Lightbits Labs is working with Intel to make its NVMe/TCP all-flash arrays almost as fast as RoCe and InfiniBand options, which require much more expensive cabling.
And Intel Capital has plunked an undisclosed sum into the startup to help make it happen. Eran Kirzner, Lightbits Labs CEO, issued a quote: “We are excited to partner with Intel Corporation, and our joint solutions will set the bar for generating new ROI metrics for enterprise and cloud customers.”
Lightbits and Intel have demonstrated NVMe/TCP with a 146μs access latency by using Intel’s Ethernet 800 Series network adapters with application device queues (ADQ) technology.
ADQ sets up application specific queues so that high priority packets get dealt with faster. The companies say ADQ enables NVMe oF/TCP to achieve distributed storage performance in the same range as RDMA-based protocols, meaning 100 – 120μs.
The test system also showed up to 70 per cent increase in IOPS at high queue depths. A white paper describes the testing in much more detail.
The topology Lightbits used in performing NVMe/TCP with ADQ cluster testing.
Composable
The two are so pleased with this that Lightbits will optimise its LightOS to work with an all-flash array using five Intel technologies:
Optane persistent memory
Intel 3D NAND SSDs based on QLC flash
Xeon Scalable processors with artificial intelligence (AI) acceleration capabilities
Ethernet 800 Series Network Adapters with Application Device Queues (ADQ) technology.
Intel’s FPGAs to handle dedupe and compression
The idea is that Optane DIMMs will make the array operations faster, QLC SSDs will send capacity up and cost/GB down, and the AI-optimised Xeons will speed machine learning and similar work. The LightOS and array will provide a composable and disaggregated storage system.
Remi El-Ouazzane, Intel Data Platforms chief strategy and business development officer, said: “The data centre is being transformed, with disaggregation and composability of resources being essential to meet the efficiency requirements needed to address the explosion of data.
“Our differentiated hardware capabilities coupled with Lightbits innovative NVMe over Fabrics software gives our joint customers an exceptional economic solution to address this strategic inflection point.”
Neither Intel nor Lightbits is providing any details on what composability will mean in practice.
Micron’s latest quarterly results show it climbing out of a revenue trough, on the back of strong sales of quad-level flash. But the pandemic and the US trade ban on Huawei are expected to put a crimp on growth next quarter
The US chipmaker reported revenues of $6.06bn for the fourth fiscal 2020 quarter, up 24.4 per cent on the year-ago $4.87bn. Net income was $988m, up 76 per cent on last year’s $561m.
Sanjay Mehrotra
Micron CEO Sanjay Mehrotra noted “strong DRAM sales in cloud, PC and gaming consoles and an extraordinary increase in QLC NAND shipments.”
In the earnings call he said “economic recovery from the sharp recession in calendar Q2 is under way, but the pace has been limited by the continuation of the pandemic. … our short-term outlook has weakened due to a combination of factors.”
He cited the pandemic and US government restrictions on exports to Huawei, which had accounted for 10 per cent of Micron’s Q4fy20 sales.
Micron sees the market improving throughout calendar 2021, driven by 5G, cloud and automotive industry growth, but its own revenue growth is almost slowing to a halt next quarter. It thinks it will see revenues of $5.2bn plus or minus $200m – up one per cent.
Also Micron said there was a risk of NAND over-supply in 2021 as suppliers could build too much manufacturing capacity.
By the numbers
Full fy2020 revenues were $21.4bn, compared to fy19’s $23.4bn, down 8.5 per cent, with profits of $2.69bn, down 57.4 per cent on the year-ago’s $6.3bn.
Financial summary:
Gross margin: 34.9 per cent
Free cash flow: $111m
Operating cash flow: $2.27bn vs $2.23bn a year ago.
Net cash: $2.613bn
CAPEX: $2.16bn vs $2.23bn a year ago
Micron has two basic product lines; DRAM and NAND. DRAM revenues in the quarter were $4.37bn, up 29 per cent from the year-ago $3.39bn. NAND revenues of $1.53bn, up 27 per cent Y/Y.
