Amazon Web Services (AWS) has integrated Hammerspace data orchestration to imrove its globally distributed render farm capabilities through AWS Thinkbox Deadline.
According to an AWS blog, AWS Thinkbox Deadline serves as a render manager that allows users to use a mix of on-premises, hybrid, or cloud-based resources for rendering tasks. The platform offers plugins tailored for specific workflows. The Spot Event Plugin (SEP) oversees and dynamically scales a cloud-based render farm based on the task volume in the render queue. This plugin launches render application EC2 Spot Instances when required and terminates them after a specified idle duration.
SEP supports multi-region rendering across any of the 32 AWS regions. With a single Deadline repository, users can launch render nodes in a preferred region. This offers regional choices to optimize costs or prioritize energy sustainability. For effective distributed rendering, the necessary data for a region’s render farm must be appropriately positioned.
Along with Hammerspace, AWS has introduced an open source Hammerspace event plugin for Deadline. This plugin utilizes Hammerspace’s Global File System (GFS) to ensure data is transferred to its required location.
Hammerspace setup to sync file systems across two AWS Regions
The data is categorized into render farm content data and associated metadata. Hammerspace’s GFS Anvil Nodes consistently synchronize the metadata, which is typically a few kilobytes for larger files, across the involved regions. However, the content or payload data is synchronized by Hammerspace Data Service Nodes (DSX) when needed.
The nodes operate as EC2 instances, with DSX nodes using AWS S3 for staging. By default, payload data remains in its originating region. If required in another region, the primary DSX node transfers it to the destination node. Hammerspace’s directives and rules can be configured to establish policies for this process, ensuring files are pre-positioned as required.
The AWS blog has information about how Hammerspace and Deadline handle file:folder collisions to ensure a single version of file truth.
This is an example of how using Hammerspace can avoid the network traffic load of continuous file syncing and gives Hammerspace something of an AWS blessing.
Samsung made a splash at Flash Memory Summit (FMS) 2023 by highlighting advancements such as a 256TB SSD concept, a petabyte SSD architecture with Flexible Data Placement, SSD bandwidth management to address the “noisy neighbor” challenge, and the introduction of their new PCIe gen 5 SSD.
The annual event in Santa Clara serves as the prime venue for NAND, SSD hardware, software, and CXL suppliers to unveil their cutting-edge technologies. Consistent with its showing in 2022, Samsung once again dominated this year with a significant exhibition stand accompanied by multiple presentations.
The PM9D3a SSD, equipped with an 8-channel controller, is slated for release over the upcoming four quarters. It will offer capacities ranging from 3.84TB to 30.72TB, with various form factors, including 2.5 inches. Samsung informed attendees that the PM9D3a has a sequential read bandwidth up to 2.3 times superior to the existing PM9A3 with its PCIe gen 4 interface. This results in roughly double the random write IOPS.
The PM9A3 has a sequential read bandwidth reaching 6.95GBps. This suggests a groundbreaking 16GBps for the PM9D3a – a PCIe gen 5 SSD sequential read bandwidth previously unseen. For perspective, Fadu’s Echo achieves 14.6GBps, while Kioxia’s CM7-R reaches 14GBps.
While the PM9A3 achieves 200,000 random write IOPS (and 1.1 million random read IOPS), the PM9D3a is projected to deliver up to 400,000 random write IOPS, we’re told. Kioxia’s CM7-V and CD8P drives, both PCIe gen 5, achieve up to 600,000 and 400,000 IOPS, respectively. This puts Samsung’s 400,000 IOPS figure into a competitive context.
Samsung highlighted that the drive possesses telemetry capabilities, transmitting data back to the company throughout its lifecycle, aiding in troubleshooting, though this feature was introduced at the previous year’s FMS.
Samsung had previously touched upon its petabyte-scale architecture (PBSSD) at an OCP APAC tech day in Seoul. At FMS 2022, a 128TB QLC SSD was displayed, hinting at the progression towards the petabyte scale.
The architecture features Flexible Data Placement (FDP), which enables server hosts to manage data placement within the flash blocks of the drive. Data with specific attributes can be grouped together, enhancing the drive’s longevity by minimizing the write frequency, thus reducing write amplification. Samsung demonstrated this feature in collaboration with Meta using open source software.
Another feature of the PBSSD allows drives to regulate performance for each accessing virtual machine by setting traffic limits – dubbed Traffic Isolation. This ensures no single VM monopolizes the drive’s performance capabilities, ensuring balanced responsiveness.
