TR Forums

https://blocksandfiles.com/2019/12/03/m ... or-drives/

Should be exciting when this technology trickles down to mainstream drives. Both RAID 0 and 1 with just two drives. That would be pretty convenient, especially for those with large Steam libraries that can be stored on the RAID 0 volume while the RAID 1 volume can be home to the OS and important data.

You'll need an enterprise-class HBA though, as these are incompatible with the SATA ports on consumer gear. These are SAS only (no SATA), and use the SAS "dual LUN" feature to make the drives appear as two drives to the OS. So you'll need to shell out for a high-end discrete disk controller, and you'll also need an open PCIe x8 (or better) slot to put it in.

I doubt this tech will ever get implemented on SATA drives, as the SATA equivalent of dual LUN (SATA port multiplier) sucks big-time, and the consumer market has almost entirely moved on to SSDs for anything other than external backup solutions.

Hopefully they will add an extra SATA port on the drive itself and maybe updated storage drivers will take care of making the drive appear as two independent physical devices.

That would violate SATA spec for connector placement. All drives must use exact same connector location for compatibility with hot swap bays and docks.

What possible motivation do HDD manufacturers have to even do this? Enterprise customers are fine with SAS (and even prefer it, since it has 2x the burst transfer rate); and the vast majority of consumers don't give a crap about mechanical HDDs any more unless they're in an external enclosure with a USB interface. I doubt they'd make back the extra engineering cost if they released a SATA version.

Like host-managed SMR, this is an enterprise feature.

just brew it! wrote:

Like host-managed SMR, this is an enterprise feature.

Yep.

I can confirm that this isn't heading to consumer drives anytime in the near future.

Waco wrote:

I can confirm that this isn't heading to consumer drives anytime in the near future.

RATS! Praying for some hardware engineer/hacker to concoct a quick and dirty interface but that is assuming these drives are available in the retail channel by that time.

You're basically talking about some sort of dual-port SATA to SAS bridge, then. That's not going to be quick, dirty, or cheap, since it will require custom silicon.

If you really want to play with the tech, get a SAS controller. But you would be better off just getting a pair of conventional SATA drives.

Igor_Kavinski wrote:

RATS! Praying for some hardware engineer/hacker to concoct a quick and dirty interface but that is assuming these drives are available in the retail channel by that time.

The way you expressed your wish makes me suppose that the first thing you're willing to give up is reliability.

Wirko wrote:

Igor_Kavinski wrote:

RATS! Praying for some hardware engineer/hacker to concoct a quick and dirty interface but that is assuming these drives are available in the retail channel by that time.

The way you expressed your wish makes me suppose that the first thing you're willing to give up is reliability.

Well, I meant that once someone has a quick solution available and it catches on, others will jump in and competition will lead to better reliability. But someone has to initiate this tsunami and usually hacker types are more likely to take a crack at solving challenging problems.

The quick solution is to purchase a SAS controller.

I seriously doubt that anyone with enough of a clue about storage tech to be capable of designing such a device could be convinced that it is a good idea to do so.

just brew it! wrote:

The quick solution is to purchase a SAS controller.

I seriously doubt that anyone with enough of a clue about storage tech to be capable of designing such a device could be convinced that it is a good idea to do so.

Are you gonna lie that you never accomplished something amazing just coz you were feeling bored? There are enough brilliant people out there that someone might just be tempted to do it, even if just to kill some time. Look at all the great emulation projects especially http://emulation.gametechwiki.com/index.php/UltraHLE which would have been considered an impossibility by the majority of nerds. Not everyone does things for profit or only if it seems like a good idea.

By the way, a story from my youth. I was taught C/C++ in university on Turbo C 3.00 (it was the early 00s!). I was mesmerized by programming. Learned as much as I could and tried to do something practical with it. The graphics library of Turbo C was not part of our curriculum. That only meant I was drawn to it even more. So I used the library functions to draw some geometric shapes. And then it occurred to me to use these shapes to create a graphical demo of the Towers of Hanoi. I won't lie, recursion is still alien to my mind. But I looked at the function and I was like, all I need to do is give the output of the recursive textbook towers of hanoi function as an input to my graphical shape drawing function. So I started hacking away in my spare time and about a month later (I was stumped somewhere in my unwieldy code), I had a eureka moment while tending to Nature's call in the toilet (of all the places!). I hurried back to my PC and changed a few lines in my code and lo and behold! It worked like a charm. That was the day I realized that you can do anything if you are motivated enough and willing to keep at it, no matter how long it takes.