There are four business units selling DRAM and NAND and Wells Fargo senior analyst Aaron Rakers provided numbers to his subscribers;
Computer and Networking BU – $3.02bn revenues up 59 per cent Y/Y with greater than doubled bit shipments to the cloud customers as on-premises demand declined with weak enterprise server demand.
Mobile BU – $1.46bn revenues up 4 per cent Y/Y. 5G phone adoption will drive the market higher.
Embedded BU – $654m revenues down 7 per cent Y/Y with a predicted recovery in the automotive market coming.
Storage BU – $913m revenues, up 8 per cent Y/Y as consumer SSD sales grew to a record and data centre SSD revenue grew strongly.
The pandemic appears to have cut car sales which affecte revenues for Micron’s embedded BU.
Dell EMC today updated the high-end PowerMax storage array to run more VMware Virtual Volumes (vVOLs), migrate data to the public cloud and improve disaster recovery capability.
vVOLS and SRDF/A
With today’s update, VMware Site Recovery Manager (SRM) 8.3 can access a PowerMax array replication feature called SRDF/A, to automate VM movement between sites. SRDF maintains real-time or near real time copies of data on a production storage array at a remote site. It can work synchronously or, with the/A version, asynchronously.
A disaster recovery version of SRDF – SRDF/Metro – pairs the remote site target with the primary site. The primary and secondary site arrays may appear as a single virtual device for presentation to a single host or host cluster. If one of the primary arrays is unavailable, a new Smart DR facility maintains data resiliency by copying data from both primary arrays in a cluster to one remote array.
A new CloudMobility feature enables a PowerMax array to move, via snapshots, data to the public cloud or on-premises object storage for cheaper long-term retention. CloudMobility targets are AWS, Azure and Dell EMC’s ECS object storage system. This is a two-way system and data can be returned to the PowerMax array. Using a Dell EMC vApp, PowerMax admins can move snapshot data from S3 object storage to Amazon’s elastic block storage (EBS).
Dell EMC has added Thales end-to-end encryption software to PowerMax arrays.
All these new PowerMax features are generally available.
Dell Technologies has added a raft of VMware-support across its Cloud Platform, VxRail, ObjectScale and PowerProtect systems.
The updates, announced today at VMworld 2020, include:
Cloud Platform and VxRail hyperconverged systems now support the latest VMware Cloud Foundation, vSphere, vSphere with Tanzu and vSAN releases. VxRail now supports cloud-native workloads via Tanzu.
PowerProtect Data Manager backup has added support for Tanzu, VMware’s software to run containers and virtual machines side-by-side.
VMware Cloud on PowerEdge servers now support remote workforces with VMware Horizon virtual desktop software, gets a larger node, new compliance certifications, and supports VMware HCX workload mobility.
The new ObjectScale software, directly deployable through vSphere, is available for preview in an early access program.
Dell Technologies Cloud Platform (DTCP) is an integrated VMware Cloud Foundation and VxRail offering, and available with subscription pricing. VMware Cloud Foundation combines vSphere (compute), vSAN (storage), NSX (networking), and vRealize Suite (management).
There is tighter integration between VMware Cloud Director and DTCP, which reduces time to provision and manage multi-tenant cloud environments. According to Dell, cloud service providers can create virtual data centres in minutes and deploy them in seconds.
PowerProtect
PowerProtect Data Manager (PPDM) now protects Tanzu Kubernetes clusters, Kubernetes clusters in vSphere, Tanzu Kubernetes Grid (TKG) and Tanzu Kubernetes Grid Integrated (TKGI) on-premises and in public clouds.
PowerProtect and Tanzu
PPDM admin staff use vSphere workflows to assign data protection policies. PPDM has a new snapshot and data mover feature, to prevent the pausing of a VM during backup, in preview. PowerProtect VM backup stops being disruptive and, when enacted, this will ensure all VMs are protected and active all the time.
