Bechtle pulls HPE Primera array from datacenter over performance issues

By

-

April 13, 2022

German IT services company Bechtle replaced a faulty HPE Primera array in its data centre with an Infinidat equivalent when HPE support could not resolve a firmware problem expeditiously.

Bechtle, headquartered in Neckarsulm, north of Stuttgart, is a $6 billion turnover multinational IT products and services supplier with 80 offices, 12,700 employees and more than 70,000 customers. It used a Primera array as part of its hosting environment for managed services.

The firm said it first encountered a problem with its Primera storage array in 2021 which prevented it from performing properly as the workload scaled up. Talking at an IT press tour event in Israel this month, Eckhard Meissinger, Bechtle’s head of Datacenters, said there were “massive performance issues with the existing storage system” – the Primera system.

He added that the “frequency of the faults led to a loss of trust among our customers” and “we had to pay a lot of contractual penalties.”

Meissinger said he was told by Primera Level 3 support that the problem affected 1 percent of HPE’s Primera customers. HPE offered a 3PAR replacement array as a solution to the issue, but then – according to Bechtle – subsequently withdrew the offer as it suffered the same firmware glitch.

Both the Primera and 3PAR arrays rely on a proprietary ASIC to carry out functions such as zero detect, RAID parity calculation, checksums and some deduplication tasks, accelerating them far beyond the speed of software carrying out the same functions.

As the location of the fault was in firmware, it seems the Bechtle workload encountered an issue in the Primera and 3PAR ASIC subsystem.

HPE was unable to fix the problem over a number of weeks, because it was a major bug fix issue, according to Bechtle. As a result, Bechtle went out to tender for a replacement system, eventually choosing one from Infinidat. It bought a 1PB InfiniBox array.

The array became operational in January this year after three weeks of testing. Thousands of Bechtle clients use the system and – to date – it has exhibited 100 percent availability, and performance close to or equal to an all-flash system. Bechtle said that with a capacity utilization of 70 percent, the savings compared to competitors such as NetApp and Pure are about €35 per TB. When utilization is higher than 70 percent, the claimed savings are about €40-45 per TB.

*Bechtle InfiniBox read and write latency record. The mini-spikes are DRAM read cache misses*

Bechtle is now in the process of becoming a public Infinidat reference customer.

That’s quite a turnaround as Bechtle was honoured as Hewlett Packard Enterprise’s Global Solution Provider of the Year for Germany, Austria, Switzerland, and Russia in June 2019, with HPE CEO Antonio Neri present at the ceremony. Bechtle was also HPE’s DACH Solution Provider of the Year 2020.

We have asked HPE what was the nature of the problem and why it could not be fixed in a timescale that met Bechtle’s requirements. An HPE spokesperson replied: “HPE as a policy does not comment on our customers.”

MAID

By

Chris Mellor

-

April 12, 2022

MAID – Massive Array of Idle Disks to store persistent but little-used data on mostly spun-down disks. Envisage a disk array with a cache of spinning drives and a residual archive of idle, non-spinning drives. The idea is to have a faster access archive than a tape library with lower electricity draw than an array of fully spinning drives. If a data read request is not present on the cache drives, then the data is located on an idle drive and it is spun up and then the data accessed. Startup Copan used the MAID concept in its disk drive arrays in the late 2000s but crashed. Its assets were bought by SGI which persisted with the technology in 2010. It didn’t take off.

StorageMAP charts uncharted unstructured data oceans, estimates carbon footprints

By

Chris Mellor

-

April 12, 2022

Datadobi claims its new StorageMAP product can chart entire file and object storage estates, revealing a customer’s unknown data arsenal and shining light on their low-value or useless data that is wasting IT spending and hiking up the carbon footprint.

The uncontrolled accumulation of file and object data by private businesses and public sector bodies is putting them under increasing budget pressure. Filers and object systems are filling up with uncharted oceans of data.

What a business doesn’t know it has cannot be used – or even stored in the right place. StorageMAP has a file and object scanning engine that’s part of Datadobi’s migration product and produces metadata about unstructured data that can be viewed through a management GUI.

Datadobi CTO Carl D’Halluin said: “Enterprises cannot handle the pace of unstructured data growth on their own. Managing unstructured data across today’s heterogeneous environments is a specialist activity,” which needs tools such as StorageMap.

One of the first things customers can find with it is the number files and objects they have, who they belong to and where they are stored. Customers might find, for example, multi-terabyte folders belonging to ex-employees which are still being backed up – perhaps because the offboarding process didn’t work as intended.

Once the resident techies know what they have and where it’s stored, they can start assessing its value – ie, the number of accesses in the last week or month – and take action if it is on the wrong tier of storage. They can find the redundant, obsolete, and trivial (ROT) data and delete it, reducing storage costs. They can assess the risk profile of the data, and the type of action to take on it.

