Home Blog Page 391

Amazon buys NVMe startup E8

By
Chris Mellor
-

Amazon is buying the NVMe-over-Fabrics startup E8 Storage for $50m-$60m.

The news is reported in the Israeli Globes IT news website and the acquisition price is an estimate.

E8 was founded in 2014 by CEO Zivan Ori, VP R&D Alex Friedman, and Ziv Serlin, Director of Systems Architecture. It has raised $18m in funding in two rounds and has about 25 employees in Tel-Aviv.

Zivan Ori.

E8’sproduct technology provides NVMe-oF access to an all-flash array giving accessing applications low latency access to data at high-bandwidth. Optane (3D XPoint) media is supported.

E8 Storage array

Its software can run on servers from Dell, HPE and Lenovo. The E8 array, a scale-out, 2U, 24 x 2.5-inch slot, dual processor server system, is integrated with ThinkParQ’s BeeGeeFS clustered parallel file system to deliver files faster to high performance computing customers. 

E8 reported a SPEC SFS2014 benchmark result using Optane in August 2018 to demonstrate its speed.

Blocks & Files thinks AWS may well use E8 technology inside its own infrastructure. It may have ideas about using it in its on-premises edge devices.

When the acquisition closes E8’s Israeli team will join AWS in its Tel-Aviv facility.

Another NVMe-oF startup, Excelero, also has a BeeGeeFS integration. CEO and co-founder Lior Gal told us: “E8’s acquisition by AWS is a huge validation that NVMe-oF technologies are here to stay as the future of storage networking and innovation. We are delighted that AWS, which is always being followed by everyone, is now publicly showing commitment to scale-out NVMe using intelligent clients to achieve scale.”

Samsung feels effects of DRAM and NAND memory glut

By
Chris Mellor
-


Samsung revenues are still suffering the effects of the global memory glut but demand uplift signs are apparent.

The company today reported DRAM and NAND sales revenues down 34 per cent year-on-year at ($10.44bn for the second 2019 quarter.

Samsung’s overall sales for the quarter were $47.63bn, a 4 per cent fall y/y, with profits of $4.4bn, down 53.3 per cent from $9.57bn a year ago.

The memory products form part of Samsung’s Semiconductor business division. It does not separate out DRAM and NAND product revenues so we have no insight into their relative health.

Samsung said weakness and price declines in the memory chip market persisted as the previous quarter’s inventory reductions by major data centre customers continued to reduce sales, despite a limited recovery in demand.

The company attributed the uptick to data centre customers resuming DRAM purchases and mobile applications adopting higher-capacity product. Price elasticity increased NAND demand across all applications.

Persistent uncertainties

Samsung’s outlook for the second half of the year sees persistent uncertainties in the memory business, although demand is seen growing further as major applications adopt larger capacity product under strong seasonality.

Its memory strategies for the second half are:

  • DRAM: Focus on flexible management of product mix to respond to demand from different applications; enhance technological competitiveness via stable 1ynm ramp up 
  • NAND: Strengthen profitability by mass producing 6th generation (100+ layer 3D NAND) V-NAND and expanding supply of premium products (such as Enterprise SSDs) based on 5th generation (90+ layer 3D NAND) V-NAND.

The company did not mention production capacity cuts, a tactic deployed by South Korean competitor SK hynix.

Samsung did not elaborate on to the ongoing trade dispute between Japan and South Korea which has resulted in restricted supply of chemicals and some equipment needed in semiconductor processing. Possibly these are the persistent uncertainties it sees in the memory business.

The A-Z of Asigra and Zadara’s anti-ransomware appliance

By
Chris Mellor
-

A backup appliance that resists ransomeware has been developed by Asigra and Zadara.

The Asigra Cloud OpEX Backup Appliance comprises Asigra’s Cloud Backup V14 software with the Zadara Cloud Storage Platform, a virtual private storage array.

Effectively the array is an inbuilt high-availability target store to hold Asigra backups.

It’s sold on a subscription basis and can be deployed on premises, at a colocation facility, or in the public cloud. The Asigra appliance software converges data protection and cyber-security to counter malware attacks on backup data. 

