April 12, 2018

iXsystems launches TrueNAS M-Series

iXsystems, well known system builder and active open source player, just announced a new line of product named TrueNAS M Series. We met and visited the company last December during The #25 IT Press Tour and discover a strong player with deep expertise and proven track record chosen by hundreds of client companies.

iXsystems now unveiled TrueNAS M Series to address the capacity and performance need for new rapid growing applications environments.

Two 4U models M40 and M50 with NVMe and NVDIMM storage entities to boost performance in write and read mode coupling as well 14TB HDDs. The M40 uses SAS SSD and the M50 considers NVMe SSD for read cache, but these 2 models use NVDIMM for write cache. Systems arrive also with 100GbE and 32Gb Fibre Channel networking.

In term of interfaces, M Series expose NFS, SMB, AFP, iSCSI, FC and S3 protocols. It also supports remote copy to public clouds such Amazon S3, GCP, Azure and Backblaze B2 and potentially with any S3 compatible object storage solutions.

Advanced configuration is pretty impressive for the M50 with up to 3TB of RAM, 32GB of NVDIMM, 15TB of NVMe flash for reach cache and 4 x 100GbE ports.

iXsystems clearly redefines storage appliances and servers market segment with these 2 models coupling advanced technologies. The performance and capacity gains are significant if we compare with the Z Series. Pricing starts at $20k for an entry model and 1PB can be stored for less than $100k giving the GB for less than 10 cents.

The team has also introduced an object storage solution based on Minio, the fast growing open source object storage.

This product introduction represents a strategic step for the company and we expect that this M Series line will impact positively the 2018 revenue.

April 9, 2018

Full Flash NAS for Qumulo

Qumulo, leader in scale-out NAS, met last december with the 25th IT Press Tour, just unveiled the P-Series, its NVMe all-Flash file storage product line aligned with the NAB Show.

This new product iteration consists of NVMe SSDs, Intel Xeon 6126 processors and of course 100Gb networking still with the QF2 software layer.

A P-Series cluster has to be deployed with a minimum of 4 nodes (max. is 1000 nodes) able to deliver 16GB/s per 4 nodes cluster i.e 4GB/s per node. We think and hope this throughput is linear but we're surprised by the limited capacity per node: model P-23T for 23TB (12 x 1.92TB NVMe SSD) or model P-92T for 92TB (24 x 3.84TB NVMe SSD) per node so less than 400TB raw and 266TB usable (2/3 EC efficiency) for the entry level cluster in the larger size. 1PB usable requires 1.5PB raw and 17 nodes P-92T.

Qumulo P-23T or 92T respectively with 23 or 92TB
Interesting approach for an established and recognized leader, Qumulo is clearly #1 in M&E with many, many references and the 10 largest Hollywood Studios as clients.

This product competes with Avere Systems, now Microsoft, Dell EMC Isilon F800, Pure Storage FlashBlade, Quantum Excellis Scale-out NAS, Rozo Systems RozoFS or WekaIO Matrix. We also remember Arkologic (see my old posts in Oct. 2012 & Jan. 2014) who tried to build and sell a full flash NAS 5 or 6 years ago and finally disappeared.

But times have changed and the market is now ready, users really need such solution now.

April 4, 2018

New era at DataCore

DataCore, leader in storage virtualization with SANsymphony for a few decades and more recently active player in the Software-Defined Storage wave has shaken its management layer recently.

I just discovered that George Teixeira, CEO and co-founder of the company, has updated his LinkedIn profile and is now executive chairman. This event invited me to check the rest of the team and I realized that DataCore is entering a new era with several new team members.

So George moves from CEO to Executive Chairman, the new CEO is Dave Zabrowski who was CEO of Cloud Cruiser before the acquisition by HPE last year and before he was CEO of Neterion. At the time I wrote that post his LinkedIn profile was not updated.

I also discovered a new CMO on the web site, Gerardo Dada, who came from SolarWinds and before Rackspace.