Storage admins can identify data that could feed analysis routines in the public cloud and move it there; or add their own tags (custom metadata) to file and objects so that they can track environmentally sensitive aspects of data storage. StorageMAP helps organizations to meet their CO₂reduction targets through carbon accounting of unstructured data storage in the cloud and in the datacenter, or so DataDobi claims.

Reducing an enterprise’s carbon footprint is becoming a more pressing concern for board members and CEOs. Knowing, for example, the carbon footprint of datacenters in different countries, and knowing where data is stored, means techies can consider moving it from a high carbon footprint datacenter to a lower one. They can also measure and monitor the carbon intensity of your unstructured data storage.

*A StorageMAP carbon footprint screenshot*

Vinny Choinski, senior lab analyst at ESG, said: “One of the biggest challenges I see IT operations facing today is the fact that they do not have a single view into the entire data storage environment or the ability to organize and restructure their data quickly and easily. Without that, it is practically impossible for IT to improve operational efficiency and to make their piles of unstructured data work for the business.”

StorageMAP is sold as a “pay-as-you-grow” model that allows customers to first understand their unstructured data environment before committing to take any necessary actions required by the business. Datadobi sells purchasable Action Add-ons such as a migration package.

Other suppliers provide software which can be used to index and locate files and objects, such as Data Dynamics with StorageX file lifecycle management SW; Hammerspace and its global file data environment; and data lifecycle management supplier Komprise.

Krishna Subramanian, co-founder, COO and president of Komprise, took issue with having separate functions within a lifecycle offering: “Analytics is key to effective data management but perhaps even more important is how it is incorporated. The industry practice of a separate analytics function that has to tag data and a separate migration module that moves data and disrupts user access is cumbersome and difficult to adopt.”

In her view: “We find one platform for analytics, automated policies derived from the analytics and transparent data management is key to exabyte-scale adoption.”

Datadobi said its functions are integrated and StorageMAP is a fully-fledged data management platform, putting companies in control of their data’s cost, carbon footprint, risk, and value.

Tiered backup storage player ExaGrid turns in Q1 numbers

By

Chris Mellor

-

April 12, 2022

Privately-owned ExaGrid has reported what it says is its sixth consecutive quarter of revenue growth, as it fights for market share with on-premises disk-based, purpose-built backup appliance makers.

The company provides scale-out, 2-tiered backup storage with a front-end disk-cache, the Landing Zone or performance tier, which writes backup data directly to disk for the fastest backups, and restores directly from disk for the fastest restores and VM boots.

Data is then moved to a long-term deduplicated Retention Tier to reduce the amount of retention storage and its cost. ExaGrid’s growth is accelerating, the company said, and it is hiring over 50 additional inside and field sales staff worldwide.

As a private firm, its profit/loss accounts are not available for public scrutiny.

President and CEO Bill Andrews said of the latest results: “ExaGrid’s continued momentum is the result of the ongoing expansion of our sales team and working with our channel partners across the globe. We continue to see that organizations are realizing that primary storage disk is too expensive for longer-term retention and that traditional deduplication appliances don’t deliver the backup and restore performance needed to ensure that data is always available.”

We’ve tabulated the numbers in each of ExaGrid’s recent quarterly results announcements:

*Note that these are ExaGrid announced numbers, hence they can be approximate, as with the active customer count, or even missing, as with the Q4 2020 revenue growth rate.*

ExaGrid claims it is consistently winning in competitive bids against:

Low-cost primary disk storage – more expensive than ExaGrid, inadequate dedupe ratios,
Dell EMC PowerProtect/Data Domain,
Veritas 5330, 5340, etc.,
HPE StoreOnce,
Flash/SSD-based systems – 5x the price of low-cost, primary storage disk

Andrews told us that some companies are pushing “flash for backup storage as that is all they have to sell. It is far too expensive for backup.”

He went on to claim ExaGrid sees “Rubrik or Cohesity when the customer is changing the backup app because they have to replace the backup app and the backup storage, at the same time, to compete. We see them in about 10 percent of the deals. ExaGrid goes in with a backup application: mostly Veeam and sometimes Commvault or HYCU if the customer is replacing both their backup application and backup storage at the same time.”

“In summary, disk is too expensive for backup due to retention and deduplication appliances are too slow and not scalable. ExaGrid’s Tiered Backup Storage… take[s] the pros of disk (ExaGrid Landing Zone) and the pros of deduplication appliances (ExaGrid repository) and offers both with scale-out storage.”

The suppliers who can beat ExaGrid on speed will be the flash-based backup systems such as Pure’s FlashBlade, the VAST Data Universal Storage system in data protection guise, Infindat’s high-end enterprise InfiniGuard system with DRAM caching and three dedupe engines, and StorOne.

