DapuStor Officially Launches High-Capacity QLC eSSDs up to 64TB

[Nomad76]

As AI accelerates data expansion, enterprises face increasing challenges in managing large volumes of data effectively. Tiered storage solutions have emerged as the preferred approach for balancing performance and costs. Solid State Drives (SSDs), with their fast read and write speeds, low latency and high power efficiency, are becoming the dominant storage selection for data centers and AI servers. Among SSD techniques, QLC SSDs offer unique advantages in costs and storage density, making them particularly suited for read-intensive AI applications. Therefore, high-capacity SSDs, such as 32 TB and 64 TB models, are gaining traction as a new storage solution in the market.

Read-Intensive Applications: Mainstream Enterprise SSD Use Cases
According to the latest research from FI (Forward Insight), up to 91% of current PCIe SSDs deployments are used in applications with DWPD of less than 1, and the share is expected to reach 99% by 2028. this shift underscores the increasing prevalence of read-intensive applications and data centers, a space where QLC SSDs excel.



Power Efficiency: the Foundation for Large Storage
DapuStor QLC eSSDs are designed to handle massive data workloads from core to edge, supporting ultra-high capacities of 15.36 TB to 61.44 TB. This high storage density delivers greater economic value and more environmentally sustainable storage solutions. In this case, DapuStor offers two 30.72 TB QLC models: the J5000 and J5060, both upholding multiple mapping schemes (4 KB / 8 KB / 16 KB). Additionally, DapuStor QLC SSDs deliver impressive full-capacity random read performance of up to 1,500K IOPS, rivaling mainstream high-performance TLC SSDs achieving both efficiency and low power usage

DapuStor QLC SSDs: Key Features

• 4 KB / 8 KB / 16 KB Mapping Granularity: Multiple mapping options, with 4 KB adapting to applications automatically and 8 KB / 16 KB offering cost savings.
• Low Read Power Consumption: Power consumption during reads is as low as 12 W, significantly reducing client energy costs.
• High Read Performance: Up to 1,500K IOPS in read scenarios.
• Optimized QLC R/W QoS: Prioritizes read scenarios, with optimizations for Charge Trap NAND Flash and enhances QLC endurance and power consistency.
• Enhanced Power Loss Protection: Ensures data integrity during unexpected power failures, with capacitor self-check mechanisms to protect data.
• Dual-Port Support: Facilitates system maintenance and upgrades, making it ideal for core storage applications.
• Optimized Write Buffer Algorithm: Direct data writes to QLC minimize data movement, improving write efficiency.

Improved Endurance: DapuStor QLC SSD J5000
The DapuStor QLC SSD J5000 significantly outlasts traditional HDDs in endurance. In sequential write scenarios for large data blocks, a 32 TB QLC SSD, if written to capacity once per day, would take over 11.5 years to reach its endurance.

Lower TCO: 16 KB High Mapping Granularity
For QLC SSDs, in addition to the high-endurance J5000, DapuStor also introduces J5060, a 16 KB granularity mapping version, offering a maximum capacity of 61.44 TB. This version is well-suited for clients with large sequential data processing needs. This high mapping granularity technology further underlines the cost advantage of DapuStor QLC SSDs, effectively reducing the TCO (Total Cost of Ownership).



Why J5060 Excels: Principles of 16 KB Mapping Granularity
When host IO writes are smaller than 16 KB (e.g. 4 KB), the FTL reads an additional 12 KB to combine with 4 KB, forming a 16 KB block for NAND write operations, causing a 4x write amplification factor (WAF) that impacts SSD performance and endurance. In contrast, when the host writes size matches or a multiple of mapping granularity (as the graphic shows), IOs can write directly to physical blocks with no write amplification. Although this requires advanced upper-layer management, it provides substantial benefits: the 16 KB mapping version requires only a quarter of the DRAM needed by the 4 KB version at the same capacity, breaking through capacity limitations.

Currently, the DapuStor QLC SSD J5 Series, with capacities of 32 TB and 64 TB, are officially available. For detailed product information, please visit en.dapustor.com/product/14.



