89 replies to this topic

[anudanan]

Why is a SoC driver closed source? I think it is linked to the linux kernel and then a part which must follow the GPL licence?

 

Or I´m wrong?

[WanWizard]

The entire SDK is closed source, and under heavy NDA from Broadcom. Only clients of Broadcom have access to it. Drivers are delivered in binary form only, no GPL in sight (and there is no requirement to do so).

[anudanan]

I use now the uno4kse with an external USB HDD and use the internal disk only for a movie store and not for recordings. Current my wife has started cutting and moving jobs on the movies stored on the internal hdd and I´ve see very often that the writing rate to the internal disk ist very poor. Very often the disk writes onyl 1-2 Mbyte/s during cutting and moving. in parallel the current three recording movies on the external drive writes constant 7-9 Mbyte/s.

 

In the situation where the writing rate is very poor on the internal disk the top tool shows very often I/O load from 50-90 percent. Very strange.

[MiLo]

I think the SATA driver is part of the regular Linux kernel, the closed Broadcom driver is only the DVB and Video parts.

[anudanan]

Some interessting information about the linux Kernel version on the openpli 6.1 with uno4kse  Until now I thought the Kernel is version 3.14, but today I´ve seen also openpli use like VTI the kernel 4.1.20. 

 

root@Schlafzimmer:/var/volatile/log# cat /proc/version

Linux version 4.1.20-1.9 (build@bouwserver2) (gcc version 6.2.0 (GCC) ) #2 SMP Tue Dec 19 01:32:18 CET 2017

root@Schlafzimmer:/var/volatile/log#

Edited by anudanan, 2 January 2018 - 10:57.

[WanWizard]

VU+ only updated the BSP recently, after our codebase was frozen in preparation of 6.1-release. So the kernel update is for the next release.

 

I sincerly doubt however that this will make any difference, if the kernel SATA driver was responsible for very slow throughputs, the entire Linux community would be kicking and screaming. This is not the case, sofar you are the only one complaining.

[anudanan]

I´ll test in one of the next days the 2,5 inch seagte drive with the poor performance issue in a housing to use it with USB. Then I can say very shure if the problem is the drive itself or the sata interface inside the uno4kse box.

 

But I think the reason is not inside the hdd drive

 

Do you now how often the SATA driver for the BCM7252S SoC is used by any oher boxes in the past?

 

For the VU+ boxes it is very new with these SoC. But I don´t know which other boxes used them in the past with SATA interface

Edited by anudanan, 2 January 2018 - 14:00.

[WanWizard]

Assuming the SoC exposes a standard hardware port, the driver used is the same on all boxes.

 

To test the raw throughput on the box, use

dd if=/dev/zero of=/media/<device>/test.img bs=512 count=1000

To test buffered reads:

hdparm -t /dev/<device>

To test cached reads:

hdparm -T /dev/<device>

[anudanan]

The raw performamce with dd is 50 Mbyte/s

 

buffered read performance with hdparm-t  is 125 Mbyte/s

 

Cached read with performance with hdparm -T is 800 Mbyte/s (with baracuda). With Firecuda (SHDD) it is  1000Mbyte/s

 

 

From my point of view these values are ok. My external 3,5 inch drive has nearly the same values

I´ve seen the write performance only is ok, the read performance only is ok, but parallel read and write at the same time ist poor (moviecut does it or move moving and recordings do it)

[WanWizard]

So it is unlikely to be a hardware and/or an OS issue.

[anudanan]

I´ve done some test with many parallel  dd jobs for reading, writing and copy of files to/from my internal HDD

 

The performance looks.ok

 

But what is the reason for the poor performance with the internal HDD when I use moviecut, moving files and recording files and why is it ok with the USB drive. In the moment I´m helpless and perplexed.

[WanWizard]

No errors in your logs? Did you run a disk check to rule out any physical issues with the disk?

 

You also might want to install smartmontools, and do a SMART check of your disk, just to be sure.

[anudanan]

Yes, I´ve done diskchecks, make new fileystems. And I´ve the problem with two different 2,5 inch drives. (seagate baracude and firecude both 2TB) in two different uno4kse boxes

I´m excited on the test result with these 2,5 inch drives in an external housing for USB connected tests. Maybe I can report the results on thursday.