Zadara co-founder and CEO Nelson Nahum was ebullient about the new system: “Our mutual partners and customers are raving about the Cloud OpEX Backup Appliance as it is the first of its kind to protect backup data against the ominous threat of ransomware.”

The Zadara-Asigra appliance may be the first such appliance to protect against ransomware but it’s not the first backup product to resist ransomeware.

Other backup suppliers, such as Acronis, have combined cyber security and backup software to protect against ransomeware and other malware. Backup suppliers with support for tape-based backup stores, such as Veeam and Quantum, mention the air gap between IT systems and offline tape as an effective anti-ransomware measure too.

The Asigra OPEX Backup Appliance is available through authorized partners and pricing starts at pennies per GB per month for a virtual appliance configured with 30TB of storage and the ability to scale to multiple petabytes.

Dropbox loves shingled disk drives

By
Chris Mellor
-

Dropbox has used shingled magnetic recording (SMR) hard drives in its data centres for a year now and today has let the world know how it is getting on with the technology. To cut to the chase, it loves shingled disk drives, which have saved it money and enabled it to keep up with data growth.

Dropbox adopted 14TB shingled disk drives from Western Digital in June 2018 for Magic Pocket, its in-house multi-exabyte storage system. Since then the company has used other vendors’ SMR drives and has told all a blog about its shingled disk drive experience. 

In its year of SMR drive use, Dropbox said moving to SMR has yielded

  • 20% plus savings overall compared to non-SMR disk drives,
  • It has been able to increase the density of its disk capacity, faster than the growth of data,
  • It has increased the number of disks from 60 to 100 on a single machine while maintaining the same CPU and memory.
  • Made 13 SMR-related contributions to the open-sourced libzbc SMR write library.

Dropbox said close to 40 per cent of its data will be on SMR by the end of the year. It envisages moving to 18, 21 and 24 TB drives, which tells us that Western Digital or other suppliers will introduce such SMR drive capacities.

Mingles in the Shingles

Shingled disk drives take advantage of disk write tracks being wider than read tracks. They partially overlap write tracks and so enable more tracks to be written and more data to be stored on a shingle disk compared to an ordinary drive.

Although shingled media recording (SMR) disks hold more data they are slower to write as groups of overlapping  tracks must be written sequentially in fixed 256MB zones, not just one track at a time.

Dropbox said it added an SSD staging area for data in preparation for fixed zone and sequential writes. Also “the size of the write zones we’ve set up in Magic Pocket, with 1 GB extents of data, fit perfectly with the 256 MB zones used to split up SMR drives.”

The code Dropbox uses writes directly to the disks without a filesystem, and it uses the Libzbc open source library to do this.

In its blog today Dropbox forecast “future technologies will likely use the same or similar API as SMR, where disks will be able to achieve greater densities by writing continuously to a limited number of zones of the disk at any given time. They may be microwave-assisted magnetic recording (MAMR) drives or heat-assisted magnetic recording (HAMR) drives—but they will have the same interface and we’ll be able to use the same software architecture.”

This implies multiple read:write heads working in parallel.

Formulus Black consciously uncouples from Ubuntu

By
Chris Mellor
-

Formulus Black has issued a major release of Forsa that sees the software unbundled from Ubuntu OS and adding support for Optane memory and more CPU sockets to run apps faster.

The in-memory app accelerator startup’s software re-encodes application code into Formulus bit markers, effectively deduplicating and enabling it to run in memory, as a Logical Extended Memory (LEM) unit with no swapping or paging external IO and working set data in memory, thus speeding up execution time.

Another way of looking at it is to say that Forsa amplifies the available memory up to fourfold so that 3TB of DRAM looks like 12TB.

Forsa 3

The latest release, Forsa 3.0, supports Ubuntu and Centos and has added support for Red Hat Linux. The Ubuntu unbundling does not affect performance, the company said. V3.0 also gets:

  • Cascade Lake SP and Optane DIMM support,
  • Scalable beyond 2 CPU sockets, with 4 and 8 supported and 32 being checked,
  • Logical Extended Memory stretched to support multiple nodes,
  • More granular instant copies ((Blinks) of virtual machines or LEMs,
  • Docker container support.

