz, Medium Timing)
Revision 40h, Header Type 00h, Bus Latency Timer 40h
Self test 00h (Self test not supported)
PCI Class Storage, type IDE (ATA)
PCI EIDE Controller Features :
  BusMaster EIDE is supported
  Primary   Channel is in native mode at Addresses 0 & 1
  Secondary Channel is in native mode at Addresses 2 & 3
Subsystem ID 76401462h Unknown
Subsystem Vendor 1462h Micro-Star International Co., Ltd. [MSI]
Address 0 is an I/O Port : F090h..F097h
Address 1 is an I/O Port : F080h..F083h
Address 2 is an I/O Port : F070h..F077h
Address 3 is an I/O Port : F060h..F063h
Address 4 is an I/O Port : F050h..F057h
Address 5 is a Memory Address (0-4GiB) : FEB0B000h..FEB0B3FFh
System IRQ 19, INT# A
New Capabilities List Present:
  Unknown Capability (Code 12h)!!
  Unknown Capability (Code 13h)!!


[Roderick Klein]

Ask Arca Noae, David has done a lot of research and made very good WIKI article how to install ArcaOS under Virtualbox.

https://www.arcanoae.com/wiki/arcaos/installation-planning/virtual-machine-configuration/oracle-vm-virtualbox/
"In Storage, the defaults are fine. Note that an IDE controller is used, and note that the VDI is on IDE Primary Device 0, and the DVD is on IDE Secondary Device 0. This is the recommended configuration as the DVD should not be on the same controller (Primary vs Secondary) as the VDI. Add additional controllers and disks as necessary. The IDE controller is the fastest emulation. Other devices such as AHCI will work also but the emulation is slower."
It contains more usefull tips how to get the best performance out of Virtualbox.

Roderick

[Lars]

add /N /F switches to OS2AHCI.ADD. I understood that in particular when using JFS the /F switch will improve performance as it ignores any write-through setting of the FS in order to prevent lengthy disk operations.

[Dariusz Piatkowski]

Hello Lars,

add /N /F switches to OS2AHCI.ADD. I understood that in particular when using JFS the /F switch will improve performance as it ignores any write-through setting of the FS in order to prevent lengthy disk operations.

I'm on AHCI v2.09 (Jun_2023) which claims to do '/N' by default...however, I will give the '/F' option a go and see if that makes a difference.

Thank you!

[Dariusz Piatkowski]

Hello Roderick,

Ask Arca Noae, David has done a lot of research and made very good WIKI article how to install ArcaOS under Virtualbox....

I'm running on real hardware here, and in my case the MSI 990FXA-GD80 motherboard has two SATA controllers:

1) JMicron
- nothing is hooked up to this

2) ATI SB850 chipset
- everything is hooked up to this (when the BIOS is configured for AHCI mode)

As best as I can tell my motherboard does not support having software (driver) selectable SATA vs AHCI device mode, you set it to either one at BIOS level.

[Steven Levine]

If this system is under support, I would recommend opening a ticket a providing David with testlogs.  There are a large number of possible reasons for the speed difference.

A couple of comments on DANIS506.ADD vs OS2AHCI.ADD.  For a given chipset, the codepaths for DANIS506.ADD are shorter than for OS2ACHI.ADD so performance depends on the quality of the chipset's SATA implementation compared to the ACHI implementation.  The performance also depends on how well DANIS506 understands the specific SATA chipset.  Finally, as you found, you can really slow down DANIS506 with a bad choice of options.  Less is usually better.

In a VM, DANIS506 is going to out perform OS2AHCI because the code path is shorter and the hardware is emulated.

[Andi B.]

In any case I would check copying the same file from command line. This doesn't give you the nice progress bar and you can't interrupt the copy operation but I'm pretty sure David will only look at this when cli copy shows similar results.

LarsenCommander copy use big buffers which are dynamically adjusted in size. I didn't my best to optimize the algorithmen for different scenarios. But  only with my systems and my disks (various SSDs and HDDs). You should rule out Lcmd before opening a ticket.

Edit typo.

[Dariusz Piatkowski]

Hi Andi,

In any case I would check copying the same file from command line. This doesn't give you the nice progress bar and you can't interrupt the copy operation but I'm pretty sure David will only look at this when cli copy shows similar results.

LarsenCommander copy use big buffers which are dynamically adjusted in size. I didn't my best to optimize the algorithmen for different scenarios. But  only with my systems and my disks (various SSDs and HDDs). You should rule out Lcmd before opening a ticket.

That's a great point, which I believe I have an answer to (well, almost an "answer" short of the actual CLI test): I had originally done a copy of the profile through WPS...but oohhhh booooyyyy did that take a long time!!!

At which point in time I thought to myself: umm, nope, I've done this before and literally a couple of mins later it was done...so I then went to LCMD to see what's actually going on in terms of actual throughput.

Good call though, let me try that next.

EDIT
====
'/F' test completed, looks to have been the answer to the issue I had ran into, see attached results from the very same LCMD test.

[Lars]

Ok, good. So we can guestimate that other ADD drivers, including DANIS506.ADD/IBM1S506.ADD also ignore the write-through flag ...