As its latest marker on the road to a 1PB SSD, Samsung unveiled its 256TB drive concept using QLC (4bits/cell) NAND technology. Solidigm has been a prominent advocate for this format, evident from their recent 61.44TB D5-P5336 SSD. Samsung’s message is clear: they are in the race for the long haul. Impressively, their 256TB drive reportedly consumes seven times less power compared to the energy used by eight 32TB SSDs to achieve a combined 256TB.
Airbyte has made available connectors for the Pinecone and Chroma vector databases as the destination for moving data from hundreds of data sources, which then can be accessed by artificial intelligence (AI) models. Now Pinecone and Chroma users don’t have to struggle with creating custom code to bring in data – they can use the Airbyte connector to select the data sources they want.
…
Cloud backup provider Backblaze has announced Q2 2023 revenue of $24.6 million, up 19 percent year-over-year, with a loss of $14.3 million. B2 Cloud Storage revenue was $10.8 million, an increase of 39 percent year-over-year. Computer Backup revenue was $13.8 million, an increase of 7 percent year-over-year.
In July 2023, following approval by Class B stockholders, all outstanding shares of Class B common stock were converted into shares of Class A common stock (on a 1:1 basis), thereby eliminating the company’s dual-class share structure and providing all shareholders equal voting rights. The change underscores Backblaze’s commitment to good corporate governance and being a shareholder-friendly company by simplifying the capital structure and administration processes. Q3 outlook is for revenue between $25.0 million to $25.4 million.
…
We asked Broadcom’s PR agency if Broadcom was committed to Project Monterey, the Dell-VMware-Nvidia initiative to have VMware’s vSphere hypervisor run on Nvidia BlueField DPUs. The EU Commission investigating the competitive effects of Broadcom’s intended VMware acquisition has possible doubts about this going ahead. Broadcom’s reply via its agency was: “I’m afraid Broadcom can’t answer this one. Pardon.”
…
Cohesity has appointed Eric Brown as its CFO. Brown led Informatica’s initial public offering as CFO, raising $1B in proceeds at a $10B enterprise value. Srinivasan Murari has been hired to be chief development officer. He most recently served as vice president of engineering at Google Cloud, where he was responsible for the productivity of approximately 20,000 software engineers encompassing compute, storage, networking, security, databases, analytics, AI/ML, Kubernetes, serverless, cloud console, platforms, and fleet management. Previous Cohesity chief development officer Kirk Law is now an advisor to the CEO.
…
IDC ranked Cohesity as the fastest-growing of the top 25 data protection supplier in 2022.
Veeam was next, followed by – surprise – IBM and then Commvault. Dell and Veritas exhibited negative growth.
…
DataStax announced the availability of vector search capabilities in v7.0 DataStax Enterprise (DSE), its on-premises database built on Apache Cassandra. This enables any enterprise to build applications with the power of generative AI in its own datacenter. This follows its recent announcement of vector search GA for its Astra DB cloud database. Vector search in DSE is available today as a developer preview. Get started here.
…
Dell has announced a Partner First Strategy for Storage, which designates more than 99 percent of Dell’s customers and potential customers as partner-led for storage sales. Dell is compensating Dell sellers more when transacting storage through a partner, and quadrupling the number of storage Partner of Record-eligible resale accounts for more predictability of engagement. We haven’t heard about any drop in Dell storage sales heads or their redeployment.
…
There are user complaints in a Dropbox forum that the online storage provider is throttling capacity on its costliest Advanced service offering advertised as offering “As much space as needed.”
Eg; user Nigelbb;
We have asked Dropbox what is going on.
…
SaaS data protector Druva has made it onto the 2023 Forbes Cloud 100 List, which recognizes the world’s top private cloud companies. This list is published in partnership with Bessemer Venture Partners and Salesforce Ventures.
…
Federica (Fred) Monsone, founder and CEO of storage PR company A3 Communications, has been appointed as a member of the Advisory Board for the Flash Memory Summit.
…
Google Cloud says its Chronicle CyberShield enables governments to build an enhanced cyber threat intelligence capability; protect web-facing infrastructure from cyber attacks; monitor and detect indicators of compromise, malware, and intrusions; and rapidly respond to cyber attacks to limit widespread impacts. In addition, it enables governments to raise threat and situational awareness, build cybersecurity skills and capabilities, and facilitate knowledge sharing and collaboration to raise the bar for security at a national level. CyberShield includes consulting services from Google Cloud and Mandiant to further assist governments.