This is a new era for DataCore perfectly illustrated by a new logo with new color, no more red but a kind of green.

We expect to see now more changes both in local teams, in product and market strategy. We'll see, probably now a realistic exit could be considered...

March 20, 2018

Penguin Computing develops strong storage partnerships

Penguin Computing, alternative player in HPC based on open source solutions, continues to iterate its storage solutions. The company has introduced it SDS product line named FrostByte storage solutions based on partners' products.

Aligned with different needs and uses cases, FrostByte uses:
  • ThinkParQ BeeGFS for HPC,
  • Red Hat Ceph Storage for object and block in cloud infrastructure, M&E and data lakes,
  • Red Hat Gluster Storage for VM and containers, big data and analytics and
  • Lustre for HPC and big data analytics as well.

March 15, 2018

New Era for Storj with now a seasoned CEO

Storj Labs, one of the few P2P storage player now promoting Etherum blockchain based storage, expands its team with the recruitment of its new CEO. Ben Golub, former CEO of Docker and in the past CEO of Gluster, just arrived to launch a new era for the company. He is often associated with exit as he did one for Gluster selling the open source storage company to Red Hat for $136M in 2011. We expected the same at Docker but nothing happened, is it because founders are too greedy or other reasons? We saw similar behaviors from other companies led buy french people who finally didn't accept acquisition proposal, missed the window and entered in trouble zone. He also board member of Minio, te leading open source object storage player, revolutionizing that category. So far the company has raised a total of $33M of funding.

Users share their local free storage space with other users in a secure way. Data are sharded or chunked, encrypted and distributed between farmers (people who contribute and share their free space) i.e among storage nodes. Storj now enables 60PB of data across 150,000 storage nodes present in more than 200 countries. The company charges $0.015/GB/month, which 3 times more expensive than Wasabi and Backblaze B2 even if these offering are different and we don't include traffic.

It reminds me several projects: Adanam, KerStor, Wuala, Symform, Blockade, SpaceMonkey, AeroFS, Tudzu, Connected Data, Ugloo, UbiStorage, CloudPlan, AetherSore and SiaTech but only a few of them are still alive. Storj is one of the best solution for sure available probably at the right time.

March 8, 2018

Robin Purohit now CEO of Griddable.io

Griddable.io, a SaaS company building a new approach for synchronous data integration, was founded in 2016 by Rajeev Bharadhwaj.

The company has raised a Series A and decided to recruit a full time CEO with Robin Purohit. Rajev and Robin has a story together as Veritas acquired Rajeev's company Ejasent early 2004. This acquisition was one of the key ones for Veritas to sustain the Utility Computing strategy before everyone speak about Cloud Computing. Rajeev was the founder of Aryaka Networks and KiteDance.

The genesis of the solution came from Databus, a LinkedIn originated Apache project. Griddable goal is to offer a global synchronisation service across the enterprise for on-premises and cloud entities.

February 13, 2018

Fantastic program for the IT Press Tour #26

The IT Press Tour, the leading press event for IT press, just unveiled the list of participating companies for the 26th tour scheduled the week of 26th of February.

We recognize a few companies we already met in the past but also several new ones who drive some new IT segments.

Topics will be around big data, analytics, databases, data preparation, cataloging and insights, applications monitoring and performance management, in-memory computing, continuous development and GDPR soon in place in Europe.

Here is the list:
  1. Aerospike, leader in high performance NoSQL database,
  2. Alation, reference in data driven analytics for enterprises,
  3. Anaplan, pioneer in corporate planning management,
  4. Datadog, key player in modern applications monitoring and performance management,
  5. Frame, a leading secure cloud workspace platform,
  6. GridGain Systems, In-memory computing platform leader built on Apache® Ignite™,
  7. Harness.io, new player in Continuous Delivery-as-a-Service,
  8. MapD Technologies, top vendor in GPU-based database,
  9. and Waterline Data, fast growing actor in data discovering and cataloging.
This edition will be again huge with a dense program and top innovators. I invite you to follow us with @ITPressTour, #ITPT, various publications and reporters Twitter handles.


