OCZ unveils 'first' 2.5-inch SSD with 1TB, plus PCI Express-based hybrid

The big SATA interface SSD drives are interesting for folks with high performance needs and fat wallets. However the PCIE hybrid is a step backward. It apparently requires a Windows only driver. That is not any help for most of us on this site and I hope it is not a sign of things to come. Needless to say it is not something I will even consider.

I'm trying to think of a rational reason that a PCIE device or harddrive wouldn't be OS independent.

it's probably because their solution is simply a hack.

the drive has to be used as the primary disk, otherwise the driver will cache the access from your real primary disk and not from the hybrids own disk.

this means that the card simply presents two drives to the OS and the driver is used to intercept access to the primary disk and cache it on the SSD. so the driver doesn't even depend on the disk being stuck to the card. it is not a hardware interface, but a software one.

presumably the same could be done in a linux driver, completely independent of the hardware. the driver could present a partition that is set up from two physical partitions: a large data partition on a slow disk and a fast cache partition on a fast disk. of course such a linux partition would let me chose which disks to use. to configure the partition the already existing raid api could be used.

greetings, eMBee.

Yep, their solution is certainly a hack if it needs a special driver. To be OS-independent, it'd need an onboard CPU (like a small ARM CPU) to do the work this windows driver is doing, and present the whole unit to the computer as a SATA or SAS drive. Basically, they're saving money on the ARM CPU by having the host CPU do all the work. It's similar to software RAID boards ("fakeraid"), which do the RAID processing on the host CPU instead of offloading it to a secondary CPU, but then require a special driver. Like those, Linux could implement this stuff in a special driver, but they just haven't done so yet. With the fakeraid adapters, they did because there were so many of them, they were popular, and architecturally they were all pretty much the same so you didn't need a special driver for each different board.

though i am not suggesting to write a driver for this device like drivers for raid devices, but rather something like the generic raid support that is not written for raid devices but works on any disk.

similar to

Quoting:mkraid /dev/md0; mdadm --create --verbose /dev/md0 --level=linear --raid-devices=2 /dev/sdb6 /dev/sdc5

i'd like to do:

Quoting:make-cache-partition /dev/cp0 --cache-device /dev/ssda1 --data-device /dev/sda5

this would create a new partiton /dev/cp0 which uses /dev/sda5 as its data partition and /dev/ssda1 as its cache.

greetings, eMBee.

Since SSD drives are getting popular and ubiquitous these days, and this isn't the first instance of using flash drives for caching, I'm surprised someone hasn't already made a Linux kernel driver to do just this.

Take a look in the kernel source, at Documentation/filesystems/caching/fscache.txt. It might be close to what you're talking about.

(And I might be out in left field....)

no, i think you are in the right field, this looks really like something that could be what we are looking for here. at the worst it is hardcoded to assume that the original filesystem is really a network filesystem, then it would take a bit of hacking to make it accept local drives, but other than that it looks like a good fit,

even the part where it says that it must be possible to operate without cache and handle files larger than the cache and it needs to be able to access parts of the file without loading the whole thing, these all seem to make this suitable for local usage. of course i am no filesystem expert, but for someone who wants to have a go at this, this driver looks like a good choice to start.

greetings, eMBee.

Aside from the Windows-only driver problem, that hybrid PCIe drive looks like a terrible design: where's the mechanical support? A 3.5" drive is not a lightweight item, and there doesn't appear to be any kind of mechanical support for the thing, as they mounted the drive on the opposite end from the mounting bracket. So the entire weight of the drive will be hanging off the PCIe slot. This probably isn't a problem in a horizontal, desktop case, or perhaps a 4U server chassis, but I haven't seen a "desktop" case in at least 10 years now, if not 15, since towers took over. In a tower case, the PCIe slots are sideways, so any cards plugged into them are horizontal. So this drive will be putting a large torque onto that tiny PCIe x4 connector, which it certainly wasn't designed for. It might have been workable if they had mounted the drive right next to the mounting bracket, but they didn't.

And with a Windows 7 driver, what good is it going to be when Windows 8 comes out? What guarantee do you have that the mfgr will bother to make a Win8 driver for what will be older models when Win8 arrives? This is why buying Windows-only stuff is a bad idea. Tons and tons of perfectly good hardware have been thrown in the trash because mfgrs didn't bother to supply drivers for newer Windows versions.