[Erik Slagter]

VU+ only updated the BSP recently, after our codebase was frozen in preparation of 6.1-release. So the kernel update is for the next release.

There is no linux-kernel update pending in the VU+ BSP.

[anudanan]

I´ve done yesterday evening some tests with the 2,5 inch drive in an external housing over USB. The result is not so clear as I expected. So I will do more tests this evening.

[WanWizard]

There is no linux-kernel update pending in the VU+ BSP.

 

I meant drivers, the kernel hasn't been updated in years...

[anudanan]

My tests with the my two seagate 2,5inch firecuda ST2000LX001 and baracuda ST2000LM015 show no difference between the performance over the internal SATA interface and the externaöl USB3.0 interface. I´ve tested also an Seagate Expansion portable 2,5inch drive with 1TB. This has nearly the same performance as the other both 2,5 inch drives. 

 

With 14 recorded movies and in parallel cutting und moving movies I have interrupts in the recordings. But last week the same tests with an external samsung HD204ui inside an USB3.0 housing works better but I haven´t tested is now.

 

The conclusion is that everthing inside the driver seems to be ok. It looks like that the 3,5 inch drive (5400 rpm) ist faster than the 2,5 inch drive (5400rpm). I don´t now why but I have no other explanation for that. Maybe that a 2,5 inch SSD is faster but today they are to expensive. I hope the price will go down in the future

 

I use now the 2,5 drives for archive movies and the external as the working drive.

[Erik Slagter]

Some interessting information about the linux Kernel version on the openpli 6.1 with uno4kse  Until now I thought the Kernel is version 3.14, but today I´ve seen also openpli use like VTI the kernel 4.1.20. 

 

root@Schlafzimmer:/var/volatile/log# cat /proc/version

Linux version 4.1.20-1.9 (build@bouwserver2) (gcc version 6.2.0 (GCC) ) #2 SMP Tue Dec 19 01:32:18 CET 2017

root@Schlafzimmer:/var/volatile/log#

Here is claimed that some other image has (VU+) drivers for kernel 4.1.20. Which quite surprises me. I guess the example is from a model by another brand, that has a newer kernel.

Edited by Erik Slagter, 5 January 2018 - 09:55.

[anudanan]

The output from cat /proc/version came from my uno4kse with openpl i 6.1rc

[anudanan]

I´ve a look at the partition table on my HDDs and I see, that the 2,5 inch seagate baracuda ST2000LM015 drive have the partition start at 1049kB, which is not alligned so 4K sectors of the drive. My other older HD204UI has the partition start on 4096B, which is 4K sector alligned. These drive I´ve initialized with openpli 4.0

 

Did you make changes in openpli 6.1 to set the partition start on  1049kB instead of 4096B

 

I´ve red that modern drive used 4K sectory and it is useful for the performance to set the partition start to  a 4K alligned sector.

 

 

root@Wohnzimmer:/etc# parted -l

Model: ATA ST2000LM015-2E81 (scsi)

Disk /dev/sda: 2000GB

Sector size (logical/physical): 512B/4096B

Partition Table: msdos

Disk Flags:

 

Number  Start   End     Size    Type     File system  Flags

 1      1049kB  2000GB  2000GB  primary  ext4

 

 

 

Model: Seagate Expansion Desk (scsi)

Disk /dev/sdc: 2000GB

Sector size (logical/physical): 512B/512B

Partition Table: msdos

Disk Flags:

 

Number  Start  End     Size    Type     File system  Flags

 1      4096B  2000GB  2000GB  primary  ext4

 

 

I´ve tested that with an Seagate expansion drive 1TB which I´ve initialized with openpli some days ago and that have also the partition start on 4096B

 

 

Model: Seagate Expansion (scsi)

Disk /dev/sdb: 1000GB

Sector size (logical/physical): 512B/512B

Partition Table: msdos

Disk Flags:

 

Number  Start  End     Size    Type     File system  Flags

 1      4096B  1000GB  1000GB  primary  ext4

 

Edited by anudanan, 6 January 2018 - 15:48.