Bad I/O performance Highpoint Rocket 1520 SATA controller (kernel 2.6.11.12)

Bad I/O performance Highpoint Rocket 1520 SATA controller (kernel 2.6.11.12)

[Pedro Venda]

[-- Attachment #1: Type: text/plain, Size: 2373 bytes --]

hi,

I've been having bad I/O performance with this SATA controller card
(Highpoint Rocket RAID 1520). Kernel version is 2.6.11.12.

First I used it with 2 samsung 80G P.ATA drives (with adapters) and it 
was not so good... about 15Mb/s uncached throughput for each disk. Now 
I've got 2 new Maxtor 200G native SATA drives attached and... the numbers 
didn't change.

The old samsung drives now attached to a promise Ultra133 TX2 P.ATA 
controller are quite faster than before and smoke the SATA drives... 
weird.

What follows are some (modified) snips of dmesg:

hde: SAMSUNG SP0802N, ATA DISK drive
hde: max request size: 1024KiB
hde: 156368016 sectors (80060 MB) w/2048KiB Cache, CHS=16383/255/63,
UDMA(100) DISK drive
hdg: [same as hde]

hdi: Maxtor 6B200M0, ATA DISK drive
hdi: max request size: 1024KiB
hdi: 398297088 sectors (203928 MB) w/8192KiB Cache, CHS=24792/255/63,
UDMA(33)
hdk: [same as hdi]

PDC20269: IDE controller at PCI slot 0000:00:03.0
     ide2: BM-DMA at 0xb000-0xb007, BIOS settings: hde:pio, hdf:pio
     ide3: BM-DMA at 0xb008-0xb00f, BIOS settings: hdg:pio, hdh:pio
HPT372A: IDE controller at PCI slot 0000:01:02.0
     ide4: BM-DMA at 0x8400-0x8407, BIOS settings: hdi:DMA, hdj:pio
     ide5: BM-DMA at 0x8408-0x840f, BIOS settings: hdk:DMA, hdl:pio

Now, hdparm -tT tests show what I mean (results are median (not average,
the middle value) of 3 successive runs):

/dev/hde:
  Timing cached reads:   728 MB in  2.01 seconds = 362.97 MB/sec
  Timing buffered disk reads:  158 MB in  3.02 seconds =  52.27 MB/sec
/dev/hdi:
  Timing cached reads:   772 MB in  2.00 seconds = 385.67 MB/sec
  Timing buffered disk reads:   42 MB in  3.04 seconds =  13.80 MB/sec

One could think: well, your samsung drives got faster (~4x) and your
Maxtor drives are faulty. Not quite. If I re-run the tests plugging the
Maxtors via SATA-P.ATA adapters into the promise card, they score about 
63Mb/s each for "Timing buffered disk reads" (no snip for that, just 
believe me).

Is there anyone with an explanation or should I assume the Highpoint
SATA controller card... sucks. What could be a better replacement? Are
Promises SATA150 TX4 any good (on linux, of course)? or maybe SATA300 TX4?

best regards,
pedro venda.
--
Pedro João Lopes Venda
pjvenda < at > arrakis dhis org
http://arrakis.dhis.org

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: Bad I/O performance Highpoint Rocket 1520 SATA controller (kernel 2.6.11.12)

[Pedro Venda]

Roger Heflin wrote:

>>hde: SAMSUNG SP0802N, ATA DISK drive
>>hde: max request size: 1024KiB
>>hde: 156368016 sectors (80060 MB) w/2048KiB Cache, CHS=16383/255/63,
>>UDMA(100) DISK drive
>>hdg: [same as hde]

>>HPT372A: IDE controller at PCI slot 0000:01:02.0
>>     ide4: BM-DMA at 0x8400-0x8407, BIOS settings: hdi:DMA, hdj:pio
>>     ide5: BM-DMA at 0x8408-0x840f, BIOS settings: hdk:DMA, hdl:pio

> 1520/1640's are low end, Highpoint 1810, and 1820 are real raid
> and quite fast, but need their extra driver to work.  The newer
> 2220 are quite good also, they are probably also quite a bit
> more money.

hi roger, thanks for the reply,

besides being 100% true, it's irrelevant. I'm not using the card's raid 
features. As a standalone drive controller, it should be no different 
from a non-raid version (it's HPT302 instead of HPT372) and therefore 
should do the job normally i.e. with proper I/O speeds.

> I won't use a 1520/1640.   I think they are also fakeraid.

yes, they are. I'd be interested in finding out decent and well 
supported (linux opensource drivers, preferably in kernel tree) low-end 
(cheap) non-raid SATA controllers for 2/4 drives.

best regards,
pedro venda.
--
Pedro João Lopes Venda
pjvenda < at > arrakis dhis org
http://arrakis.dhis.org