View at TechPowerUp Main Site | Source

 

[Bwaze]

Prices in the range from "if you have to ask, you can't afford it" to "several internal organs"?

Meanwhile in consumer space we're getting SSD series that end at 2TB, and they're telling us that's "vast", "yuge", etc...

 

[Patriot]

Prices in the range from "if you have to ask, you can't afford it" to "several internal organs"?

Meanwhile in consumer space we're getting SSD series that end at 2TB, and they're telling us that's "vast", "yuge", etc...

just get a m.2 to u.2 adapter and join the fun. Pick up a 1 gen old enterprise drive for about the same as current consumer drives... without the endurance worries... and consistent performance.

 

[maxfly]

Was just reading about enterprise 61tb drives that run $7k iirc. Only gleaned through the thread so don't recall the particulars unfortunately.

 

[Octavean]

just get a m.2 to u.2 adapter and join the fun. Pick up a 1 gen old enterprise drive for about the same as current consumer drives... without the endurance worries... and consistent performance.

Got link?

Not the M.2 to U.2 adapter. I'd like to see a link to 1st gen high capacity Enterprise SSD drives at consumer prices.

 

[Patriot]

Got link?

Not the M.2 to U.2 adapter. I'd like to see a link to 1st gen high capacity Enterprise SSD drives at consumer prices.

Previous gen, not 1st gen.... 1st gen would be pcie 2? those aren't worth anything.
Decide what capacity you want. 3.84/6.4/7.68 TB, what endurance, 1/3 dwpd, your performance requirements, 3/6GB/s and start looking for the drives on ebay.
Always get smart data, never get sub 100% drives, its not linear wear. 3dwpd drives are the safest, avoid drives from china due to chia.

I run a pcie gen4x8 pm1735 6.4tb that can be had for $500-700ish though I only see the gen4x4 u.2 version presently for that. Know what you want, look and be patient and use best offer. I got mine for 500.
If you want cheaper... go gen3. You can get crazy capacities and high endurance. Especially if you are willing to deal with crazy form factors...

SK Hynix Sampling New PCIe 4.0 96L SSDs, 128L 4D NAND Enterprise SSDs

www.anandtech.com

sk hynix 16tb ssd nvme ruler e1.l server enterprise PE8111 | eBay

Used, working when pulled, low hours, can provide photos of inside chips, thanks

www.ebay.com

 

[unwind-protect]

WTF is an "eSSD"?

 

[Wirko]

WTF is an "eSSD"?

It's a SSD with eQLC inside. But eQLC sounds so ridiculuous that everyone's trying to avoid saying it.

Then again, I'm sure there are many, many use cases in the datacenter where the actual use is close to write-once, so 300 guaranteed full rewrites would be sufficient if read speed is high, power consumption is low, and reliability is good.

 

[GabrielLP14]

Seems interesting, i'll check it out what's the hardware inside it.
Might be using their Dapustor DPU616 controller, i don't know, but it'd be my guess.

 

[Scrizz]

WTF is an "eSSD"?

enterprise SSD. It has nothing to do with the NAND type of the SSD. The term eSSD is more common with the Asian SSD vendors.

DapuStor and Memblaze Target Global Expansion with State-of-the-Art Enterprise SSDs

www.anandtech.com

Samsung to unveil industry’s highest-capacity eSSD in November - KED Global

Samsung Electronics Co., the world’s largest memory chipmaker, plans to unveil the industry’s highest-capacity enterprise solid-state drive (eSSD),

www.kedglobal.com

Seems interesting, i'll check it out what's the hardware inside it.
Might be using their Dapustor DPU616 controller, i don't know, but it'd be my guess.

What are your thoughts on noisy neighbor testing for enterprise SSD reviews?

https://www.microncpg.com/content/dam/micron/ssd-products/9400/flyer/technical-brief/micron_9400_ssd_tames_noisy_neighbors.pdf

Maybe even some Flexible data placement stuff in the future.

https://files.futurememorystorage.com/proceedings/2024/20240806_FARP-101-1_Hands.pdf

 

[A&P211]

Prices in the range from "if you have to ask, you can't afford it" to "several internal organs"?