Cohesity is moving into disaster recovery (DR), with SiteContinuity, an automated offering that integrates with its data management software.
SiteContinuity converges backup, continuous data protection, and DR, thus eliminating the need for separate product purchases. Zerto, a DR specialist, has already seen the appeal of this and expanded into backup in April last year.
A Cohesity spokesperson told us: “Enterprise IT commonly has to manage anywhere from three-seven application tiers, in multiple locations, with different service levels – all of which are mission-critical, and have a recovery time objective SLA of a maximum of 15 mins downtime allowed. … IT departments often rely on different applications to … meet SLAs. For example, a mission-critical app may rely on VM-based replication (a bit like Zerto), and a business-critical application and low tier apps are protected by backup replication.”
Cohesity thinks its single product will be cheaper and easier to manage.
SiteContinuity
Steve Culy, senior systems engineer at Navis, a Cohesity customer, issued a supportive quote: “The new solution from Cohesity is very timely as it allows us to protect our mission and business-critical applications on a single platform. We can now automate our business continuity and disaster recovery strategy, all from Cohesity’s data platform, making it easier for us to manage SLAs, and reduce downtime with automated, rapid failover and failback.”
SiteContinuity announcer
SiteContinuity enables users to recover a single file, whole application, or an entire data centre with a few clicks and near-zero downtime.
Journal-based recovery, based on immutable snapshots, can restore to any point in time before a disaster struck. At failover and/or restore time, Cohesity Helios’s machine learning algorithm helps identify a clean point in time to restore, based on data size and/or type changes. This helps protect the DR site from ransomware and other malware.
SiteContinuity is managed through a global UI and has a unified policy framework that applies policies across application tiers, service levels, and environments. Its architecture enables web-scale scalability. The SiteContinuity DR capability can be tested with end-to-end automated non-disruptive disaster recovery tests, which include audit trail reporting.
SiteContinuity will be generally available in the fourth 2020 quarter.
Global external storage sales in the second 2020 quarter were flat year on year despite Covid and remained steady on the previous quarter. All-flash array revenues grew four per cent Y/Y to take up 45 per cent of the pot, while disk drive and hybrid HDD/SSD array revenues fell three per cent, according to Gartner number-crunchers.
Cutting the numbers another way, secondary storage revenues climbed 44 per cent Y/Y to account for 16 per cent of the external storage market. Backup and recovery storage array revenues grew one per cent while primary storage declined six per cent.
Gartner estimates storage vendors collectively shipped 11.86 exabytes in Q2, a jump of 33 per cent Y/Y. This includes a 35 per cent increase in HDD and hybrid shipped capacity and a 19 per cent rise in shipped AFA capacity. Shipped primary storage capacity increased 19 per cent.
Wells Fargo senior analyst Aaron Rakers provided subscribers with charts based on the Gartner report, and we reproduce a couple here. The first one shows supplier share trends.
Chart based on Gartner external storage report
The second chart shows revenue share by vendor.
Chart based on Gartner external storage report looking at Q2, 2020.
Three companies grew market share: Huawei, up 72 per cent to 10.8 per cent share; IBM up 30 to 35 per cent (our estimate) to 16 per cent share; and market leader Dell EMC, up four per cent to 27.3 per cent share.
NetApp, HPE, Pure Storage and Hitachi have each suffered two quarters of external storage revenue decline, roughly coinciding with the Covid-19 pandemic. NetApp fell 4.4 per cent to 17.7 per cent share; Pure Storage declined five per cent to 11.2 per cent share, and HPE fell 25 per cent to 7.8 per cent share.
Kioxia, the world’s second-largest NAND chip maker, has called off its planned $3.2bn IPO, citing “continued market volatility and ongoing concerns about a second wave of the pandemic”.