Igor_Kavinski wrote:

just brew it! wrote:

The quick solution is to purchase a SAS controller.

I seriously doubt that anyone with enough of a clue about storage tech to be capable of designing such a device could be convinced that it is a good idea to do so.

Are you gonna lie that you never accomplished something amazing just coz you were feeling bored? There are enough brilliant people out there that someone might just be tempted to do it, even if just to kill some time.

It requires multiple channels of high speed LVDS operating in the multiple GHz range, which means either an ASIC or an expensive FPGA. I'm not sure you can even get FPGAs fast enough to handle the internal data paths at ANY price (I kind of doubt it), and the test equipment required to debug it is going to cost thousands of $. You'll also need some sort of embedded processor (and associated firmware) to handle the protocol conversion. And after all that expense and (likely) hundreds (if not thousands) of hours of effort, you'll have something that might or might not work reliably, and isn't particularly useful even if it does.

Yeah, nope.

just brew it! wrote:

It requires multiple channels of high speed LVDS operating in the multiple GHz range, which means either an ASIC or an expensive FPGA. I'm not sure you can even get FPGAs fast enough to handle the internal data paths at ANY price (I kind of doubt it), and the test equipment required to debug it is going to cost thousands of $. You'll also need some sort of embedded processor (and associated firmware) to handle the protocol conversion. And after all that expense and (likely) hundreds (if not thousands) of hours of effort, you'll have something that might or might not work reliably, and isn't particularly useful even if it does.

Yeah, nope.

Sometimes having intricate knowledge of the inner workings of something might actually hinder you from thinking out of the box. Devices designed by engineers are hacked by people who themselves would have a hard time getting paid anywhere near those engineers.

Sure, occasionally you'll encounter the "I didn't know this was supposed to be impossible, so I just went ahead and did it!" effect. But most of the time, if a problem looks complicated to someone with knowledge of the field, it is.

Dual-actuator drives are a meme. It is the grasping the straws of a dying medium. You are throwing in more points of failure and making it more difficult to do disaster recovery for at best, a 100% I/O throughout gain. It is cheaper/easier to slap two HDDs together via RAID/JBOD if your goal is to increase I/O throughput. The only advantage that dual-actuator drives have over two single-acutator HDDs running in a RAID/JBOD setup is density. Dual-actuator drives aren't even a new idea either. They were experimental multi-actuators units back in early days of HDDs. They didn't go beyond testing units for obvious reasons.

Igor. If you want your Steam library to load faster, put it on SSD.

Krogoth wrote:

The only advantage that dual-actuator drives have over two single-acutator HDDs running in a RAID/JBOD setup is density.

Actually, no. Density (GB per drive bay) is no better than normal drives, as this is limited by how many platters you can cram into a 3.5" form factor. You already mentioned the only real advantage earlier in your post - up to 2x performance, depending on workload.

Krogoth wrote:

You are throwing in more points of failure and making it more difficult to do disaster recovery for at best, a 100% I/O throughout gain. It is cheaper/easier to slap two HDDs together via RAID/JBOD if your goal is to increase I/O throughput.

Not really - the failure domain is pretty identical to a standard dual-port SAS drive. The hardware is only very slightly different.

They are a play to get back to older standards of IOPs/TB for particular customers.

just brew it! wrote:

Sure, occasionally you'll encounter the "I didn't know this was supposed to be impossible, so I just went ahead and did it!" effect. But most of the time, if a problem looks complicated to someone with knowledge of the field, it is.

Yeah, and usually that happens in fields like 3D printing which are still relatively new. Of course there's still open problems in the hardware field, but we left behind those that could be done on a table top a long time ago.

Igor_Kavinski wrote:

Sometimes having intricate knowledge of the inner workings of something might actually hinder you from thinking out of the box. Devices designed by engineers are hacked by people who themselves would have a hard time getting paid anywhere near those engineers.

And sometimes nobody will do these "impossible" things because they don't make any sense. If you want a "fake" dual actuator drive, just buy two HDDs and slap a RAID array across them. Partition 1 on each == RAID 1 (for redundancy that's actually real) and partition 2 on each == RAID 0 (for speed or whatever you were trying to accomplish).

Doing a RAID 1 across two LUNs in a dual actuator drive would be a VERY bad idea if you think it buys you anything other than protection from a particular sector going bad. If you get a head crash, the whole drive is likely to go out anyway.