…
Hitachi Vantara has achieved the Amazon Web Services (AWS) SAP Competency status.
…
HPE has accepted the resignation of EVP and CFO Tarek Robbiati, effective August 25, 2023. He’s leaving to become CEO at RingCentral. HPE has named Jeremy Cox, SVP corporate controller and chief tax officer, as interim CFO while the company and its board of directors conduct a search for a new CFO.
James Whitemore.
…
Ex-NetApp CMO and EVP James Whitemore has become CMO at iPaaS (Integration Platform-as-a-Service) supplier Celigo. Gabrielle Boko became NetApp’s CMO in January.
…
N-able, which supplies data protection for MSPs to sell to their customers, reported Q2 2023 revenues of $106.1 million, up 16 percent Y/Y, with profits of $4.5 million compared to $4.3 million year ago. Subscription revenue was $103.4 million, also up 16% Y/Y.
…
Nvidia solidified its lead in AI-focussed GPU/CPU chips by announcing its coming upgraded Grace Hopper GH200 device composed of dual Grace144-core Arm CPUs and Hopper H1000 GPUs with 282GB of HBM3e memory and 1TB of LPDDR5x DRAM. The GH200 can be scaled up to 256 units in the DGX GH200 config. It’s aimed at the AI/ML inferencing market. The upgraded GH200 commences sample ship by end of 2023 with volume production to commence in C2Q24
…
Trillion row analytics database supplier Ocient says it has demonstrated a 171 percent year-over-year growth so far this year.
…
Research house TrendFocus reports that memory suppliers are boosting their production capacity, both DRAM and HBM, in response to escalating orders from Nvidia and CSPs, with “a remarkable 105 percent annual increase in HBM bit supply by 2024.”
It believes that primary demand is shifting from HBM2e to HBM3 in 2023, with an anticipated demand ratio of approximately 50 percent and 39 percent, respectively. Demand in 2024 will heavily lean toward HBM3 and eclipse HBM2e with a projected HBM3 share of 60 percent. SK hynix currently holds the lead in HBM3 production, serving as the principal supplier for Nvidia’s server GPUs – but Samsung will catch up.
…
Storage Newsletter published Trendfocus HDD unit and capacity ships for Q2 2023, confirming the dire results recorded by Seagate and Western Digital. A table provides the numbers:
…
Middle East cloud provider G42 Cloud selected VAST Data to build a central data foundation for a global network of AI supercomputers and to store and learn from hundreds of petabytes of data. A single VAST Data cluster will support multiple ExaFLOPs of AI supercomputing capability in one of the world’s largest AI clouds.
…
Data protector Veritas has announced a Veritas Managed Service Provider Program to offer managed service providers (MSPs) a set of incentives, a range of training and enablement, and flexible pricing models to expand their offerings and capitalize on the growing demand for data protection and cloud-native cyber resilience solutions.
…
Identity security company Veza announced Capital One Ventures and ServiceNow Ventures have investments in Veza, bringing total financing to $125 million. The money will be used for product development, integrations for enterprise systems, and increasing go-to-market capacity. Veza’s product gives security and identity professionals a complete understanding of who can do what with data, across all enterprise systems so customers can find and fix permissions that are risky, dormant, or non-compliant. The product ingests permissions metadata from enterprise systems, including SaaS apps, data systems, custom applications, and cloud infrastructure. This month, Veza passed 125 integrations.
…
Wannabee bounce-back-from-disaster replication firm WANdisco will hold its AGM on August 30 with potential name change to Cirata being suggested. Sky News reports WANdisco ex-chairman and CEO Dave Richards and ex-CFO Erik Miller are refusing to return their 2022 bonus payments. WANdisco’s board wanted them repaid because it believes the bonus payments were unjustified as they were based on completely wrong 2022 revenue reporting which nearly crashed the business.
…
Weebit Nano has qualified its ReRAM module for automotive grade temperatures – confirming its suitability not only for automotive but also high-temp industrial and IoT applications.
ScaleFlux is upping its computational storage drive’s capacity from 12TB to a logical 24TB.
The company provides CSD 3000 drives featuring an SFX 3000 ASIC, a system-on-chip with twin quad-core ARM processors, hardware compute engine, and a PCIe gen 4 interface. It provides transparent compression/decompression, increases the drive’s logical capacity, and also its endurance by reducing the drive’s write storage burden. We have seen a presentation deck it will use at FMS 2023 where it is exhibiting this week in Santa Clara.
