Intel Optane...The ultimate SLOG device?

[NickF]

Has anyone seen this?
https://www.youtube.com/watch?v=bO_fh450u6Y&t=29s

This looks pretty cool... I am excited to see where this technology goes.. Maybe a cheap reliable SLOG... or if its really good a better L2 ARC.
http://www.intel.com/content/www/us/en/architecture-and-technology/intel-optane-technology.html

http://www.anandtech.com/show/11227/intel-launches-optane-memory-m2-cache-ssds-for-client-market

Anyone thoughts?

 

[MatthewSteinhoff]

That was my first thought, too: extremely low latency and high endurance. Seems like a really good SLOG.

I hear a lot of complaints about the cost compared to the storage space but I don't think they understand how best to deploy this technology.

I'm not going to be an early adopter but I'm certainly excited.

Cheers,
Matt

 

[Arwen]

Uh, guys, did you read the part about 280 Mega Bytes per second write speed?

That makes it less suitable for a SLOG, which is write mostly. Some of cheaper
SATA SSDs have write speeds up to 540 Mega Bytes per second.

Now the L2ARC is a different story, as it is a read mostly. So the 1,200 Mega
Byte per second read speed makes a world of difference compared to high
performance SATA SSDs.

 

[WorBlux]

Waking up an old thread, trying to dig out some more info.

Only found one useful benchmark. http://www.storagereview.com/intel_optane_memory_review

While it doesn't have a lot of Raw I/0 speed, it looks like could be very suited to some SLOG workloads considering how cheap it is, Particularly the database benchmark. It seems to be competitive with a large number of small writes intermixed with reads. I'd be thrilled if anyone had any more relevant benchmarks.

 

[Arwen]

If I understand ZFS SLOGs, you might be able to stripe several to get better apparent speed. A single transaction LOG entry will still go to a single device, (or Mirror of SLOG devices). However, the next LOG entry would be to the next SLOG device when they are striped.

(Of course I don't have access to the zpool manual page right now... so I could have mis-remembered.)

 

[Ericloewe]

Waking up an old thread, trying to dig out some more info.

Only found one useful benchmark. http://www.storagereview.com/intel_optane_memory_review

While it doesn't have a lot of Raw I/0 speed, it looks like could be very suited to some SLOG workloads considering how cheap it is, Particularly the database benchmark. It seems to be competitive with a large number of small writes intermixed with reads. I'd be thrilled if anyone had any more relevant benchmarks.

Well, SLOG is a write-only workload, so don't read too much into those numbers.

 

[WorBlux]

Well, SLOG is a write-only workload, so don't read too much into those numbers.

Right, I just think it'd be interesting to see some real SLOG benchmarks, definitely considering you can get a mirror in for $100/ $160. Maybe not the ultimate option, but perhaps right for a few niches.

 

[Ericloewe]

Not really, since they don't have power loss protection, making them near-useless as SLOG devices.

 

[WorBlux]

Not really, since they don't have power loss protection, making them near-useless as SLOG devices.

Thanx. Seems like a weird move for Intel not to include that, considering thier RST driver uses it as a write cache. Though I'm guessing its not much more of a risk than spinning rust and power loss.

 

[Ericloewe]

It's consumer-grade through and through.

 

[sWORDs]

Not really, since they don't have power loss protection, making them near-useless as SLOG devices.

There is no DRAM on XPoint, so no caps needed.

 

[Ericloewe]

There is no DRAM on XPoint, so no caps needed.

I might hesitantly buy that on an enterprise controller, but not on a low-end consumer device. There always has to be some state that needs to be maintained and it's not just a matter of throwing together a controller and calling it a day.

 

[sWORDs]

I might hesitantly buy that on an enterprise controller, but not on a low-end consumer device. There always has to be some state that needs to be maintained and it's not just a matter of throwing together a controller and calling it a day.

Where my Intel 750's have it, my 900p hasn't and none of the Optane drives do as you can see in the tear down of the enterprise P4800x here:

https://www.anandtech.com/show/11930/intel-optane-ssd-dc-p4800x-750gb-handson-review
"Aside from the number of 3D XPoint memory packages and the number of dies they contain, the PCBs of the P4800X and the 900p differ only in the presence of a few diagnostic LEDs on the P4800X. Neither model has any large power loss protection capacitors because the Optane SSDs don't use any volatile caches and write immediately to the 3D XPoint memory itself."

And I do get why phase change wouldn't require it while NAND does.

Intel does state it for the P4800x in ark, but that's most likely a mistake.

 

[Ericloewe]

As I said, I can accept that they can get away with not needing capacitors in their high-end controllers. It's possible.

However, I do not accept without independent testing that the low-end controllers sacrifice some performance for data integrity on power loss. While the drives might not have a large DRAM, they do need some sort of cache, since they cannot pipe data from the PCIe bus to the ICs. By acknowledging that write before it's actually finished, they can eke out a little bit more performance out of a low-cost controller, in a market where power loss protection is simply not used.

my 900p hasn't

It's not a low-end consumer device, it's a high-end consumer device trivially adapted from the enterprise one. I can imagine them not bothering to change the firmware just to eke out a tiny extra bit of performance there, since that would require extra validation and would be mostly meaningless in practice.

Intel does state it for the P4800x in ark, but that's most likely a mistake.

It's more likely to just be reuse of an equivalent field for flash SSDs, signifying that the drive can withstand that scenario. It was recently changed on the consumer model - it could be pure marketing or it could signify different firmware behavior.