In our view ExaGrid could rearchitect its front-end Landing Zone to use SSDs and potentially match the speed of the Pure, VAST, Infinidat and StorOne systems, while retaining its disk-based, globally deduped Retention Tier. The purpose-built backup appliances from Dell EMC, HPE and Veritas face speed, global dedupe and scale issues, which is where ExaGrid is hoping to make wins. They would need major architectural makeovers to match ExaGrid and systems from Pure, VAST and StorOne.

Storage news ticker – 12 April

By

Chris Mellor

-

April 12, 2022

….

In-cloud Saas data protector Clumio announced news of its Clumio Protect for Amazon DynamoDB, with granular recovery, improved audit readiness and reduced TCO for long-term backups. Clumio said it delivers cloud customers a fast setup process, turnkey ransomware protection, one-click recovery of their databases, centralized compliance reporting, and actionable insights to reduce data risks. Clumio Protect for Amazon DynamoDB can backup Amazon DynamoDB tables and items in less than 15 minutes with no upfront sizing, planning, or additional resources needed to setup in AWS accounts, Clumio added. The service will be generally available April 21, 2022.

…

Druva has gained Nutanix Ready certification for its SaaS data protection offering (DataResiliency Cloud) for Nutanix AHV HCI hybrid cloud. It covers protection for Nutanix workloads across data centers, remote offices, and Nutanix Clusters on AWS through a secure, automated, and simplified platform that requires no additional hardware or software. It works across platforms, applications, and geographies. Druva claimed it’s the only SaaS-based offering with support for Nutanix, and includes image-based backups, support for Prism Central and Prism Element, application-consistent snapshots of Nutanix AHV VMs, reduced cost with long-term retention and a CloudCache secondary storage to retain optional local backup copies for up to 30 days for stricter RTO/RPO or industry-specific compliance.

…

The Evaluator Group has announced its 2022 EvaluScale Insights for Container Management Systems for Platforms and Services. The EvaluScale Insights is based on deep research published by Evaluator Group with technology comparison and selection criteria known as EvaluScale Comparison Matrices. The EvaluScale Insights examines and ranks the vendors as Leaders, Challengers and Developing with a weighting based on the EvaluScale selection criteria and business drivers. The 2022 EvaluScale Insights Leaders: Container Management Platforms Leaders (see diagram above;)

Container Forward: Red Hat Open Shift, SUSE Rancher and D2IQ
Infrastructure Leverage: VMware Tanzu
Fast Track Win: Red Hat Open Shift, Azure Stack and D2IQ

Container Management Service Leaders (see diagram below);

Container Forward: Google GKE/Anthos Azure AKS, SUSE Rancher, VMware Tanzu
Infrastructure Leverage: Google GKE/Anthos, Azure AKS, Mirantis, VMware Tanzu
Fast Track Win: AWS EKS, Google GKE/Anthos, Azure AKS, Red Hat OpenShift, VMware Tanzu

…

An Intel VMD driver for vSphere can be used to create NVMe RAID1 setups. Intel Volume Management Device (VMD) works with the Xeon family of processors to enable additional control and management of NVMe devices. Intel VMD is similar to an HBA controller but for NVMe SSDs. Hot Plug, LED management, and error handling are some of the features available. The Intel-enabled NVMe driver for vSphere now supports RAID 1 volumes for boot and data. There’s more data here.

…

Kioxia KumoScale Software v2.0 is built around the NVM Express over Fabrics (NVMe-oF) protocol and delivers high performance NVMe flash storage as a disaggregated networked service. It includes additional bare metal deployment options, seamless support for OpenID Connect 1.0, and support for NVIDIA Magnum IO GPUDirect Storage (GDS). GDS enables a direct data path for direct memory access (DMA) transfers between GPU memory and storage, which avoids a bounce buffer through the CPU. KumoScale software behaves as a storage adapter to GDS.

OpenID Connect is an identity layer on top of the OAuth 2.0 protocol that allows Clients to verify the identity of users and session based on the authentication performed by an Authorization Server for service account permissions. V2.0 also includes CSI raw block support, and an embedded Grafana storage analytics dashboard.

…

Satori, which provides a Universal Data Access Service for cloud-based data stores and infrastructure, is adding support for three relational databases – MySQL, CockroachDB and MariaDB. Users can now control access to data stored in these databases in a streamlined, secure and non-disruptive way. Yoav Cohen, CTO at Satori, said. “As new choices like CockroachDB and MariaDB see rapid adoption within the enterprise, we also saw the need to meet the growing demand for managing access to sensitive data in these databases while realizing the vision of universal data access. Now Satori customers can hook into these popular services, alongside other leading data platforms including Amazon Redshift and Snowflake, without having to build such capabilities themselves.”

