Chipset combo to enable ripping multiple SATA optical drives behind port multiplier?

[byrdman]

Searched on "port multiplier" and got 5 results that didn't address this... anyway, to the pain:

Found a good deal on 5x HL...GH24NSC0 drives and a 5-bay tower. Performance is abysmal, 2x to 5x each disk with all disks running. Bottleneck appears to be behind the port multiplier. If I run 2 DVD drives in the tower and my laptop drive, I get 36x, 31x, 16x (last is laptop). If I run all 3 in the tower, I get 6x, 6x, 6x. Crap is evil... Enabling C2 drops it to 1x or less. Nero CDSpeed gives essentially the same results. In ripping all drives CPU utilization stays <15% at all times.

First port multiplier was an Oodelay USB/eSATA 1 to 5 card. Flat out didn't work, contacted them and they essentially said "Sorry, you can't run USB to optical". Got an ExpressCard->eSATA for the laptop and now it works, abysmal perf. Contacted again, got a note saying essentially that the RAID wouldn't allow it to work, needed another card. Got that, same abysmal perf. Then they just said sorry, too many variables, should get an actual duplicator. I don't care about duplicating discs.

Did some investigation into power saving, PIO vs DMA, etc, it all checked out OK. Saw some stuff on StackExchange and Tom's Hardware regarding chipset compatibility...

The ExpressCard uses the Marvell 88SE9128 chip (verified running PCIe 2.0), current port multiplier is a JMicron JMB575. Getting any useful info about these chips is a PITA. Everything is a marketing brochure. If I had some way of getting an errorlog or performance counters, or trace arrays, this would be much easier. (Would also help to take the Mindshare class on SATA.) Seems like SiliconImage chips might work. One person (I think here or Club Myce) recommended a DATOptic card with a SIL-3726, but the link to Amazon showed that it was discontinued, Addonics had a SIL-3726 card, but couldn't answer basic questions.

Which brings me to PCPitstop, which had a port multiplier based on a Marvell 88SM9785. These people were less clueless, but were a reseller, not an integrator. Marvell's page seems to imply optical support.

Does anybody have experience with Marvell 88SE9128 --> Marvell 88SM9785 for multiple optical drive ripping?

More importantly, has anyone slapped >3 SATA optical drives into a box, run a single eSATA or USB cable to it, and gotten all the drives to rip at >20x behind a port multiplier? If so, what are the chipsets of the HBA and the port multiplier?

FWIW, running Thinkpad W530, i7 class proc, 16GB mem, Win7SP1, dbPowerAmp 15.3

thanks in advance for any advice...

[Spoon]

Re: Chipset combo to enable ripping multiple SATA optical drives behind port multipli

You would be best getting a cheap tower unit, CPU is not important (anything Pentium from the last 4 years will do) and stick 4 drives to real sata ports.

[byrdman]

Re: Chipset combo to enable ripping multiple SATA optical drives behind port multipli

Not exactly ideal for getting it to my laptop drive, but I guess I could sneakernet it with USB sticks.... except that I already bought the box and drives

More fundamentally, I wonder if SATA optical drives work behind a port multiplier or is there a hole in the SATA spec such that different implementers could build port multipliers that optical drives couldn't work with. (ATAPI???) I'd expect a 1->5 port multiplier to allow 70-80% of 1's bandwidth to be equally utilized by the fanout. Looks like here, something is being clogged up since a SATA-I interface should be able to absorb 5 DVD-ROMs pretty well, much less SATA-III, even if it is transaction-bound rather than bandwidth-bound.

I went ahead and ordered the Marvell-based PC-Pitstop part. Since both HBA and MP chips are from the same company, maybe they have been simulated together... or not. We'll see what that does.