ScaleFlux FMS 2023 presentation deck slide
The company is increasing the CSD 3000’s maximum raw capacity by a third to 16TB. The SoC’s so-called capacity multiplier function extends this by up to 24 TB by expanding its logical NVMe namespace. E1.S form factor support is coming soon but is not available yet.
ScaleFlux has a PCE gen 5 interface product in development, which should arrive in 2024. The company says it is continuing to integrate its ASIC and firmware with partner products and services. There is a new focus on AI workload support, with ScaleFlux saying its compression provides more data for AI training. It cites MLCommons to assert that AI training sets are significantly compressible. It can optimize checkpoint data during training and improve handoffs between GPUs.
ScaleFlux and AI data set compressibility
It says it’s working on improving queuing and Flash Translation Layer functions and has announced system integrator partnerships coming to sit alongside an existing Nvidia GPUDirect relationship. There will also be support for new drive form factors and even higher capacity points. The ASIC is going to be developed as well, with joint-development and collaboration partners.
Comment
ScaleFlux raw capacities appear to be behind those of off-the-self SSDs. Although it is promising to increase raw capacity to 16TB and logical capacity to 24TB, customers can buy larger, 30TB, SSDs now, and even 61.44TB SSDs from Solidigm.
The company says it can boost a 16TB raw capacity to 24TB in the near future and so speed AI training workloads, while Solidigm says it provides 61.44TB . Pure Storage provides 24 or 48TB DFM drives today and has a route to 300TB by 2026. IBM has a 38.4TB FLashCore Module with on-board compression and other functions boosting it to 116TB effective capacity.
ScaleFlux has to deliver more than the benefits of a compressed small standard SSD, and its partnerships with database, silicon and other software vendors are probably key to that. Compression on its own might not be enough.
But this is a capability that will have to be enhanced with every raw capacity increase coming from the NAND suppliers, as they increase 3D NAND layer counts.
The company’s integration and partnerships with a vertical market and edge computing focus look to be vital in sustaining its growth.
We have asked ScaleFlux about these points and will update things when we hear back.
Lenovo is using Cigent software to hide PC data in secure and invisible vaults to prevent data theft.
The PC maker’s ThinkShield Data Defense, based on Cigent software, has prevention-based defenses embedded directly into secured storage devices that use self-encrypting drives (SED). This, Lenovo says, is better than software-only alternatives and also policy-based data loss prevention (DLP). Multi-factor authentication (MFA) adds an extra layer of security.
Tom Ricoy, Cigent CRO, said: “Detection-based endpoint security solutions continue to be bypassed by adversaries. … Using secured storage built into Lenovo devices and file encryption with multi-factor authentication (MFA) for file access, Cigent and Lenovo are able to help mitigate even the most sophisticated attacks.”
Lenovo commercial laptops ship with SED storage devices that follow the TCG Opal 2.0 specification. SEDs, even though the name self-encrypting drive implies so, do not in and of themselves “self-encrypt”. The encryption is kicked off by additional software; the Cigent software in this case, which is used to implement encryption (full drive or partial drive encryption) on the SEDs.
Any file protected by ThinkShield Data Defense that is attempted to be copied, opened, accessed, moved, deleted, etc. by malware or a malicious user with remote or direct physical access to the PC should be stopped by MFA. This requires that the user authenticates with their 2FA/MFA selected factor, such as PIN, or Yubikey, or Duo, etc. Malware and malicious users, lacking the selected factor, cannot authenticate as a trusted user and file access is denied.
The Cigent software can also create so-called Secure Vaults. Think of these as additional partitions/drives, like D:\ or F:\ drive on your PC in addition to your C:\, where data can be stored. When the vault is locked, the drive itself makes that portion of the storage itself completely inaccessible to the Operating System.
The data in the vault is completely inaccessible, and invisible to malware, we’re told. Lenovo insists that no tool known to man can see the data, even a hex reader, such as Win Hex, that looks at the sectors of the drive. This, it claims, prevents data theft from a Lenovo PC or laptop that uses ThinkShield Data Defense.
Find out more about the Cigent Data Defense Secure Vault here.
Panmnesia, a technology firm based in South Korea, unveiled its pooled CXL memory system at Flash Memory Summit (FMS) 2023. In a live demonstration, the system showcased performance metrics more than three times faster than an RDMA-based system while running a Meta recommendation application.