…

Vehera LTD, trading as Storage Made Easy today announced a record year of bookings, growing over 100 per cent year-on-year. This included its highest quarterly bookings ever in the fourth quarter and its largest seven figure sale. Company CEO Jim Liddle said: “During the pandemic we have seen companies shift to use on-cloud object storage in combination with on-cloud or on-site file storage. This shift has meant that companies required a flexible global namespace to enable employees to easily and transparently work with their hybrid file and object data sets wherever they reside.” SME’s File Fabric product meets that need “without any lock-in enabling companies to continue to directly access their data assets and take direct advantage of low cost analytics and other applications offered by large cloud vendors such as Amazon, Microsoft and Google.” The business is owned by Moore Strategic Ventures and did not make public it’s actual financial results.

…

StorPool Storage was listed as the 413^th fastest-growing company in Europe as part of Financial Times’ in-depth special report focused on organizations that achieved the highest compound annual growth rate in revenue between 2017 and 2020. StorPool achieved a CAGR during the queried timeframe of 69.29 percent. This was nearly double the minimum average growth rate – 36.5 percent – required to be included in this year’s ranking. StorPool Storage is a primary storage platform designed for large-scale cloud infrastructure. It converts sets of standard servers into primary or secondary storage systems and provides thin-provisioned volumes to the workloads and applications running in on-premise clouds with multi-site, multi-cluster and BC/DR capabilities .

…

VMware said it has worked closely with other tech firms, including Pensando Systems, to deliver a single infrastructure platform across silicon and OEM vendors in the DPU space under the auspices of Project Monterey. AMD’s purchase of Pensando seemingly validates VMware’s belief in the future of the data center as a distributed and heterogeneous architecture – which will continue to evolve by embracing newer types of devices and accelerators. VMware sdaid it is actively engaged with Pensando Systems to jointly work on a distributed data center architecture.

…

VMware announced VMware Cloud Flex Storage, a natively integrated, elastic and scalable disaggregated cloud storage that mitigates the scalability issues faced by users on VMware Cloud on AWS. With VMware Cloud on AWS, cost economics were less than ideal and there were challenges associated with scaling its default vSAN storage. The only way to scale is to add or remove hosts, which means the addition or removal of CPU and memory in chunks, regardless of whether it was needed or not. VMware Cloud Flex Storage adds on-demand capacity and easy cost-effective scalability that requires no more than just a few clicks. Read a Gestalt IT blog for more info.

…

Replicator WANdisco said it saw strong trading in Q122 following significant contract wins for its LiveData Migrator (LDM) solution, both directly and via key cloud channel partners including Azure, AWS and IBM, and analytics partners Databricks and Snowflake. Eg; see Canadian National Railway. Its unaudited Q122 bookings are $5.8 million: up from $1.1 million a year ago. Ending RPO (Remaining Performance Obligations) is expected to be approximately $14 million for Q122: up significantly from $4.2m a year ago. Continued high visibility of it’s near-term business pipeline underpins confidence in strong 2022 trading.

…

HPE’s data protector Zerto has published a ransomware report about an ESG study revealing that gaps in readiness are seriously impacting the ability of many organizations to manage and recover from attacks. While organizations recognize that one of the best protections against a ransomware attack is the ability to recover from it, many are still struggling to counteract ransomware when prevention has failed. So, natch, buy Zerto’s services to help. Find out more in its downloadable e-book, “The Long Road Ahead to Ransomware Preparedness.”

…

Kaseya buying Datto for $6.2 billion

By

Chris Mellor

-

April 11, 2022

Who knew cloud backup was so hot? Storage conglomerate Kaseya is buying Datto out of the hands of private equity for a phenomenal $6.2 billion.

Datto IPO’d in October 2020 with private equity business Vista Equity Partners holding 69 percent of the shares. Datto signalled it was looking for a takeover or another private equity buyer in March and a month later it has succumbed to Kaseya’s advance.

Fred Voccola, Kaseya’s CEO, said in a statement: “Kaseya is known for our outstanding track record of retaining the brands and cultures of the companies we acquire and supercharging product quality. We couldn’t be more excited about what lies before us – Kaseya and Datto will be better together to serve our customers.”

The all-cash transaction will be funded by an equity consortium led by Insight Partners, with significant investment from TPG and Temasek, and participation from other investors including Sixth Street. Under the terms of the agreement, Datto stockholders will receive $35.50 per share in a transaction that values Datto at approximately $6.2 billion. The offer represents a 52 per cent premium to Datto’s stock price of $23.37 on March 16, 2022.

Datto’s current stock price is $34.70, reflecting the Kaseya bid, and its market capitalisation is $5.7 billion.

Kaseya, which was hit by a ransomware attack last year is a portfolio business selling infrastructure management, data protection and security products to SMBs through MSPs. It includes the Unitrends backup business and the Spanning in-cloud backup offering, both of which promise integration opportunities with Datto.

Datto CEO Tim Weller said: ”I’m encouraged by the continued investment in the rapidly-expanding global MSP community, and this transaction is another important validation of the channel.”