System image copies, called Blinks, are written to SSDs – 3TB raw memory taking seven minutes – and can be reloaded for a system recovery or as part of normal power-up, saving application load time.

Amplifying memory

in a phone briefing this week, Formulus Black COO Jing Xie told us Forsa 3.0 uses Optane Direct Mode and spreads the byte-addressable range of data across multiple CPUs without the application having to know.

Going beyond 2-socket support was important to Formulus Black, he said. Forsa’s v3.0’s app performance, with memory pooled across the processors in a NUMA (Non-Universal Memory Architecture) scheme, scaled more linearly than a NUMA system without Forsa, he added.

As well as amplifying memory, Forsa can reduce the number of CPU cores required to reach a particular performance level. This can be helpful with software such as VMware which is licensed on a per-core basis.

Formulus Black is investigating a 2-node high-availability scheme with a LEM able to failover from node to node. This involves replicating a LEM or a VM from node to node and will be good for customer-developed applications which need high-availability.

Performance boost

Cascade Lake and Optane support improved a MySQL application’s 8,025 transactions per sec on a 4-socket system. When a Forsa 3.0 version of that application ran on the same hardware; it was rated at 12,000 tps.

A customer called etherFAX was able to render documents 3.4x faster with Forsa, with average document render time dropping to 89.4ms. IO-intensive apps benefit the most. Querying a week’s average fare and tip amount for NYC TAXI using the PostgreSQL database took 320.6 secs with an Ubuntu virtual machine using NVMe SSDs. It took 4.67 secs with Forsa; 68.7x faster

Comment

Startups like Formulus Black with radically new software technology have to show customers that their software really does do what it says it can do,

It has has accumulated some case studies and performance whitepapers, such as one talking about 22.9 transactions per minute with SQL Server. 

Shipping its app as an OS (ForsaOS) was a limitation and the unbundling of the OS extends Forsa’s applicability to non-Ubuntu shops. The increased CPU socket support and Blink granularity make Forsa both more manageable and more scalable than before.

Red Hat Linux support will extend market reach significantly and Optane support should help Formulus Black again. It may also help Intel sell more Optane persistent memory as Forsa enables a chunk of Optane memory for amplification. This enables it to hold deduped applications which need more Optane capacity than is available in their raw state. A 4x Optane DIMM amplification by Forsa could be viewed as an effective Optane price cut to 25 per cent off list price.

Western Digital tunes up ActiveScale object storage arrays

By
Chris Mellor
-

Western Digital has updated its ActiveScale object storage arrays, adding larger disks and a faster operating system.

WD sells two ActiveScale systems, the entry-level and modular P100 and the larger integrated or monolithic X100. Both are controlled by ActiveScale OS and use 12TB helium-filled disk drives in their scale-out pizza box-style enclosures.

ActiveScale rack cabinets

The updated models run v5.5 of the OS and use 14TB disk drives to increase array capacity. These provide 16 per cent greater capacity than 12TB drives and cost 2.5 to 5 per cent more. Their energy footprint is the same as the 12TB drives and they can fit into existing arrays.

We can safely assume that WD will support 16TB drives in the future, to give a further capacity jump.

The X100 is up to three times faster, with up to 75GB/sec bandwidth. WD claims this is the best in the object storage industry for secondary data – aka cold files.

With a new scale-out architecture providing near-linear performance growth, the system can scale up to a 9-rack behemoth with 74PB of raw capacity – more than 5,200 disk drives.

ActiveScale OS v5.5

ActiveScale OS moves from v5.4 to v5.5 and gets up to 3x performance boost from improved data path handling. It also gets a data pipeline notification service.

It has file performance balancing, with files and objects stored in different buckets. File placement is optimised for faster IO performance, particularly with large files.

There are two management facilities, with ActiveScale Cloud Management managing an ActiveScale instance. ActiveScale System Manager manages a group of ActiveScale instances.