The CXL technology behind this was developed in partnership with the Korea Advanced Institute of Science & Technology (KAIST) located in Daejeon. Panmnesia has crafted a comprehensive CXL framework that includes CXL CPU, Switch, and Memory expander modules. This framework possesses 6TB of DIMM-based CXL memory. The company’s strategy involves marketing its CXL hardware and software intellectual property products to CXL system developers and manufacturers.
Dr Myoungsoo Jung, Panmnesia CEO, said: “We’re excited to introduce our innovative multi-terabyte full-system CXL framework at this year’s Flash Memory Summit. Our commitment to pioneering cutting-edge solutions is demonstrated through our CXL IP, which we believe will significantly enhance memory and storage capabilities within data centers.”
The demonstrated system has a capacity three times larger than the 2TB Samsung/MemVerge-based pooled memory system also displayed at FMS 2023.
The Panmnesia framework system chassis has two CXL CPU modules, highlighted in the diagram above, three lighter colored CXL switch modules, and six 1TB memory modules or CXL endpoint controllers, forming a single DIMM pool.
On the software side, the system runs on Linux and consists of CXL hardware device drivers, a virtual machine subsystem, and a CXL-optimized user application. Interestingly, the virtual machine software component creates a CPU-less NUMA node within the memory space.
A video slide deck shows the performance of this CXL framework system in a movie recommendation application akin to Meta’s recommendation application. This is again compared to an RDMA-based alternative where the servers involved had no added external memory.
The video starts with the loading of user and item data (during Tensor initialization) and subsequently used a machine learning model for movie suggestions. The Panmnesia system completed the task 3.32 times faster than its RDMA counterpart.
An additional benefit of Panmnesia’s system is its modularity; DIMMs within memory modules can be substituted, allowing for a scaling up of memory capacity using larger-capacity DIMMs rather than incorporating additional memory modules.
Western Digital has upgraded its OpenFlex storage solution, integrating faster dual-port SSDs and the new RapidFlex fabric bridge for improved performance.
OpenFlex is a 2RU x 24-bay disaggregated chassis designed with NVMe SSDs, offering Ethernet NVMe-oF access. It supports both RoCE and NVMe/TCP. The Data24 3200 enclosure is equipped with the Ultrastar DC SN655 drive, an enhancement from last year’s SN650 model. Both drives incorporate the WD/Kioxia BiCS5 flash with 112 layers in TLC (3bits/cell) configuration, reaching a peak capacity of 15.4TB. These drives are also compatible with PCIe gen 4.
Kurt Chan, VP and GM of Western Digital’s Digital Platforms division, said: “When combining RapidFlex and Ultrastar into the Data24, Western Digital gives organizations a transformative, next-generation way to share flash and bring greater value to their business.”
Clockwise from top left: OpenFlex Data24 3200 chassis, DC SN655 SSD, RapidFlex C2000 Fabric Bridge card, and A2000 ASIC
The SN655 provides up to 1,100,000/125,000 random read/write IOPS and sequential read/write speeds of up to 6.8/3.7GBps. In comparison, the older SN650 model achieved up to 970,000/109,000 random read/write IOPS and a sequential speed of 6.6/2.8GBps.
Both the SN650 and SN655 feature a U.3 format, come with a five-year warranty, and support 1 drive write per day. The SN655 is available in three capacities: 3.48, 7.68, and 15.36TB, whereas the SN650 was available in 7.68 and 15.56TB options.
The RapidFlex C2000 Fabric Bridge, powered by the A2000 ASIC, serves as a PCIe adapter. This device can export the PCIe bus over Ethernet and boasts two 100GbitE ports linked to 16 PCIe gen 4 lanes. An additional advantage of the C2000 is its ability to function in initiator mode, adding to its existing target mode. This feature enables clients to deploy more cost-efficient and energy-saving initiator cards in their servers, negating the need for a traditional Ethernet NIC for NVMe-oF connectivity.
With the C2000’s capabilities, the Data24 3200 can connect to up to six server hosts directly, eliminating the necessity for a switch device.
The OpenFlex Data24 3200 NVMe-oF Storage Platform, along with the RapidFlex A2000, C2000 FBDs, and the Ultrastar SN655 dual-port NVMe SSD are currently available for sampling.
Rubrik has announced its acquisition of Laminar, a data security posture management provider, for an undisclosed amount.
Update: Laminar acquisition price added. 9 April 2024.
It says the combination will integrate cyber recovery and cyber posture across enterprise, cloud, and SaaS platforms. As part of this expansion, Rubrik plans to establish an R&D center in Israel, concentrating on cybersecurity. Rubrik’s existing R&D facilities are in India and the United States.