The purchase is currently expected to conclude in the second half of 2022, subject to customary closing conditions.

Nearline disk shipments flatten as COVID sales boom cools down

By

Chris Mellor

-

April 11, 2022

Researchers at TrendFocus reckon that all disk drive shipments declined sequentially in Q1, with nearline-only shipments growing year-on-year but declining slightly quarter-on-quarter and all other categories seeing double-digit percentage quarter-on-quarter declines.

Total disk ships in the quarter ranged from 53.0 million to 55.2 million – 54.1 million at the mid-point – which would be down 16 per cent year-on-year

There were an estimated 17.5 million to 18 million nearline HDDs shipped in the quarter – 17.75 million at the mid-point – which compares to 16.5 million a year ago and 18 million last quarter (Q4 calendar 2021) and 19.75 in the quarter before that (Q3 2021). Nearline unit ships have declined over the past two quarters, although the latest quarter is still higher than the year-ago 16.5 million.

Wells Fargo analyst Aaron Rakers told subscribers that “Assuming an ongoing increased average TB/drive (estimated 13.5TB/drive vs 13.1TB/drive in Q421), we would estimate that nearline capacity ship growth slowed to the mid/high-20 per cent year over range vs +60 per cent year over year in Q421.” He estimates that nearline HDD revenue in the quarter was >$3.5 billion, up over 12 per cent year-on-year.

The other HDD categories:

Mission-critical HDD shipments were less than 2.9 million, falling nearly 20 per cent quarter-on-quarter;
2.5-inch mobile and consumer electronics (CE) HDD ships were under 16 million, declining more than 15 per cent quarter-on-quarter;
3.5-inch mobile and CE HDD ships were about 18 million, down over 10 per cent quarter-on-quarter.

Mission-critical disk drives are being replaced by faster SSDs. Rakers points out there was a measurable decline in notebook demand during the quarter. People bought fewer notebooks, rather than buying the same number of notebooks as before but with SSDs fitted instead of disk drives. That sent the notebook HDD segment downwards. The desktop drive decline was attributed to seasonal weakness and worries about the economy.

TrendFocus also reckons Western Digital grew its HDD unit ship share by 200 basis points quarter-on-quarter – that’s two per cent, to an about 37 per cent share. That compares with Seagate staying flat at around 44 per cent and Toshiba declining two per cent or so to about 19 per cent. Rakers thinks that WD’s share gain comes from increased nearline disk shipments.

We have charted TrendForce disk shipments over the past few quarters:

*Blocks & Files chart using TrendFocus numbers.*

The TrendFocus numbers for a quarter are estimated and often get revised in subsequent quarters, so this chart is only a rough indication of what’s going on. Most nearline drives are bought by hyperscalers these days and their buying patterns can be lumpy. Having said that, it will be interesting to see if growth resumes over the next two or three quarters, or stays flat.

Intel’s not updating XPoint media for gen 3 Optane products

By

Chris Mellor

-

April 11, 2022

Intel will use existing gen 2 3D XPoint media in its third generation Optane SSDs and Persistent Memory products, and gen 4 Optane devices will likely be freed from Xeon dependency.

These points were revealed by Kristie Mann, VP and GM at Intel’s Datacenter and AI (DCAI) group, and other presenters at a Tech Field Day Session on April 8. Optane is Intel’s storage-class memory using 3D XPoint media – it is faster than NAND but slower than DRAM.

Gen 1 Optane SSDs and Persistent Memory DIMMs were introduced in 2018 with gen 2 devices (P5800X SSD and PMem 200 Series DIMMs) coming in 2020. These used gen 2 3D XPoint media with four layers of cells rather than gen 1’s two.

Mann showed an Optane roadmap slide and said: “Coming near the end of this year, we’re going to have our 300 series.”

The roadmap links Optane Persistent Memory (PMem) and Xeon CPU generations, with the current PMem 200 series synced with Cooper Lake and Ice Lake gen 3 Xeon processors.

The future Optane PMem 300 series will be twinned with Sapphire Rapids Xeons, Intel’s fourth generation of its scalable processor design. Sapphire Rapids will support CXL v1.1, an interim CXL specification before CXL 2.0, which will introduce memory pooling, which is the ability to have pools of memory remotely accessed over the CXL bus.

Mann confirmed the PMem 300 series will use PMem 200 Series Optane media: “It’s using second generation Optane media.”

That means Intel, as yet, has no public plans to develop 3rd-generation 3D XPoint media with, for example, 8 decks or layers compared to the second XPoint media generation and its 4 decks.

CXL 2.0

Some of an Optane device’s controller functions are carried out by a Xeon host CPU.

Mann said: “With Optane, we actually put part of the memory controller in the Xeon processor. So we we’ve come out with all sorts of memory control optimizations right in the Xeon processor itself.”

That means Optane can only be used in servers with Xeon CPUs.