In other words, the US government has thrown a giant spanner into the works with its export ban on the use of US-owned technology in semiconductors sold to China. Investors are sitting on their hands until they can get a better picture of the impact on Kioxia revenues.
Kioxia originally planned the IPO for October 6. Kioxia was initially valued at $20bn but this was later downgraded to $16bn, according to reports. Presumably the pricing was revised in light of investor feedback.
Kioxia is the new name for Toshiba Memory Holdings, which was sold in 2018 to a Bain-led consortium, with Toshiba retaining 40 per cent. This was a forced sell-off, prompted by enormous losses at Toshiba’s US Westinghouse Electric subsidiary, which built nuclear power stations.
Spectra Logic’s TFinity tape library can store an exabyte of uncompressed data with LTO-9 tapes, and is much cheaper than AWS Glacier Deep Archive, the company claims
Spectra Logic estimates a TFinity library with 1EB of LTO-09 data has a total accrued cost over five years of $8.05m whereas a AWS Glacier Deep Archive costs $12.12m over the same period.
Extending the storage period to 10 years results in a TFinity library total accrued cost of $8.55m. But AWS Glacier would cost an amazing $72.7m at that point.
The TFinity scales out to 45 frames or large racks and could store an exabyte of compressed data in 2016 with IBM TS1150 tapes and drives. The arrival of LTO-8 in July 2018 enabled the system to store an exabyte of compressed LTO-8 data. Now 18TB LTO-9 tapes bring the uncompressed exabyte into view, using 55,990 cartridges.
TFinity library interior
Customers would probable compress the LTO-9 data so we are looking here at a 2.5EB library.
This requires a lot of data centre floorspace – a 45-frame library occupies an area 109 feet by 3.5 feet deep.
Storing an exabyte of data costs millions of dollars, however you do it. But tape is cheaper than hard disk. Disk drives cost more than tape cartridges, and need power and cooling. Spectra suggest a 4PB disk array would need the same amount of power as a 1EB tape library.
Tables extracted from Spectra’s Exabyte storage eBook.
In an eBook, Spectra says TFinity users would start small – with a three-frame system, say – and then bulk it out, adding frames and drives as data volumes grow.
With the exception of a few hyperscalers, the notion that enterprises will store an exabyte of data on-premises seems fanciful. However, the LTO tape roadmap extends to LTO-12 with 144TB of raw and 350TB of compressed data. This means that 6,999 cartridges could store 1EB – an eighth of the LTO-9 number. The library would be correspondingly smaller and so more affordable.
Ransomware rescue! HYCU has helped save a customer from a ransomware attack and DataDobi moved a petabyte of data off an end-of-lifed Isilon array to newer NetApp kit. The hubbub of background events in the storage industry is constant – as you will see from the shorter news bites.
HYCU protects French company in ransomware attack
An unnamed French construction equipment retail and rental company was hit by a ransomware attack, while in the process of transitioning from its legacy three-tier VMware environment to a new Nutanix hyperconverged infrastructure (HCI) using AHV and HYCU for backup and restore.
On a Sunday morning, the company’s IT manager started receiving alerts indicating strange activity. The servers were all encrypted with a cryptolocker, and the cyber thief demanded a ransom of several hundred bitcoins, equal to hundreds of thousands of Euros.
The IT manager said: “While we use several types of security software, the virus apparently entered via a computer under configuration and not ready for production and it was able to propagate throughout the environment from there.”
The company’s servers were shut down and the IT manager called HYCU support. HYCU’s Fast Restore feature keeps local snapshots on the Nutanix cluster. This snapshot was not compromised by the cryptolocker, so it offered a restore point for the Nutanix environment.
He said: “The HYCU team has extensive knowledge of the Nutanix REST API, so they were able to recreate our VMs with Nutanix snapshots using API request.” They had the company’s Nutanix infrastructure up and running normally within five hours. That was in time for the work week, avoiding business disruption.
This HYCU customer did not pay the ransom.