Mike Tornincasa, Rubrik’s Chief Business Officer, said: “Rubrik and Laminar share a common vision that cyber resilience is the next frontier in data security. Laminar’s technology, ability to execute, and vision make it a perfect complement to our strategy and … roadmap.”
Established in Tel Aviv in 2020 by Amit Shaked and Oran Avraham, Laminar has secured $67 million through seed funding (Nov 2021) and two-part Series A funding (Nov 2021 and June 2022). Given its rapid growth, the acquisition price likely exceeds the company’s total funding. [Rubrik subsequently admitted in its IPO filing (2 April 2024) that it paid $105 million for Laminar; $91 million in cash and the remainder in shares, according to Calcalist.]
The combined companies will develop an integrated product, ensuring clients benefit from enhanced cyber recovery and posture capabilities.
Rubrik is no stranger to the implications of cyber vulnerabilities. In March, they experienced a data breach through the Fortra GoAnywhere managed file transfer service, attributed to a zero-day vulnerability.
Laminar’s first product, released in February 2022, emphasizes data visibility, privacy, and governance across multiple public cloud environments to mitigate data breaches and compliance issues. Their current offerings provide comprehensive data discovery, classification, and protection across major cloud service providers and data warehouse environments.
Laminar Data Catalog screenshot
In the realm of data security posture management, the focus is on detection, prioritizing based on risk, and safeguarding sensitive data. Rubrik, which experienced a breach through the aforementioned Fortra incident, is enhancing its cybersecurity measures in partnership with Zscaler to prevent unauthorized data exports. This acquisition furthers Rubrik’s capabilities in breach prevention.
Rubrik and Laminar claim the merger will:
Proactively improve cyber posture to stop cyberattacks before they happen by knowing where all of an organization’s data lives, who has access to the data, and how it’s being used.
Expand focus beyond just network and endpoint security to include cloud and data security.
Prepare for more sophisticated cyber threats with AI-driven technology.
Amit Shaked, CEO and co-founder of Laminar, said in a statement: “There is a dark side to digital transformation in the form of shadow data, and more businesses are realizing they can’t protect against what they can’t see – leaving them vulnerable to cyberattacks… The combination of cyber posture and cyber recovery will help create a cyber resilient future where organizations can take on any threat, at any stage of the attack.”
Suppliers are showcasing a 2TB pooled CXL memory concept system at the Flash Memory Summit 2023 in Santa Clara.
H3 Platform has developed a co-engineered 2RU chassis that houses 8 x Samsung CXL 256GB memory modules. These are connected to an XConn XC50256 CXL switch and use MemVerge Memory Machine X software to manage the CXL memory – both pooling and provisioning it to apps in associated servers. The Computer eXpress Link (CXL) is a sophisticated PCIe gen 5 bus-based connectivity protocol. Its v2.0 version supports memory pooling and facilitates cache coherency between a CXL memory device and systems accessing it, such as servers.
JS Choi, Samsung Electronics’ VP of its New Business Planning Team, said: “The concept system unveiled at Flash Memory Summit is an example of how we are aggressively expanding its usage in next-generation memory architectures.”
Unlike conventional DRAM DIMMs that are sold directly to server OEMs, CXL pooled memory requires CXL switches, enclosures, and management software. Industry leaders like Micron, SK hynix, and Samsung view CXL pooled memory as a method to diversify the datacenter memory market beyond the typical server chassis. Their goal is to foster a comprehensive ecosystem comprising CXL switches, chassis, and software suppliers to propel this emergent market.
MemVerge, Samsung, XConn and H3 Platform unveiled a 2TB Pooled CXL Memory System for AI
XConn’s XC50256 is recognized as the inaugural CXL 2.0 and PCIe Gen 5 switch ASIC. It has up to 32 ports, which can be subdivided into 256 lanes, a total switching capacity of 2,048GBps and minimized port-to-port latency.
According to H3 Platform, its concept chassis provides connectivity to eight hosts, with the MemVerge Memory Machine X software dynamically allocating it. Additionally, an H3 Fabric Manager facilitates the management interface to the chassis.
This MemVerge software mirrors the version used in Project Endless Memory in collaboration with SK hynix and its Niagara pooled CXL memory hardware in May. If an application requires memory beyond what’s available in its server host, it can secure additional allocation from the connected CXL pool through MemVerge’s software. A Memory Viewer service provides insight into the physical topology and displays a heatmap of memory capacity and bandwidth usage by application.