The coming third generation Optane SSD, Lake Stream in Intel’s code-name, will be developed and arrive in the Sapphire Rapids era and overlap with Grand Rapids. The Optane SSDs are not closely tied to Xeon generations whereas the Optane PMem have been synchronised with Xeon CPU families. Mann said this could end: “Once we’re a fully compliant CXL persistent memory type of device, we may also have the flexibility to decouple from Xeon.”

The Granite Rapids iteration of the Xeon processor will support CXL 2.0, at least that’s Intel’s intention, and a future Optane PMem product – logically the PMem 400 Series – will support CXL 2.0 as well. Mann said: “And at that point, that’s when we want to have a CXL type of device for persistent memory. In that timeframe.”

In more detail she said: “So let’s put it this way: this CXL standard is still in development. So I would say we have very limited visibility into what types of devices and functionality are being built into all of the various … processors and controllers out there.

“Our approach right now, just like it has been with starting with a minimum viable product in the first generation and moving forward, is that we’re we’re doing all of the design, the validation, the optimization around Xeon. And we’re not actively going out and trying to connect to other types of processors, but it is going to be a CXL-compliant device. And so depending on how everybody implements these things, there could be compatibility. Just we’re not designing for, or guaranteeing it at this point in time,” Mann said.

Three other points. First, Intel is going to make it possible to address smaller portions of Optane memory. Intel Fellow and Senior Software Engineer Andy Rudoff said: “The things you see listed here, like optimising the power utilisation generation over generation, or making finer granularity access to our media, that’s the cache line access size.”

Secondly, CXL 2.0 should see performance increases: “Optane and CXL, I gotta say, are like made for each other … So the RAM has the DDR bus all to itself. Octane has CXL and now we get concurrency, those two things operating at the same time that we didn’t have before. So we should see some great improved performance by splitting that apart.”

Lastly, when Optane is available on CXL 2.0, the existing Optane applications should still work, said Intel.

KVCache

By

Chris Mellor

-

April 10, 2022

KVCache – Key-Value Cache is a mechanism used to store past Gen AI large language model (LLM) layers’ activations (keys and values) during the inference phase. It allows LLMs to bypass recomputation of these activations, improving performance. The cache serves as a repository to “remember” previous information, the pre-computed key and value pairs, reducing the need to reprocess entire sequences repeatedly. The memory-based KVCache applies to the attention mechanism of a transformer model (LLM). This attention layer computes relationships between input tokens (words or subwords) using three items: queries (Q), keys (K), and values (V). During text generation, the model processes one token at a time, predicting the next token based on all previous ones. Without caching, it would need to recompute the keys and values for all prior tokens at every step, which is an inefficient process.

Keys (K) represent the “context” or features of each token in the sequence. Values (V) hold the actual content or information tied to those tokens.

By maintaining a KVCache, the model only computes K and V for the new token at each step, appending them to the cache, and then uses the full set of cached K-V pairs to compute attention scores with the current query (Q). This drastically reduces computational overhead.

OEM

By

Chris Mellor

-

April 9, 2022

OEM – Opto-Electronic Memory. These are devices that use light – visible, ultra-violet or infra-red – signals and properties to store and retrieve data. They feature optical-to-electrical and electrical-to-optical transduction.

OEM more generally stands for Original Equipment Manufacturer and OEMs manufacture and supply IT systems and components for use by upstream suppliers. The Opto-Electronic Memory use of OEM is specific to the memory area.

CTERA: Not storage, but a cloud data services biz

By

Chris Mellor

-

April 8, 2022

CTERA is leaving its storage roots behind and becoming a cloud-based data services business selling to large enterprises.

Update. Nasuni comment added 12 April 2022.

That was the message presented to IT Press Tour attendees in Tel-Aviv by co-founder and CEO Liran Eshel and his team.

CTERA is a 14-year old company, VC-funded with its last round worth $30 million in 2018. It has raised $100 million in total and is getting close to being profitable and thus self-funding. CTERA was founded by Eshel and VP for R&D Zohar Kaufman after running an appliance company in the SMB network security area. They built a cloud file storage gateway appliance and software providing file sync and share, and security.

CTERA now presents itself as a Cloud Data Services supplier with a layer of services above its edge filers and edge-to-core-to-cloud global file system:

Programmable deployment is based on a DevOps SDK, which enables developers to automate system rollout. CTERA says an edge filer can be deployed in 90 seconds using the SDK. Discovery and migration supports customer file and object discovery across their IT estate. It can then migrate files to the cloud, using network transmission for SMB customers and Amazon Snowball for larger projects.

Eshel says CTERA thought about working with Datadobi on migration but realized it could do things better by taking advantage of its edge filers. It deduplicates and compresses every block sent up to the cloud using them.