Geo-spreading – the copying of data between three remote ActiveScale installations – has been enhanced to add eventual consistency, via Asynchronous replication, to the existing strong consistency. When an ActiveScale site ingests data it is distributed to the other two geo-spread sites in the background. This is useful for long-distance disaster recovery.

With data pipeline notification the completion of a media ingest session can be sent to Apache Kafka in real-time and a customer application started up to process the data. Metadata indexing can also be run on completion.

According to Erik Ottem, WD’s senior director for product marketing, this is the first step to dig ActiveScale arrays deeper in workflows and we can expect to hear more in the future.

V5.5 has a golden copy feature whereby the ActiveScale data is regarded as the single version of the truth. This is farmed out to analysis and test-and-development users to ensure they use appropriate data for their activities – in essence this is similar to the copy data managers such as Actifio and Delphix.

Lastly, power users can manage v5.5 via Amazon S3 so they can run ActiveScale as they wish without involving admin staff.

Comment

Western Digital is serious about the storage array system business, updating the IntelliFlash primary storage arrays two weeks ago and now refreshing ActiveScale.

Dell, Hitachi Vantara, IBM, NetApp and Pure Storage think it is necessary to integrate an in-house object storage business with their general storage product offerings. HPE takes the view that it only needs to partner, and does so with Cloudian and Scality.

Like the mainstream storage system players, WD has its own in-house offering with ActiveScale and, so too does DDN, with its WOS offering complementing its HPC, Nexenta and Tintri storage businesses.

Standalone object storage suppliers like Caringo, Cloudian and Scality are facing a maturing market centralising on S3 and file access protocols. Their task is to enhance their differentiation as general object storage gets commoditised underneath S3 and file access.

Perhaps HPE will help one of them out by suggesting acquisition discussions.

Google Cloud serverless function object access is four times faster than AWS

By
Chris Mellor
-

AWS S3 is four times slower than Google Cloud Services when serverless functions access object storage.

So says LightStep Research in a report seen by Blocks and Files entitled “How Google Cloud Storage processes cloud functions 4x faster than Amazon Web Services.”

Auther James Burns points out that serverless functions in the cloud are stateless and get their data either from data passed to the function at call time or from an external data store accessed over a network.

He looked at a serverless Google Cloud Function in Google Cloud Services and traced the time it took to initiate a connection to an object datastore in GCS and also AWS S3, and then fetch data.

A connection request made from REST APIs secured with Transport Layer Security (TLS) involves four network round trips. There are three to set up the connection; TCP SYN, TLS Client Hello, and TLS Change Cipher Spec, and then a fourth, HTTP GET, to retrieve the data.

Total connection time in a data centre is single to double digit milliseconds. If the data is in another data centre or a cloud provider, a round trip can take tens to 100s of milliseconds.

If an existing connection can be re-used three round trips are saved and the total time is quicker. The less time taken the faster the function executes.

With AWS, Burns found the initial connection setup was followed by re-use, saving time. But with GCS the initial connection re-used an existing one (so-called connection re-use insertion) and so more time was saved.

Burns concluded Google Cloud Functions’ reusable connection insertion makes the requests more than four times faster both in-region (same data centre) and cross-region (remote data centre) accesses. Here are his numbers;

In-region total:

  • GCS – 37ms
  • S3 – 156ms

Cross-region total:

  • GCS – 216ms
  • S3 – 909ms

The takeaway here is that serverless functions referring to external object datastores should execute faster in the Google Cloud than AWS. 

LightStep Research was set up by ex-Google microservices people to provide performance management services for microservices and serverless cloud applications.

Introducing Enmotus, micro-tiering and server speed-up

By
Chris Mellor
-

Two-tier hierarchical sub-file management can accelerate slow capacity storage to fast performance storage speeds and speed up servers on the cheap.

This is what Enmotus’s Virtual SSD server software does. It combines a fast storage tier with a slower one into a singe virtual drive. The software dynamically moves high-access rate data blocks to the fast tier, giving the overall impression of a single large fast tier.

