PCI 4/8 port

[tucansam]

The older Promise Tech Fast Track 4 and 8 port PCI SATA-II controllers can be had on ebay for a song, and I have a four port model in an older machine I used to test unraid before I rolled it out, so I know they work. 

 

I'd like an M1015, but its not going to happen at the moment.

 

So my question is, how much is the PCI bus going to limit my speeds?  I don't need blistering speeds, but let's pitch two scenarios:

 

- Fast Track has four drives attached, and two or three clients try to stream BluRay rips from the same drive on that controller.  Do able or not?

- Fast Track has four drive attached, and two or three clients try to stream BluRay rips from two or three rives on that controller.  Do able or not?

 

I'm just trying to think of a worst-case scenario, to see if the controller can handle it.

 

Thanks.

[Joe L.]

The PCI bus will limit combined throughput to ALL disks connected through it to 133MB/s.  If you have 10 disks, then you'll get 13MB/s parity calc/check speed.  That is your worst case.

 

The PCI bus will not really be an issue when accessing one or two disks (typical reading or writing activity)  Reading from 4 disks will be pushing it, but if the bit rate on the Blue-Rays is not too high, it might make it, a lot depends on the controller.  Reading 4 streams from 1 disk might be pushing it, as the disk head will really be seeking, and the limit might be the ability of the disk to keep up.  (Just don't even think of doing a parity check while attempting this...)

 

Joe L.

[Johnm]

Actually, in said example. the PCI drives, if not the whole sever is probably out of commission during parity checks. due to bandwidth issues.

 

2TB at 13MB/s ? eek.. thats a long time to be down.

 

As mentioned. for a few drives, you should be ok. I would not try to write to any other drive on the PCI bus while streaming multiple blurays.

 

[Joe L.]

Actually, in said example. the PCI drives, if not the whole sever is probably out of commission during parity checks. due to bandwidth issues.

My older server is PCI based.  (PCIe did not exist when it was purchased) It has mostly IDE drives (SATA was barely out when its MB was designed)  The parity disk and one other is SATA, but they are older 1TB drives.

2TB at 13MB/s ? eek.. thats a long time to be down.

Not down at all.  I've never had the parity check affect use of the array.  Movies play just fine.  My parity check speed is between 12 and 16 MB/s

As mentioned. for a few drives, you should be ok. I would not try to write to any other drive on the PCI bus while streaming multiple blurays.

agreed.

[tucansam]

OK... Some testing.

 

I have three disks on the onboard SATA controller, and four disks on the PCI Promise four-port controller.

 

I can stream two blurays at the same time via the same disk on the Promise, or via two disks on the Promise.  If I add a third bluray, I get stuttering (didn't try streaming from the onboard).

 

So assuming I don't have more than two clients streaming at the same time, or maybe a third client streaming a dvd instead, I should be OK.

 

As for parity checks, the parity drive is on the onboard SATA.  I'm just looking for a cheap(er) way to add more member disks.

 

Still, the eight-port models are $80 or so on Ebay, while the M1015 is just a bit more.

 

The AOC-SASLP-MV8 is on newegg for $109, but I read this review which gives me concern:

 

--

Pros: This is a very inexpensive 8-port SATA card with (mostly working) Linux. I've run them for years in Linux file servers using patches from LKLM, but no stock kernel driver works under load. I've emailed the Marvell driver engineer and begged him to get the patches into shape for merging with upstream, but this hasn't happened during the past 2-3 years, and I'm swtiching to LSI.

 

Cons: Linux drivers *appear* to work, but under load they can *time out*. This can introduce *silent* data corruption (e.g., one bit error in a 1GB file -- enough to change the md5 hash, but not enough to be noticed by many applications).

 

Other Thoughts: I really had a lot of hope for this card under Linux, but have finally given up. The low-end LSI card is more than twice the cost, but it performs substantially better without annoying timeouts (or, with an unpatched kernel, data loss).

--

 

Can anyone speak to this?  I'd but the Supermicro today just for the speed and newness if the above review is not entirely correct.