CTERA Insight provides information about files in CTERA’s filesystem and their content. No-code data pipelines can be set up to filter and extract data sets then feed them to analytics applications in the cloud or Lamda functions. Data is extracted from compressed and deduped and written to a new dataset for this purpose.

Cloud Analytics can extract data from CTERA’s immutable S3 buckets using filters, rehydrate it, and send it to analytics applications using an S3 data export microservice. CTERA also provides multi-layer virus scanning and malware protection combining on-access edge scanning with a cloud-based ICAP service for later detected threats. The ransomware protection involves continuous replication into immutable S3 buckets, zero-day detection, and instant recovery using rollback.

CTERA competition

CTERA sees itself mostly competing with NetApp, aiming to replace its filers. But it also competes with Nasuni and Panzura, fellow companies in the cloud file services and collaboration space.

Eshel thinks CTERA can scale much farther than Nasuni, with support for thousands of endpoints. He suggested Nasuni installations typically amounted to 30 or so endpoints and thinks CTERA has superior deployment and automation capabilities to Nasuni.

In general, we can assume it is easier for CTERA, or Nasuni and Panzura, to sell to customers who have on-premises filers and no connected public cloud file system or collaboration resources than to try and replace each other. Greenfield should be an easier sell than brownfield.

NetApp is the giant to beat for CTERA, as for Nasuni and Panzura. Nasuni has raised around $229 million, with a $60 million round earlier this year. Our thinking is that it is second in revenue behind CTERA in this distributed cloud file system services market, with Panzura in third place. (Update; see the Nasuni comment below.)

Veteran Affairs

In December 2021 CTERA won a place in Peraton’s Department of Veteran Affairs bid for a $497 million contract to provide infrastructure-as-a-managed-service (laaMS) for storage and computing facilities across the US and globally. CTERA, which provides the only global filesystem included on the Department of Defense Information Network (DoDIN) Approved Products List (APL), will deliver file services for mission-critical workloads, connecting up to 300 distributed sites (hospitals) to the VA Enterprise Cloud powered by AWS GovCloud (US).

This involved 220PB of data from business operations to medical imaging. It uses edge filers, based on Cisco HyperFlex hyperconverged systems, and CTERA-supported 80TB Amazon Snowball boxes being used to ship data up to the AWS GovCloud for S3 storage. This cloud-based storage replaced NetApp storage systems and HPE servers.

HPE and NetApp bid a system using 3PAR tier 1 and NetApp StorageGRID tier 2 storage through system integrator Thundercat. It failed to win, even after Thundercat protested the contract being awarded to Peraton.

Eshel thinks CTERA is taking share from NetApp because it is growing faster. As a supplier of caching, cloud-connected edge filers, CTERA says it is about much more than storage, with cloud-based data services being an essential part of its deployments. CTERA is a cloud-centric data services business whereas, in our view, NetApp is an on-premises storage supplier with a very good cloud connectivity story.

Eshel said he has no plans to raise more VC money. He wouldn’t reveal his run rate. The Veteran Affairs was a huge win for CTERA and other wins – such as Iron Mountain (hundred of endpoints worldwide), and the Thales-led London Underground digital modernization – suggest to us it has attained a run rate amounting to multiple tens of million dollars a year. Data services is bringing in the dollars.

Nasuni

On seeing this article Nasuni told me;

On market and growth aspects, Nasuni can point to evidence in the public domain (such as number of LinkedIn employees) which would suggest that it is substantially larger in terms of revenue, employees and customers than CTERA
Nasuni is nearing a revenue milestone that will see it surpass CTERA and Panzura substantially
Nasuni’s last funding round added to the company’s cash balance of $100 million which suggests there has been little cash burn from operations over the last 2-3 years
On Nasuni scalability, Nasuni customers AECOM (the world’s biggest engineering and design firm) and Omnicom (the world’s largest advertising company) can confirm that Nasuni Edge instances are running in hundreds of sites that are all managed through Nasuni’s central management console
Regarding endpoints, CTERA appear to mean their desktop sync client running on home user desktops, rather than caching instances serving up files in branch or remote office locations – so they seem to be comparing apples with oranges somewhat there
Nasuni competes with NetApp in the enterprise 95 percent of the time and Panzura and CTERA only in SMB.

NetApp on the CloudOps acquisition trail

By

Chris Mellor

-

April 8, 2022

NetApp has bought eight companies in two years as Anthony Lye, EVP and GM of the Cloud DataServices Business Unit, builds up a cloud data operations facility.

The aim is to provide customers with the means of using various cloud resources optimally and cost-efficiently. They don’t have to get down and dirty in the weeds of evaluating, for example, which of 475 AWS compute instances to use, how to reserve instances, or optimize Spark.

None of this has much to do with NetApp’s traditional core focus on on-premises storage, and adjacent focus on providing its storage facilities in the AWS, Azure, and Google clouds.