UMass uses DataDobi for Isilon to NetApp switch
UMass Memorial Health Care data centre in Massachusetts, uses a Dell EMC Isilon scale out network-attached storage platform to manage a variety of data, including medical applications and patient records. Earlier this year the organisation planned was to migrate the data from the Isilon and onto new, higher performing, and more secure NetApp storage.
Kevin Davis, Senior Storage Engineer, Systems Administration, and Information Service, said: “The Isilon had over a petabyte of NAS data, but any engineer that is confronting having to move data across platforms, let alone this much data, understands the problems that can arise can be insurmountable.”
He said: “The data is complex — there are countless disparate apps being used by the doctors and departments across the UMass Memorial Health Care network.” CIFS. SMB and NFS protocols were involved. To complicate things, the Isilon had been “end-of-lifed, meaning updates, support, and services are now virtually non-existent. So, not only is it much more expensive to support this gear, it can put data at risk.”
“I looked at NetApp XCP software, but their CIFS tool is lacking. And I had previous experience with Robocopy and rsync, but I just didn’t have the time for all the planning, trouble-shooting, and scripting that I knew would be involved with those tools. Moreover, Robocopy and rsync lack reporting capabilities, except for some very basic logs.”
He used DataDobi’s DataMigrate product instead and was pleased with the result: “The initial replication is fast, but after DobiMigrate completes it, it is ridiculously fast. You can open as many threads as your network can handle,” he said.
“Also, it is able to plow through metadata and discover changes in our databases in an incredibly fast and efficient manner. And any kind of error is also a cakewalk. DobiMigrate manages it all, even when users add something as foreign as emojis in the file name. Yes, that actually happens.”
Shorts
Security-focussed data protector Acronis and the Schaffhausen Institute of Technology (SIT) today announced the launch of a Roborace team, the Acronis SIT Autonomous Roborace Team, with Acronis being the Official Cyber Protection Partner. This is the world’s first driverless electric racing car competition.
SaaS in-cloud backup supplier Clumio said it has experienced an eight-time increase in platform customers protecting more than one data source across private cloud, public cloud and SaaS compared to seven months ago. It has tripled employee headcount, hiring more than 50 per cent of its total company roster during the pandemic. The company protects four data sources; AWS EC2, EBS and RDS, VMware Cloud on AWS and Microsoft 365.
DH2i today announced the availability of DxEnterprise for Availability Groups in the Microsoft Azure Marketplace. This enables Microsoft SQL Server availability groups (AGs) to be made highly available within and between Windows and Linux nodes and across any type of infrastructure.
DigitalFilm Tree (DFT), a creative supplier to the world’s media, tech and entertainment companies, is using Excelero’s NVMesh software as the centrepiece of a new storage architecture. DFT CTO Thomas Galyon said: “Excelero will let us store any file format we throw at it, and run on any server, with 10x faster render processing and at least 100x greater total aggregate bandwidth than our previous software. The added bandwidth lets us send project content out to clients far faster.”
IBM has published a Red Book entitled “Deployment and Usage Guide for Running AI Workloads on Red Hat OpenShift and NVIDIA DGX Systems with IBM Spectrum Scale” It shows how one ESS3000 NVMe box running Spectrum Scale can more than handle the workload of two DGX-1 Systems running 16x V100 GPUs on a real world AI workload.
Kasten has integrated the K10 Kubernetes data management platform with VMware vSphere and Tanzu Kubernetes Grid Service.
This week’s Microsoft Ignite 2020 saw the company bringing Azure Kubernetes Service (AKS) to Azure Stack HCI. A preview is available. With AKS, developers can deploy and manage containerised applications with Azure Stack HCI in the same way as they would in Azure.
NAKIVO Backup & Replication v10.1 is now available. It has HPE StoreOnce Catalyst integration and a OneDrive for Business Backup feature protects OneDrive user accounts. A Free edition gives businesses the opportunity to use Backup for Microsoft Office 365 for free for one year.