Fast storage is more expensive than slow-access storage. The main benefit of Virtual SSD (VSSD) is that customers can buy a smaller fast storage tier than otherwise and still get overall fast access to all their data in their less expensive capacity tier of storage, Ron Macleod, EMEA VP of Enmotus, said.

In an interview with Blocks & Files this week, Macleod claimed the VSSD capacity storage layer operates at the speed of the highest performance storage devices but overall cost is up to one-fifth the cost of all-flash/memory tier equivalents. It also means that less compute capacity is required to provide a higher level of performance than from capacity storage alone.

The fast storage tier can be Optane DIMMs or SSDs, NVMe SSDs or even SAS SSDs. The capacity tier can be disk drives or SSDs, and is defined in relation to the fast tier. 

Enmotus history

Enmotus was founded in Orange County, California in 2010 by CEO Andy Mills and CTO Marshall Lee. It has received $5.6m in funding from two investors: Tech Coast Angels and Exponential Partners.

The founders recognised that file-level tiering, as seen in hierarchical storage management (HSM) and information lifecycle management (ILM) products was not granular enough. Sub-file file level tiering would be quicker than moving entire files between tiers and also optimise use of expensive fast storage.

Inside the operating system with a block-tiering engine was the best place to do this – this maximises the types of applications that can benefit from the speed-up.

This software layer, implementing a virtual drive, intercepts low-level storage requests at block level, groups of either 522 or 4K bytes, before they reach the physical drives and remaps them to a fast or slower storage tier.

Data is moved between the tiers in real time as data block access patterns change.

Virtual SSD tiering technology

According to Enmotus documentation, data usage patterns do not have random IO access across the entire logical block address (LBA) range of the data set. In most cases the active data stays within the bounds of 5-10 per cent of  the LBA of the current data. The other data can be viewed as being at rest, with no need to store it on the fastest storage.

It can be stored on less expensive media and promoted when needed. Data blocks are promoted into the fast tier and demoted with:

  • 3-tier scheme – RAM, Performance drive and capacity drive – HDD/bulk SSD
  • MicroTiering policy driven data movement that operates in real time and is user-tunable
  • Fast sub 1 μs translation overhead, < 1 per cent CPU loading when storage is balanced
  • Ability to migrate SSDs in and out of different virtual disks as application loads change

Altogether the Virtual SSD software supports up to 1PB virtual drives. Enmotus said it is not caching data on the faster drives within the total but tiering it. All the media used is persistent and RAM is not used as a tier.

Its VSSD analytics software identifies the size of a customer’s working data set, so helping the customer buy the right amount of high-performance needed. 

Enmotus said storage could be dynamically shrunk or grown, older storage retired and replaced with newer faster storage while the host computer system(s) continues to use the virtual drive storage.

FuzeDrive product

Enmotus’s FuzeDrive software for AMD and Intel combines Virtual SSD software with their hardware to move data blocks into and out of the fast tier. The company said machine learning is involved.

Various Enmotus white papers discuss using VSSD and FuzeDrive software to accelerate systems. For example, one discusses accelerating Hyper-V with Intel NAND or Optane NVMe drives or Optane DIMMs.

Dell and NetApp Enmotus solution briefs, using FuzeDrive, can be found here.

Intel H10 drive

Intel introduced the H10 drive in April 2019, which combines Optane memory and QLC (4bits.cell) flash in one device. Intel’s RST software is used to combine the separate Optane and QLC flash stores on the card and present them as a single volume to accessing applications.

This is drive-level tiering, proprietary to Intel, and limited to capacities it chooses to provide. Enmotus’s Virtual SSD virtualizes multiple drives and is not limited to Intel devices. It can also work with AMD processors and Samsung’s Z-SSD drives, for example.

Comment

The greater the speed difference between fast and slow tiers, the more data access acceleration is possible. This means SSD and disk drive tiers are a great match for VirtualSSD. NVMe and SAS or SATA SSDs benefit to a medium degree and an Optane and NAND combination benefits the least.

