Re: [ata] 0568e61225: stress-ng.copy-file.ops_per_sec -15.0% regression

[Damien Le Moal]

On 2022/08/09 8:16, David Laight wrote:
> ...
>>>> Without knowing what the device adapter is, hard to say where the problem is. I
>>>> suspect that with the patch applied, we may be ending up with a small default
>>>> max_sectors value, causing overhead due to more commands than necessary.
>>>>
>>>> Will check what I see with my test rig.
>>>
>>> As far as I can see, this patch should not make a difference unless the
>>> ATA shost driver is setting the max_sectors value unnecessarily low.
>>
>> That is my hunch too, hence my question about which host driver is being used
>> for this test... That is not apparent from the problem report.
> 
> No one's fallen over the old problem and managed to limit
> the number of sectors in a read to the number of sectors
> in (IIRC) a 'multi sector' read that uses a single DMA burst?
> 
> Was always a good way of killing disk performance.
> 
> IIRC the maximum number of sectors for an ATA disk transfer is 255.

That is for super old pata/ide devices. Modern SATA/AHCI disks can do a lot more
than that with NCQ/DMA. The default max_sectors_kb for AHCI connected devices is
1280 KB = 2560 sectors, max_segments = 168 and max_segment_size = 65536 B, which
gives a reliable command size of at least 168 * 4096 = 688128 B (1344 sectors)
with 4KB page size architectures.

> 
> 	David
> 
> -
> Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
> Registration No: 1397386 (Wales)


-- 
Damien Le Moal
Western Digital Research