[plug] Linux hardware compatibility

[Michael Collard]

On Sat, 2005-01-29 at 17:30 +0400, Brad Campbell wrote:
> Michael Collard wrote:
> 
> > I hear a few people saying things like this regarding VIA chipsets but
> > I've never seen it. What are the actual problems that you have?
> 
> On the server I have 14 Hard disks, spread across 3x Promise SATA150TX4 PCI boards, 2 On board VIA 
> SATA ports and 1 on board VIA ATA Port.
> 
> If I stream a read from 2 drives on the promise controllers (any one) I can saturate the PCI bus.
> I can get sustained transfer rates of 120MB/s. If I then try across boards, 1 drive on 1 board and 
> one on another, this drops to about 70MB/s. If I spread the load across all the boards (as you do in 
> a 10 drive RAID-5) I drop to below 50MB/s.
> 
> When utilising the network interfaces at the same time, my PCI bus throughput often drops to about 
> 20MB/s. Not hard disk bandwidth. Calculated bandwith accounting for the network cards and hard disk 
> controllers. It just blows goats. This just never happened on an AMD760 chipset.
> 
> As for the HTPC. I have 2 BT878 based cards, one analogue tuner and a DVB card. The bandwidth 
> limitations there become really apparent when trying to record from analogue and DVB at the same 
> time. Again, never on an AMD760 chipset. Plus if I init one of the BT878 cards incorrectly, it locks 
> the PCI bus completely solid. Hard hang up that not even the NMI oopser can recover from. This did 
> not happen on an Intel chipset.

I've been able to replicate that problem with my DVB and BT878 cards. I
have to init the DVB card before the bt878 card or goodbye system.

I did some tests as follows regarding I/O performance, with and without
the DVB card streaming.

Created file:
dd if=/dev/urandom of=testfile
-rw-r--r--  1 quadfour quadfour 179M Jan 29 21:50 testfile

Filesystem            Size  Used Avail Use% Mounted on
/dev/hda1             8.3G  4.9G  3.0G  62% /
/dev/hda3              65G   63G  1.8G  98% /home
/dev/hdb5              63G   57G  6.6G  90% /stuff
/dev/sda1             184G   27G  157G  15% /stuff2

** Without DVB ** 

[quadfour at natalie stuff2 $ ] time mv testfile /stuff/
0.01user 0.60system 0:00.87elapsed 70%CPU (0avgtext+0avgdata
0maxresident)k
0inputs+0outputs (1major+164minor)pagefaults 0swaps

[quadfour at natalie stuff2 $ ] time mv /stuff/testfile ~
0.01user 0.61system 0:05.85elapsed 10%CPU (0avgtext+0avgdata
0maxresident)k
0inputs+0outputs (0major+165minor)pagefaults 0swaps

[quadfour at natalie stuff2 $ ] time mv ~/testfile /stuff2/
0.01user 0.61system 0:00.93elapsed 67%CPU (0avgtext+0avgdata
0maxresident)k
0inputs+0outputs (0major+165minor)pagefaults 0swaps

The 1st and third are probably caching or something. The second shows
correct performance.

** Without DVB **

-rw-rw-r--  1 quadfour quadfour 3.4G Jan 18 16:08
Mandrake-10.1-x86_64-DVD-01.iso

[quadfour at natalie stuff2 $ ] time cp
Mandrake-10.1-x86_64-DVD-01.iso /stuff/
0.23user 15.04system 2:30.64elapsed 10%CPU (0avgtext+0avgdata
0maxresident)k
0inputs+0outputs (1major+154minor)pagefaults 0swaps

** With DVB **

-rw-r--r--  1 root     root     1.2G Jan 13 16:18
mandrake10-64-130105.tar.gz

[quadfour at natalie stuff2 $ ] time cp
mandrake10-64-130105.tar.gz /home/quadfour/
0.09user 15.73system 1:07.19elapsed 23%CPU (0avgtext+0avgdata
0maxresident)k
0inputs+0outputs (1major+154minor)pagefaults 0swaps

[quadfour at natalie stuff2 $ ] time cp
Mandrake-10.1-x86_64-DVD-01.iso /stuff
0.25user 14.59system 2:41.03elapsed 9%CPU (0avgtext+0avgdata
0maxresident)k
0inputs+0outputs (1major+154minor)pagefaults 0swaps

** With DVB **

-rw-r--r--   1 quadfour quadfour 695M Oct  2 02:20
Mandrakelinux-10.1beta1-CD1.x86_64.iso

[quadfour at natalie stuff $ ] time cp
Mandrakelinux-10.1beta1-CD1.x86_64.iso ~
0.05user 7.59system 1:00.96elapsed 12%CPU (0avgtext+0avgdata
0maxresident)k
0inputs+0outputs (0major+155minor)pagefaults 0swaps


With this, there was only a slight slow down when copying data while
streaming DVB data. Also, there was a noticeable lack of response when
copying to/from the SATA drive, but I believe this is because its kinda
done with software (VIA controller) though I could be wrong.

What do you think about the above, considering that all the speeds
(except maybe involving /home because it was in use by other apps). It
also seems to me that the transfer rates were rather good compared. I
was getting up to 22Mbytes/sec, compared to your 150MBit/sec (I assume
it was bits) 

Cheers
Michael Collard