NetApp has announced Elastigroup for Azure Spot Virtual Machines. This feature simplifies and automates compute infrastructure provisioning and autoscaling while allowing cloud consumers to drastically reduce their compute spending by up to 90 per cent, along with providing high availability.
Retrospect, a StorCentric company, announced the general availability (GA) release of Retrospect Backup 17.5, which includes new cloud storage provider certifications and platform updates for Apple macOS, as well as performance enhancements to Retrospect Management Console.
ScaleMP, which virtualizes NVMe SSDs as memory, is offering its vSMP MemoryONE software for a range of AWS EC2 instances with NVMe SSDs. Available via AWS Marketplace, vSMP MemoryONE supports bare-metal instances and the newly announced vSMP MemoryONE v10 Preview supports AWS virtual instances.
LAKE Solutions, a cloud service provider in Switzerland, has deployed Scality RING, built on Hewlett Packard Enterprise (HPE) Apollo 4000 Systems delivered with the GreenLake as-a-service consumption model, to power a new backup-as-a-service offering.
The Storage Networking Industry Association (SNIA), NVM Express and DMTF has announced the SNIA Swordfish storage management specification version 1.2.1 and the 2020.3 release of DMTF Redfish provide functionality to ensure NVM Express (NVMe) and NVMe over Fabrics (NVMe-oF) technology environments can be managed entirely by Redfish and Swordfish.
StorageOS Version 2.2 is available through the Red Hat marketplace. V2.2 has been designed for organisations that want to run hyperconverged and/or multiple cluster Kubernetes environments with the underlying storage to support cloud-native workflows and stateful applications.
WD 18TB Purple and Red Pro HDDs and 1TB microSD card
Western Digital is spreading 18TB bigness into WD Purple video surveillance and WD Red Pro NAS disk drives.
These are nine-platter conventionally recorded drives that use WD’s ePMR bias current technology to push capacity to 18TB.
18TB Purple
The WD Purple 18TB drive has AI-oriented firmware, the company says. This will improve picture quality in video surveillance to help operators better detect events and people in the image stream.
Eric Spanneut, VP for WD’s client computing and smart video business, said: “AI is driving the need for more enhanced capabilities in drives to keep up with performance demands, especially as video resolution and the number of incoming streams increase.”
The WD Purple range extends from 1TB to 18TB. The low end, the 1TB – 6TB drives, have AllFrame 4K firmware and support 64 cameras. The 8TB-18TB models have different AllFrame AI firmware supporting 64 cameras and an additional 32 streams for deep learning analytics within the system to improve video streaming quality. This is for network video recorders (NVR) applications which encode and process data at the camera. How they use this is left unsaid.
The previous WD Purple maximum disk capacity was 14TB. Drive spin speed varies with capacity. The 1TB – 4TB drives spin at 5400rpm, the 6TB one at 5700rpm, and the 8TB – 18TB ones at 7200rpm. They all have a 6Gbit/s SATA interface. Data caches are 512MB (14TB-18TB), 256MB (8TB – 112TB), and 64MB (1TB – 6TB). The sustained transfer rate varies with capacity, from 110MB/sec at 1TB to 272MB/sec at 18TB.
The 1TB to 6TB drives support a 24×7 180TB/year workload, 300,000 load/unload cycles, and carry one million hours MTBF rating. The 8TB to 18TB drives support a 24×7 360TB/year workload, 600,000 load/unload cycles and have a 1.5 million hours MTBF value.
In effect, there are two classes of Purple drive: low capacity, lower reliability, slower spinning models; and mid-to high capacity, faster spinning, higher reliability versions.
18TB WD Red Pro
The 7,200rpm 18TB WD Red Pro drive essentially uses the same mechanical components, recording methods, and interface as its Purple sister.