The Cloud Business Unit was initially incubated and run by Jonathan Kissane, SVP and general manager. Lye came on board in March 2017 to make NetApp’s Cloud Business Unit profitable. When Lye was hired, Kissane became NetApp’s Chief Strategy Officer but left eight months later. Who was Lye and where did he come from?

It’s no Lye

Lye was a product marketing manager at Tivoli in the early ’90s and then became a senior director/ major accounts sales rep at Remedy Corp for four years from 1994. Then came a big move to president and CEO of SaaS company ePeople from March 1999.

In late 2005 he became Group VP and GM at Siebel Systems, which was acquired by Oracle, with Lye moving onto SVP and GM of Customer Relationship Management. He managed a unit of 3,000 people and acquired 10 companies for more than $2.5 billion.

In 2012 he joined Publicis as Global President for Digital Platforms and Product, but spent just nine months at the French multinational advertising and public relations company. He moved to become Chief Product Officer at HotSchedules, a provider of workforce and inventory management services for the restaurant industry, and became its President and CEO, leaving in late 2016 to be EVP and Chief Cloud Officer at Guidewire Software.

Guidewire supplies an industry platform-as-a-service for property and casualty insurance carriers. Lye quit after six months, jumping ship to NetApp. With hindsight, it looks like he was asked to replicate the things he did for Siebel/Oracle at NetApp in the cloud area.

This is a seasoned SaaS business executive, with experience building SaaS business units and acquiring companies to help with that.

Buying time

In January 2018 NetApp’s Cloud BU became the Cloud Data Services BU and Lye set out on the acquisition trail.

The first was Talon Software, which provided its FAST software-defined storage enabling global enterprises to centralize and consolidate IT storage infrastructure to the public clouds. NetApp said at the time that the combination of its Cloud Volumes technology and Talon FAST software meant enterprises could centralize data in the cloud while still maintaining a consistent branch office experience.

Lye said at the time: “As we grow our cloud data services offerings with solutions like Cloud Volumes ONTAP, Cloud Volumes Service, Azure NetApp Files and Cloud Insights, we are excited about the potential that lies in front of this new combined team to deliver complete solutions for primary workloads. We share the same vision as the team did at Talon – a unified footprint of unstructured data that all users access seamlessly, regardless of where in the world they are, as if all users and data were in the same physical location.”

This acquisition was a cloud product-as-a-service deal, not a cloud service operations deal. That came the next month with CloudJumper, a cloud VDI player and the first of eight such acquisitions.

NetApp has bought rapidly growing companies in what is still a fast-expanding market. These companies had great prospects and NetApp will have paid a good price for them, as the $450 million reported for Spot indicates, a 6.6x increase on Spot’s $52.6 million funding total.

We have collated, in the table below, information on the nine acquired companies, the total funding (where it’s known), as well as the reported acquisition cost.

We must emphasize that this is highly speculative, but we also applied what we felt was a more conservative 5X increase to the total funding of these eight companies, $148.65 million, to arrive at a guess of $743.25 million for the vendor’s spending. There was no funding amount for four of the acquisitions, so this table assumes NetApp spent $10 million to buy each of them, giving a total potential cost of around $783 million.

Even if the number turns out to be smaller, a nine-firm buy is a huge bet by NetApp, and is notable because the acquisitions are in a completely new enterprise market that is quite a long way from its traditional and core storage on-premises market and growing adjacent public cloud storage market. Signalling this separation, Lye’s CloudOps software products were kept out of Chief Product Officer Harvinder Bhela’s control when he was hired by NetApp in December 2021.

Lye’s BU is doing well. In March we wrote NetApp’s “Cloud Operations (CloudOps) business looks so good for NetApp that the company has increased its overall revenue targets.”

These were the numbers behind that: “NetApp thinks its public cloud annual recurring revenue (ARR) will be $2 billion by its fiscal year 2026, up from $1 billion in 2025. NetApp should now reach its $1 billion public cloud target by 2024 – a year sooner than predicted.”

These numbers show why NetApp thinks it’s worthwhile to spend hundreds of millions in its rush to build a cloud data services business before anybody else. Public CloudOps is going to become, NetApp hopes, its second goldmine after storage.

NewsPaperStorages and File System News

NewsPaperStorages and File System News

Bechtle pulls HPE Primera array from datacenter over performance issues

MAID

StorageMAP charts uncharted unstructured data oceans, estimates carbon footprints

Tiered backup storage player ExaGrid turns in Q1 numbers

Storage news ticker – 12 April

Kaseya buying Datto for $6.2 billion

Nearline disk shipments flatten as COVID sales boom cools down

Intel’s not updating XPoint media for gen 3 Optane products

CXL 2.0

KVCache

OEM

CTERA: Not storage, but a cloud data services biz

CTERA competition

Veteran Affairs

Nasuni

NetApp on the CloudOps acquisition trail

It’s no Lye

Buying time

ABOUT US

FOLLOW US