The greater the cost differential between the fast and slow tiers the larger the price/performance benefit to customers. Thus VirtualSSD software used to accelerate NAND drives with a high-priced Optane tier can be effective, even though the relative speed up is lower than using a SATA SSD tier accelerating capacity disk drive.

Go to market

Enmotus has developed partner deals with system suppliers to help bring its tiering software to customers. It aims to sell to small and medium businesses with a focus on low-latency data needs, data centres and financial trading.

Macleod said Dell is reselling the Enmotus technology. There is a consumer version of the software, with two drives supported, costing around $70.

He wouldn’t reveal enterprise pricing but said the cost, related to capacity and drive numbers, amortised across 24 drives was “amazingly low.”

Why hasn’t Enmotus grown to be bigger and more prominent? We see several reasons:

  • The funding total is small,
  • Partnership deals are not getting a lot of marketing promotion,
  • Ron Macleod is a consultant rather than a full-time employee,
  • Limited channel program,
  • Marketing focused on white papers and solution briefs.

There is no channel programme as such, with market development funding, grades of reseller and qualification programs, like ones we would associate with storage suppliers with ten or a hundred times as much funding as Enmotus.

But then they have the money for channel and marketing programs and Enmotus doesn’t. It looks like good technology that could do with being pushed harder.

Your occasional storage digest, featuring Druva, Microsoft, Veeam and StorONE

By
Chris Mellor
-

Take a jolly dash along the River Storage and visit Druva for a backup history lesson, Microsoft with ideas on formal data sharing, StorONE with Dell server storage as a service, and Veeam for customer growth numbers.

Druva data protection

In a July blog Druva’s W Curtis Preston sketched out five phases of major development in the history of data protection in the last 30 years:

  • Centralised network backup (Veritas, Legato, IBM, Commvault)
  • Target deduplication (Data Domain, Quantum)
  • Virtualization-centric backup (Veeam)
  • Hyperconverged data protection (Cohesity, Rubrik)
  • Cloud-delivered data protection as a service (Druva)

He is prepping a series of articles that will compare Druva with the other players mentioned above. We can’t wait.

Microsoft and data sharing

Businesses and organisations sharing data often have detailed and formal agreements as to what sharers can and cannot do with the shared data. Fr instance the agreements could concern people’s privacy, intellectual property rights or other limitations on use.

Microsoft’s Erich Andersen, chief IP counsel, has written a blog about such formal data sharing agreements between organisations.

He writes: “Often, agreements for broad data sharing scenarios are unnecessarily long and complex. We also think there is an important role for agreements that limit rights to computational use for AI.”

The company has developed three draft data use agreements:

  • Open Use of Data Agreement (O-UDA)
  • Computational Use of Data Agreement (C-UDA)
  • Data Use Agreement for Open AI Model Development (DUA-OAI)

The company invites interested people to read and review the drafts, with a view to formalising the agreements and making them available for use.

StorONE and Dell storage-as-a-service

Startup StorONE has announced its S1-as-a-Service (S1aaS) which integrates its S1 combined file, block and object storage software with Dell PowerEdge server hardware. Customers can start with a $999 per month deal for an 18TB all-flash array that delivers up to 150,000 IOPS in a 2U form factor, and has data protection included.

The hardware details:

  • PowerEdge R740xd rerver, a 2U 2-socket Xeon Skylake, all-NVMe flash drive system,
  • Active-Active configuration,
  • Raw Capacity: 6 * 3.84TB SSD SAS = 23.04TB,
  • Usable Capacity: 18.4 TB Useable,
  • Compression and Dedupe are included; StorONE counts the physical storage only,
  • Protocols: Block (iSCSI), File (NFS, SMB) and Object are included,
  • Block (FC) – one time extra cost for the HBA, no extra charge for the software,
  • Performance: 150,000 IOPS,
  • Scale: capacity – up to 60TB; performance – up to 300,000 IOPS.
Dell PowerEdge R740xd server

There is monthly billing and no cancellation fees. StorONE says its S1aaS offering is cheaper than any cloud storage service or on-premises storage solution in the market today.