Meanwhile in consumer space we're getting SSD series that end at 2TB, and they're telling us that's "vast", "yuge", etc...

Just my TPU picture porn collection is almost 2tb.

 

[chrcoluk]

I am a little confused from the marketing material.

Is it physically 4k and can do 16k via emulation, which would then cause rewrite operations, write amplification and drop in performance (looks like it from the performance spec).
Or is it physically 16k, and can emulate 4k which would cause the above.
Or is it something else?

 

[Scrizz]

I am a little confused from the marketing material.

Is it physically 4k and can do 16k via emulation, which would then cause rewrite operations, write amplification and drop in performance (looks like it from the performance spec).
Or is it physically 16k, and can emulate 4k which would cause the above.
Or is it something else?

As mentioned in the post, these drives are meant for read-intensive applications. Think a CDN (Content Delivery Network), Netflix that sort of thing.
In these cases, the larger IU(Indirection Unit) doesn't affect the read performance at all. Adding to that, when the data is written to these drives it is written sequentially and the IU would be "full" before writing to the drive. You might ask why would someone use a larger IU size. One reason is there are less "chunks" to keep track of for the logical to physical mapping(indirection table). All SSDs have to keep track of what LBA(Logical Block Address) is stored in each location in NAND. That mapping has to be stored somewhere, a lot of times it's in SSD DRAM. The larger your drive is the larger the mapping is which in turn means you have to spend more money on DRAM for larger drives. One way companies can lower those costs is by implementing larger "chunks"(IU). Hopefully that helps you understand a little bit of the background of these things.

 

[chrcoluk]

As mentioned in the post, these drives are meant for read-intensive applications. Think a CDN (Content Delivery Network), Netflix that sort of thing.
In these cases, the larger IU(Indirection Unit) doesn't affect the read performance at all. Adding to that, when the data is written to these drives it is written sequentially and the IU would be "full" before writing to the drive. You might ask why would someone use a larger IU size. One reason is there are less "chunks" to keep track of for the logical to physical mapping(indirection table). All SSDs have to keep track of what LBA(Logical Block Address) is stored in each location in NAND. That mapping has to be stored somewhere, a lot of times it's in DRAM. The larger your drive is the larger the mapping is which in turn means you have to spend more money on DRAM for larger drives. One way companies can lower those costs is by implementing larger "chunks"(IU). Hopefully that helps you understand a little bit of the background of these things.

It is specifically mentioning write amplification in the material is at 4x in one of the modes, thats what my question is centred around. I understand the mapping and filesystem advantages of larger chunks. I still appreciate the answer.

 

[Scrizz]

It is specifically mentioning write amplification in the material is at 4x in one of the modes, thats what my question is centred around. I understand the mapping and filesystem advantages of larger chunks. I still appreciate the answer.

Everything I mentioned was for systems on the SSD itself not the filesystem. Writing at anything smaller than the IU size will result in higher write amplification/wear. The drive has to do a read, modify, write for anything lower than the IU size(16k), 4x in the case of 4k writes. It's essentially the same as unaligned writes. It would be disappointing if this company's drive architecture doesn't have a write buffer for those writes that are smaller than the IU to pack them into the neat 16k chunks before writing to the NAND.

 

[Patriot]

Everything I mentioned was for systems on the SSD itself not the filesystem. Writing at anything smaller than the IU size will result in higher write amplification/wear. The drive has to do a read, modify, write for anything lower than the IU size(16k), 4x in the case of 4k writes. It's essentially the same as unaligned writes. It would be disappointing if this company's drive architecture doesn't have a write buffer for those writes that are smaller than the IU to pack them into the neat 16k chunks before writing to the NAND.

I am having flashbacks... Customer company using ext2 (ssd unaware) Rhel3 with a Rhel6 kernel... and thrashing the read optimized Micron m500 drives with unaligned writes.
They ran and hid, slowed down to sub 100 iops. We found every bug micron made in its firmware... lol.

Every time I hear unaligned writes... I see those shite drives...