But it has NAS-oriented firmware and has a 5-year warranty and 1 million hour MTBF rating. The WD Purple drive has a three year warranty, and MTBF rating of 1.5 million hours. So the NAS drive is warranted to last longer but is less reliable than the WD Purple 18TB. We presume this reflects a different vibration pattern in a NAS drive bay compared with a video recording system chassis.
The WD Red Pro’s maximum sustained data rates and cache sizes are basically the same as the WD Purple drives, varying with capacity, which runs from 2TB to 18TB in 2TB increments.
1TB microSD card
WD 1TB microSD card
WD today also released the WD 1TB microSD card. This will help capacity-bound video cameras take higher resolution images.
Until today the top WD Purple microSD card capacity was 512GB. The doubling comes from upgrading to 96-layer 3D NAND. The card supports up to 500 write cycles ad operates in a temperature range of -25C to 85C. Health monitoring software enables compatible cameras to estimate the card’s remaining life.
The 16TB and 18TB WD Red Pro and WD Purple drives are expected to ship in October. The 1TB WD Purple microSD card should ship in November.
WD also has the WD Blue HDD line for PC use, which tops out at 6TB. The idea of an 18TB drive for PCs has appeal but backup times could be horrendously long and there’s an awful lot of data to lose if the drive crashes. We may never see 18TB disk drives on the desktop because of these concerns.
WD announced its first 18TB disk drive, the Ultrastar DC HC550 for data centre use, a year ago. In July this year, it introduced the 18TB WD Gold brand HDD for enterprise storage systems and servers.
Seagate is shipping 18TB Exos data centre drives and expects to ship its 18TB IronWolf NAS drive this month.
Seagate is entering the object storage business with brand new CORTX software.
The disk drive maker aims to build a developer community for the open source software and has published a reference architecture for use in a Lyve Drive Rack.
Announcing the news today at Seagate Datasphere, the company said CORTX gives developers and partners access to mass capacity-optimised data storage architectures. CORTX use cases include artificial intelligence, machine learning, hybrid cloud, the edge and high-performance computing.
The object storage market has seen two entrants in two weeks – Dell EMC has joined in with ObjectScale software.
So why does the world need another object storage software technology? Seagate’s Ken Claffey, GM for Enterprise Data Solutions, said: “CORTX brings something different to other object stores in that it will uniquely leverage HDD innovations such as REMAN to reduce the likelihood of rebuild storms, HAMR to enable the largest capacity/lowest cost per bit next gen devices, and multi-actuator to retain IOPS per capacity ratios. CORTX and the community are focused on such capabilities that are required in mass capacity deployments.”
HAMR is Seagate’s Heat-Assisted Magnetic Recording drive, due to ship at 20TB capacity by year-end, and a pathway towards 40TB HDD capacities. Multi-actuator drives have two sets of read-write heads and logically divide a disk drive into two halves that perform read/write operations concurrently to increase overall IO bandwidth.
Seagate multi-actuator drive graphic.
Lyve Drive is a series of integrated, modular data storage drives, carriers and receivers for multi-stage workflow processes.
Jacques-Charles Lafoucriere, program manager at The French Alternative Energies and Atomic Agency, an early CORTX adopter said: “CORTX can very nicely work with storage tools and many different types of storage interfaces. We have effectively used CORTX to implement a parallel file system interface (pNFS) and hierarchical storage management tools. CORTX architecture is also compatible with artificial intelligence and deep learning (AI/DL) tools such as TensorFlow.”
Gary Grider, HPC division Leader at Los Alamos National Lab, also said: “I am very excited to see what Seagate is doing with CORTX and am optimistic about its ability to lower costs for data storage at the exabyte scale. We will be closely following the open source CORTX and will participate in the community built around it, because we share Seagate’s goal of economically efficient storage optimised for massive scalability and durability.”
Toyota and Fujitsu are also early CORTX adopters.
Shipments of Lyve Drive Rack and the 20TB HAMR drives are scheduled to begin in December.
Kioxia claims direct-attached performance from network-attached devices is no longer a thing of storage architects’ dreams.