Veeam customer growth rate slows

Veeam, the virtualized server backup vendor, said it surpassed 355,000 customers in the second 2019 quarter. On May 22 it said it had more than 350,000 customers and had achieved a $1bn/year revenue run rate. It also said then it was adding 4,000 customers a month. The rate appears to have slowed since then as it has added 5,000 or so customers in two months.

Most customers have taken up Veeam Availability Suite 9.5 Update 4, which has been downloaded nearly 350,000 times since its release in January.

Veeam is adding subscription pricing for all products as a business model and is also embracing the hybrid cloud. It said annual recurring revenue increased 26 per cent year-over-year in the quarter.

The partner deals announced with Nutanix and Exagrid, combining their hardware systems with Veeam software, should be followed by others in the second half of the year.

Shorts

Content management supplier Alfresco has been named a leader in a 26-criteria evaluation of 14 Enterprise Content Management (ECM) Platforms in The Forrester Wave: ECM Content Platforms, Q3 2019 report.

Israeli telecommunications provider Bezeq has replaced a Hitachi UCP and Hitachi VSP F400 all-flash array because response times were getting too liog and upgrade possibilities were limited in its 118TB data warehouse application. It’s now using Excelero NVMesh with Fujitsu servers and Mellanox 100 Gbps Ethernet switches, and getting  a 2x to 3x throughput improvement and database run times cut by up to 80 per cent

In-memory startup Formulus Black has appointed Mark Iwanowski as its new CEO, replacing Dr. Carr Bettis who stepped down as CEO and executive chairman, citing personal reasons.

In-memory supplier GridGain said subscription sales doubled during the first half of 2019 compared to the same period in 2018. Sales to new GridGain customers more than doubled during the period and total sales from new European customers grew more than 150 per cent.

Iguazio and Microsoft showcased Machine Learning Pipeline Automation on Azure Intelligent Edge for multiple industries at the Microsoft Inspire event. Demos included location-based recommendations for retail, real -time stock performance and sentiment analysis, predictive infrastructure operations and KubeFlow Pipelines at the edge.

iXsystems has expanded the FreeNAS Mini series of network storage systems with two new models and tighter integration with TrueNAS and cloud services. TrueNAS and FreeNAS Mini can be managed by True Command from a “single pane of glass”.

The NVM Express organisation has announced the release of the NVMe  1.4 Base Specification. The NVMe over Fabrics (NVMe-oF) 1.1 specification has entered into final 45-day member review. V1.4 NVMe provides improved quality of service, faster performance, improvements for high availability deployments, and scalability optimisations for data centres.

Phison Electronics will provide native support for Everspin’s 1 Gbit STT-MRAM memory in its next generation enterprise SSD controller lineup. Phison ships more than 600 million controllers annually. The MRAM support means it can protect data against power loss without the use of supercapacitors or batteries.

The RM5 has been certified for VSAN v6.7.

The SCSI Trade Association (STA) has announced the successful completion of the 18th Serial Attached SCSI (SAS) plugfest, the first for the 24G SAS storage interface standard that consists of SAS-4 and SPL-4 protocol layers. The plugfest was held June 25-27, 2019, at Austin Labs in Milpitas, California.

Amphenol Corporation, Microchip Technology, Molex LLC, Seagate Technology PLC and Teledyne LeCroy Corporation attended and provided initial specification verification support.

StorageCraft has announced its OneXafe Solo 300, a plug-and-play appliance that streams data to the StorageCraft Cloud Services for business continuity. It said it provides enterprise-class protection and recovery at an SMB price point.


Carbonite revenues soar but profits are elusive

By
Chris Mellor
-

Cloud backup storage service provider Carbonite has experienced a sudden CEO change.

Mohamad Ali is leaving the company to become CEO of the International Data Group (IDG). Chairman Steve Munford has been named interim CEO and exec chairman and the hunt is on for a new CEO.

Carbinite reported Q2 2019revenues of $121.5m, 56.4 per cent up on a year ago, boosted by the Webroot security business, bought for $618.5m in February. The company lost $11.3m (-$5.5m.)

