ReRAM boosts SSD performance 11-fold

[mashumk]

This would be neat if it turns commercially viable. I wonder when and how much will it cost me.

ReRAM boosts SSD performance 11-fold

http://www.bit-tech.net/news/hardware/2012/06/15/reram-ssd-boost/1

 

[FelixDeCat]

As someone noted in the comment section "Just when SSDs were getting affordable...BAM!"

I too wonder how long before it does hit the market.....2 years maybe?

 

[Eug]

The power savings would be very welcome. IMO, most SSDs still use far too much power.

 

[Cerb]

Where's this 11-fold number coming from? Copying the same 4K block a million times? If it can save power and improve performance great, but the article offers no indication oh what kind of test was used to come up with their numbers. How much does the ReRAM cost? How much will be needed to actually improve performance significantly? What are the probably real performance improvement numbers?

 

[DigDog]

before i even read the link ...

waaht??

meaning now SSDs read at 5Gb/s ?? or am i missing something? or is it just "advertisement jargon"? (gonna read the link now)

ok ...

so, the quoted 4.2Mb/s write speed, is that per chip ? i'm having trouble understanding how can a SSD like a Vertex be quoted at 200-500Mb/s write and then these guys say theirs is faster at 46Mb/s

confuzed?

 

[kernelc]

Where's this 11-fold number coming from? Copying the same 4K block a million times? If it can save power and improve performance great, but the article offers no indication oh what kind of test was used to come up with their numbers. How much does the ReRAM cost? How much will be needed to actually improve performance significantly? What are the probably real performance improvement numbers?

Yeah, I agree.

Moreover, enterprise servers often have RAID controller with aboundant DRAM cache that is used for this very same purpose: limit the number of total I/O operations passed to the disk (or SSD) backend.

So I see this ReRAM cache as a mean to drive cost low, rather then to obtain higher performance.

Regards.

 

[Makaveli]

before i even read the link ...

waaht??

meaning now SSDs read at 5Gb/s ?? or am i missing something? or is it just "advertisement jargon"? (gonna read the link now)

ok ...

so, the quoted 4.2Mb/s write speed, is that per chip ? i'm having trouble understanding how can a SSD like a Vertex be quoted at 200-500Mb/s write and then these guys say theirs is faster at 46Mb/s

confuzed?

I don't see why the confusion its fairly obivous they are talking about the internal chip to chip communication inside the SSD and not the disc speed you would see from the OS side.

Maybe someone can provide a more technical explanation?

 

[dave_the_nerd]

I don't see why the confusion its fairly obivous they are talking about the internal chip to chip communication inside the SSD and not the disc speed you would see from the OS side.

Maybe someone can provide a more technical explanation?

They're talking about individual NAND chips. SSDs in computers divide your data up, and read/write to multiple chips at once, just like RAID-0.

Controllers then use caching and/or compression algorithms to speed up even more.

From Wikipedia:

When multiple NAND devices operate in parallel inside an SSD, the bandwidth scales, and the high latencies can be hidden, as long as enough outstanding operations are pending and the load is evenly distributed between devices.[31] Micron and Intel initially made faster SSDs by implementing data striping (similar to RAID 0) and interleaving in their architecture. This enabled the creation of ultra-fast SSDs with 250 MB/s effective read/write speeds with the SATA 3 Gbit/s interface in 2009.[32] Two years later, SandForce continued to leverage this parallel flash connectivity, releasing consumer-grade SATA 6 Gbit/s SSD controllers which supported 500 MB/s read/write speeds.[33] SandForce controllers compress the data prior to sending it to the flash memory. This process may result in less writing and higher logical throughput, depending on the compressibility of the data.[citation needed]

 

[Batmeat]

The power savings would be very welcome. IMO, most SSDs still use far too much power.

Seriously?

http://www.tomshardware.com/reviews/ssd-330-review-benchmark,3190-8.html

 

[KillerBee]

That was weird .... http://power ?

http://www.tomshardware.com/reviews/ssd-330-review-benchmark,3190-8.html

 

[zephyrprime]

Resistive ram technology is fundamentally better than flash (floating gate) technology and will probably replace it eventually. Flash has problems scaling down to small scales but resistitive is supposed to be able to do it. Also, flash is fundamentally slow. We mask this slowness in applications by designing with high parallelism but parallelism doesn't help you with small read/writes and it hurts durability because you can't write less than a block as wide as your parallelism is. Oh, and flash is not very durable either. Resistive is durable.

 

[Magic Carpet]

Sadly, Takeuchi and his team are silent regarding commercial availability of the MLC-and-ReRAM components, but with numerous companies already investigating commercialisation of ReRAM parts - including memory giant Elpida - it surely can't be too far away.

Damn it D:

 

[taltamir]

What this article is discussing is a controller for a hybrid drive.

Current hybrid drives have a large spinning HDD with a tiny SSD acting as a cache. This causes a small increase over HDD prices due to using so little NAND.
This hybrid drive uses large SSD with tiny amount of ReRAM. It shouldn't be much more expensive, but for any application generating constant writes and filling up the cache faster then it can be flushed there is no benefit beyond those first few seconds writing into cache only. Of course in the real world such applications are not that common and if tuned correctly it will result in very very high IOPS and random write performance.

It is currently possible to get equal results to what they describe using regular ram as cache with an array of capacitors to prevent catastrophy in the case of powerloss.
So all this does is replace the very cheap regular ram + semi expensive capacitors with very expensive ReRAM with no/fewer capacitors.

For it to be commercially viable ReRAM has to be really expensive compared to NAND so you can't make an entire SSD just out of it, but not so expensive as to make regular RAM + capacitors cheaper.

Also, AFAIK there are already existing SSDs that rely on ram cache don't perform all that amazingly

 

[Makaveli]

They're talking about individual NAND chips. SSDs in computers divide your data up, and read/write to multiple chips at once, just like RAID-0.

Controllers then use caching and/or compression algorithms to speed up even more.

From Wikipedia:

Thanks dave that is the post I was waiting for. I was at work so didn't have time to expand on what I posted.

 

[taltamir]

I don't see why the confusion its fairly obivous they are talking about the internal chip to chip communication inside the SSD and not the disc speed you would see from the OS side.

Maybe someone can provide a more technical explanation?

They explicitly state and their picture also explicitly shows that this is NOT what they are talking about.

They are talking about hybrid drives where you have many NAND chips and a single ReRAM chip working as cache. NOT chip to chip communication inside the SSD.