The company has planted a Marvell Ethernet controller directly onto an SSD and fitted 24 of these drives into an EBOF (Ethernet Bunch of Flash drives) chassis.
The drives are accessed as Ethernet devices, with an NVMe twist. An EBOF is simpler to deploy than a JBOF (just a bunch of flash) platform, Kioxia argues, because it needs an integrated Ethernet switch only. A JBOF, by contrast, requires a box-controlling CPU, DRAM and Fibre Channel host bus adapters.
Kioxia Ethernet SSD and EBOF concept.
The Ethernet SSD and EBOF systems are intended for applications and workloads that need disaggregated low-latency, high bandwidth and highly available storage. Kioxia is positioning the EBOF as an affordable, well-performing box – as opposed to a high-performance system – for edge computing, enterprise and cloud data centres.
Thad Omura, VP for marketing at Marvell’s Flash Business Unit, supplied a quote: “The native Ethernet SSD combined with our switches and controllers offer data centres an EBOF solution that lowers their total cost of ownership, increases performance and reduces power as compared to alternative JBOF solutions.”
Alvaro Toledo, VP of SSD marketing and product planning at KIOXIA America, said: “The Ethernet-attached storage ecosystem is an idea whose time has come. … We are enabling the true potential of NVMe over Fabrics. This opens up a new world of possibilities for cloud data centre operators, software-defined storage providers, and server and storage system OEMs.”
Kioxia needs the EBOF to help build demand for its Ethernet SSDs and is collaborating with Foxconn-Ingrasys and Accton to bring EBOF systems to market. We suspect NVMe over TCP/IP will be supported in the future, enabling the use of cheaper ordinary, non-converged Ethernet.
Ethernet SSD
In the Kioxia set up, Ethernet SSDs are addressed as NVMe devices using RoCE (Remote Direct Memory Access – RDMA over Converged Ethernet). Any NVMe-oF-capable host server can use them.
Information about the Ethernet SSD is sketchy. The devices are supplied in 1.92TB, 3.84TB and 7.98TB capacities and output 670,000 random read IOPS – equivalent to 16 million-plus IOPS per chassis.
The SSD has single or dual 25GbitE links and supports RoCE v2 RDMA, NVMe-oF 1.1 and NVMe 1.4. The drive has a Marvell 88SN2400 controller and supports IPv4 and IPv6 architecture plus Redfish and NVMe-MI storage management specifications.
We suspect it has a 20µs latency but do not know what type of NAND it uses, but suspect it’s 64-layer3D NAND in TLC (3bits/cell) format because this was the case in a demo at the FMS 2088 event by Toshiba; Kioxia’s precursor company.
The chassis is a 2U x 24-slot box supporting 2.5-inch form factor drives. Each chassis supports 2.4 Tbit/s of connectivity throughput which can be split between network connectivity and daisy chaining additional EBOFs.
Kioxia, formerly called Toshiba Memory, promoted direct Ethernet-addressed drives and an Ethernet-accessed chassis in 2018. Each drive was rated at 666,666 IOPS and the chassis achieved 16 million 4K random read IOPS from its 24 drives – claimed at the time to be the fastest random read IOPS rate recorded by an all-flash array.
Two years later that performance level looks good-ish but not great, especially when compared to Kioxia’s own CD6 NVMe SSDs using its 96-layer 3D NAND in TLC format. These have the same capacities as well as a 960GB entry-level and 15.3 TB and 30.72TB upper levels. They operate at up to 1.4 million random read IOPS across their PCIe 4.0 interface, more than twice the Ethernet SSD’s speed.
We have seen the Ethernet-addressed storage drive concept before; notably with Seagate’s Kinetic disk drive concept with the drives implementing an object storage scheme using Gets and Puts to read and write data. This technology failed to take off, partially because the drives required host application software changes.
Kioxia is sampling the Ethernet SSD with customers. There is no word on general availability for the component or the EBOF.