The DRAM producers are all interested in CXL because it enables them to sell DRAM products beyond the confines of x86 CPU DIMM socket. Micron, for example, is sampling its CXL 2.0-based CZ120 memory expansion modules to customers and partners. They come in 128GB and 256GB capacities in the EDSFF E3.S 2T form factor, with its PCIe Gen5 x8 interface, obviously competing head on with Samsung’s 256GB CXL Memory Modules.
Micron’s CZ120 can deliver up to 36GBps memory read/write bandwidth. Meanwhile, Samsung has introduced a 128GB CXL 2.0 DRAM module with a bandwidth of 35GBps. Although the bandwidth for Samsung’s 256GB module remains undisclosed, they have previously mentioned a 512GB CXL DRAM module, suggesting a variety of CXL Memory Module capacities in the pipeline.
B&F anticipates that leading server OEMs such as Dell, Lenovo, HPE, and Supermicro, will inevitably roll out CXL pooled memory chassis products. These will likely be complemented by supervisory software that manages dynamic memory composability functions in tandem with their server products.
SK hynix has unveiled a sample 321-layer 3D NAND chip and is in the process of developing a PCIe gen 5 interface for UFS flash modules.
At FMS 2023 in Santa Clara, the company showcased its 321-layer flash chip, noting it offers a 1 terabit capacity using the TLC (3bits/cell) format. This layer count surpasses previous benchmarks set by Micron’s 232 layers and Samsung’s 236 layers, while Kioxia and WD currently have technology at the 218-layer level. YMTC’s in-development 232-layer chip is delayed due to US tech export restrictions.
SK hynix said: “With another breakthrough to address stacking limitations, SK hynix will open the era of NAND with more than 300 layers and lead the market.” The new chip is projected to enter mass production in the first half of 2025, indicating it’s in the early stages of development.
While specifics are unclear, there’s speculation on whether the 321-layer chip consists of two separate 260-layer class chips (string stacking) or is a singular stacked device, which would pose more manufacturing challenges. Given that the SK hynix 238-layer chip is a combination of two 119-layer components, a string stacking technique seems plausible.
Earlier this year, SK hynix presented on its 300+ layer technology at the ISSCC 2023 conference, boasting a record 194GBps write speed. A white paper released at the time stated: “To reduce the cost/bit, the number of stacked layers needs to increase, while the pitch between stacked layers decreases. It is necessary to manage the increasing WL (wordline) resistance produced by a decreased stack pitch.”
No technical details are included in its latest announcement.
SK hynix’s existing 238-layer generation features a 512Gb die, but with the increased density of the 321-layer technology, this will be expanded to 1Tb. This suggests potential for greater storage capacities within compact spaces for SSDs, as well as embedded UFS-type NAND drives for mobile devices and lightweight notebooks.
SK hynix also announced its introduction of a UFS v4 chip, offering transfer speeds up to 5,800 MBps, with a UFS v5 variant, capable of up to 46.4Gbps transfer speed, currently in the works.
It is also progressing towards the development of PCIe gen 6 interface drives, offering an 8GBps lane bandwidth, a significant leap from PCIe gen 5’s 4GBps. This development, in conjunction with SK hynix’s UFS advancements and the 321-layer technology, is attributed to the rising demand from AI workloads, notably influenced by platforms like ChatGPT and other large language models.
Jungdal Choi, head of NAND development at SK hynix, said: “With timely introduction of the high-performance and high-capacity NAND, we will strive to meet the requirements of the AI era and continue to lead innovation.”
In the Gartner 2023 Enterprise Backup and Recovery Solutions Magic Quadrant, leaders appear more tightly grouped than before, with notable changes in other sections of the quadrant.
The Magic Quadrant is a two-dimensional chart with “Ability to Execute” on the vertical axis and “Completeness of Vision” on the horizontal axis. The chart is divided into four quadrants: Leaders and Challengers are in the upper half, while Niche Players and Visionaries occupy the lower half. A balance between execution and vision is indicated by proximity to the diagonal line running from the bottom left to the top right.
Comparing the current Enterprise Backup and Recovery Magic Quadrant with last year’s, several key differences emerge.
Four leading companies – Commvault, Rubrik, Cohesity, and Veritas – are closely positioned. Veeam is distinct due to its superior execution capability but relatively limited vision, while Dell maintains a consistent position from the previous year, marginally trailing the other five.
A significant shift is observed with IBM. Previously in the Challengers quadrant, it has now moved to the Visionaries section, joining Druva and HYCU, both of which maintained their positions from last year.