Munford’s said the company’s “security software business performed well during the quarter, however, we continued to experience challenges in parts of our data protection business. We remain committed to capitalizing on the opportunity of combining data protection and security, while we improve the effectiveness of our go-to-market efforts and deliver on our profitability targets.”

William Blair analyst Jason Ader said “organic Carbonite sales [fell[ short due to continued execution issues in the SMB backup business”.

The chart below shows that Carbonite profitability has been challenged for quite a few quarters.

See the Webroot effect in the latest quarter’s revenues.

Mellanox switches off Q2 with bumper results

By
Chris Mellor
-

Mellanox has delivered rising revenues and profits for the second 2019 quarter.

Second 2019 quarter Mellanox revenues of $310.3m were up 15.6 per cent y/y and profits of $38.43m were up 132.5 per cent from $16.5m last year. CEO Eyal Waldman talked of record revenues and said: “We are pleased with our financial performance this quarter and the adoption of our latest 25, 50, and 100Gb/s Ethernet and 200Gb/s HDR InfiniBand products.”

He also said: “We are pleased that we’ve begun shipping 200 gigabit per second Ethernet adapters, switches, and cables to our data centre customers, and expect this to be a future revenue growth driver.”

The business produced GAAP gross margins of 64.5 per cent in the second quarter, compared to 61.4 per cent a year earlier. Operating activities generated $58.6m in cash ($46.7m). Cash and investments totalled $610.6m at quarter end, up from $552.6m three months earlier.

Fast networking is a cash generator and the need for higher switch bandwidth in the future looks solid.

Mellanox has not provided forward guidance or run an earnings conference call due to the pending Nvidia acquisition. The networking switch supplier is soon to be a fully owned subsidiary of Nvidia – with the $6.9bn deal expected to close at the end of the year. Nvidia sees a growing need for more and faster Ethernet and InfiniBand switching to deliver more data faster to its GPUs.

DriveScale gives Toshiba SSDs fungible seal of approval

By
Chris Mellor
-

Toshiba’s CD5 and CM5 NVMe data centre SSDs are certified for use with DriveScale composable infrastructure.

This infrastructure composes servers on demand from pools of available driveless servers and storage resources.

DriveScale diagram.

The DriveScale Software Composable Infrastructure uses a software control plane to compose the hardware resources, and return them to the pool when no longer needed. Flash storage resources can be configured by this software at chassis, drive and sub-drive level.

Toshiba said the DriveScale system, using an NVMe-oF link, makes its CD5 and CM5 SSDs fungible, meaning they can be re-used in composed systems. Storage nodes connected to the NVMe-oF protocol enable multiple paths from servers to the CD5 or CM5 SSDs.

Tatsuya Tanaka, VP for ecosystem and standards at Toshiba Memory America, said: “When deployed within the DriveScale Composable Platform, storage services can be delivered with greater performance, resource utilisation, economics and efficiency, all while enabling dynamic service level agreements across workloads.”

DriveScale supports several SSD vendors:

  • Intel – D4600 family
  • Micron – 9200 products
  • Samsung – PM983 and PM1725B families
  • Toshiba – KCM51 and now CD5 and CM5 products
  • Western Digital – Ultrastar DC SN200 products

The other NAND and SSD supplier, SK hynix, is still qualifying its NVMe SSDs with potential customers, with general availability expected by year-end.

Other composable system developers are Attala Systems, Dell EMC with the MX7000, HPE with Synergy, Liqid and Western Digital.

1...390391392...437Page 391 of 437

ABOUT US

Blocks & Files is a storage news, information and analysis site covering storage media, devices from drives through arrays to server-based storage, cloud storage, networking and protocols, data management, suppliers and standards. We publish news, views, opinions and analysis of storage industry events and developments, aiming to keep storage users, developers, consultants, marketeers, salespeople, support staff, analysts and everyone else involved in IT storage up to date with what’s going on.

FOLLOW US

© Situation Publishing, 2018-2024