While Acronis was categorized as a Visionary in the previous report, it now appears as a Niche Player, albeit with an improved Ability to Execute score. In the Niche Players quadrant, Zerto (recently acquired by HPE) and Microfocus have departed, while Microsoft and OpenText have made their debut, joining Acronis and Unitrends.
Scality analysts have said that, in some use cases, disk drives can use less power than SSDs – specifically when compared with high-density SSDs when the storage is in active use.
Hard disk drives (HDDs) have physical platters that are kept spinning by electric motors and also read/write heads that are moved across platter surfaces by electricity as well. Solid state drives (SSDs), in which data writing and reading relies on electrical currents, have no mechanical moving parts and so, you would assume, need less electricity. The company says this is true when HDDs and SSDs are inactive but not when data is being accessed.
Paul Speciale
Scality CMO Paul Speciale says in a blog: “Surprisingly, our research here at Scality has found that high-density SSDs don’t have a power consumption or power-density advantage over HDD. In fact, we see the reverse today.”
The Scality research was kicked off by CEO Jerome Lecat’s rebuttal of the thinking that SSDs will kill off HDDs due to SSDs having a lower total cost of ownership. Speciale then said that Scality was looking more closely at the power, cooling and rack density aspects of the debate and that examination has resulted in the discovery that active HDDs use less power than active SSDs.
A Micron 6500 ION 30.72 TB TLC SSD rated at 5 watts (idle), 15 watts (read), 20 watts (write) per drive, and priced to be competitive with Solidigm P5316 QLC drive.
It modeled two workloads. There was a read-intensive one with 10 percent idle, 80 percent reading and 10 percent writing, and a write-intensive one with the same idle time, 80 percent writing and 10 percent reading. Speciale says: “For each workload profile, drives are assumed to be in the specified power state for the percentage of time indicated. The average per-drive power calculations for each workload profile are as follows:
Power density (write-intensive): 22 TB / 6.6 watts = 3.3 TB / watt
Note: All calculations rounded to the nearest tenth.
We charted the power consumption numbers to make the comparison clearer:
SSDs had a marginal advantage in the idle state, with HDDs needing 15 percent more power, but HDDs were 68 percent more efficient than SSDs when writing data, which resulted in a 63 percent efficiency advantage in the write-intensive workload. The HDD read advantage was less, 37 percent on a straight active read comparison and 40 percent in the read-intensive workload.
Speciale says: ”This clearly demonstrates that the perception of high-density QLC flash SSDs as having a power efficiency advantage over HDDs isn’t accurate today.”
His analysts also looked at a TB/watt power-density rating, and found that the 22TB disk drive had a 19 percent read-intensive and 94 percent write-intensive power-density advantage over the 30.72 TB SSD. Speciale reckons: “This will obviously vary with other workload pattern assumptions and is certainly subject to change as SSD densities increase in the coming years.”
Speciale also caveats the results by saying: “Moreover, there are additional considerations for enclosure-level (servers and disk shelves) density and power consumption metrics, and how the cost of power affects each customer’s overall storage TCO.”
His conclusion from all this is that: “We do not see a significant power consumption difference between the drive types (SSD vs HDD) for this to be a decision criteria.”
Rack power density counts
However, as SSDs are physically smaller than HDDs: “Based on these specs, we can see that SSDs provide a significant density advantage over HDDs. Some commercially available enclosures (storage servers and disk shelves) we see available today also can provide more than a 2X advantage in rack density for these SSDs.”
“It is important to note, however, that the amount of power that can be delivered to a datacenter rack (‘rack power density’) can be a limiting factor in many datacenters. In some cases, this rack power limitation does not support fully populating the racks with these high density servers. So ultimately power delivery can become the limiting factor in achieving ultra-high levels of rack density.”
But the amount of power that can be delivered to a datacenter rack (“rack power density”) can be a limiting factor in many datacenters. Ultimately, power delivery can become the limiting factor in achieving ultra-high levels of rack density.
Scality says it is itself agnostic in general about the HDD vs SSD choice for its object and file storage: “SSDs can deliver tangible performance benefits for read-intensive, latency-sensitive workloads.” But: “HDDs will remain a good choice for many other petabyte-scale unstructured data workload profiles for the next several years, especially where price (both $ per TB and price/performance) are a concern.”
Comment
This Scality finding casts a different light on the will-SSDs-kill-HDDs debate as it could surely change the five-year TCO calculation. Read Speciale’s blog here.