February 1, 2018

Nyriad for a new data protection

Nyriad plays with its name, coming from Myriad and Raid, this is what we suppose.

GPU-based products are hot, we have met several databases vendors leveraging such processors like BlazingDB, BlazeGraph, SQream Technologies, Kinetica or MapD Technologies and now it's about storage product.

The company targets big data and high performance computing. It has its roots with the Square Kilometre Array project in New Zealand having delivered 160TB/s processing 50PB/day. It works well with parallel file storage approach such Lustre.

Partners are Nvidia, SuperMicro, Microsemi, Tyan, HPC Systems, KingSpec, Revera, Datacom, WDC and Netlist and the company is actively looking for oems.

The company develops NSULATE, a GPU-accelerated block device. This storage product is a DAS with hundreds of devices. It's compatible with Linux file systems and with ZFS + NSULATE with SHA-256 delivers 2-4x faster performance than ZFS for Lustre. We don't have a lot of details on the solution and we expect to get more info soon.

The philosophy resides in the replacement of the RAID controller by a general purpose GPU board. We notice a divergence of information between the site and the paper doc as the site mentions up to 128 parities and the paper doc up to 255, meaning that potentially volume and disk could be big and obtain lots of flexibility. Hum... In fact this is far more than we need if we compare with traditional data protection mechanisms.

The GPU on the node could be used also for compute in addition to storage functions delivering a new compute + storage convergence example.

NCRYPT is the companion of NSULATE and offers GPU-accelerated cryptographic algorithms. It includes a real-time blockchain APIs that provides encryption, hashing and cryptographic signature generation.

It deserves a deep look for sure as it will bring something new to the table against the Mojette Transform from Rozo Systems, MemoScale or the Intel ISA-L functions.


January 29, 2018

DriveScale to democratize composable infrastructure

A hot topic recently emerged on the market, the Software Composable Infrastructure aka SCI and the IT Press Tour had the privilege to meet and visit DriveScale, one of the few drivers of this industry trend.

The idea if this approach is justified by the business that needs agility and scalability inspired by public clouds. The main characteristic is well summarized by the term disaggregation with better compute and storage resources segregation and granularity, controlled by an intelligent software layer.

The goal of this data center approach is to increase resource utilization thus reducing cost, improving ROI, optimizing TCO with realtime dynamic provisioning to react to new, changing and modern workloads. In a nutshell, adopting a SCI philosophy offers a cloud-like IT operations and represents a new iteration towards fully automated data center.

You get the idea, SCI is the right term to describe an IT infrastructure controlled, designed and built dynamically by software.

SCI provides a transparent service for upper applications layers and no applications changes are required. From single, simple to complex horizontal scalability, SCI supports recent applications operating on Hadoop, Spark, NoSQL databases such Cassandra and containers as well. This is also interesting to see that a commercial approach is justified to control open source applications.

This new architecture is a new iteration towards disaggregation with better granularity tough to deliver with classic or recent tools.

Founded in 2013 by 2 SUN ex-employees Tom Lyon and Satya Nishtala, DriveScale illustrates the long time tag line of the workstation company i.e "The network is the computer". Finally SUN has in a sense anticipated it but never delivered at least at this level as what we live today was an utopia 10 or 15 years ago. They don't forget their past colleagues so it's not a surprise to see Scott McNealy, James Gosling and Whitfield Diffie as advisors. With 25 employees today, DriveScale is frugal with VC money as they raised "only" $18 million so far but clearly we can expect a new round in 2018.

The solution is organized around 4 components:
  • Adapter aka DSA, currently a SAS to Ethernet 1U appliance that bridge compute and storage resources. We expect a NVMe flavor in 2018.
  • Management Server aka DMS controls the rack SCI service and is available as bare-metal or VM Linux instances.
  • Agents are lightweight Linux processes running in user mode on servers controlled by DMS.
  • and Central is a cloud-based administration to control and manage deployments globally.
DMS receives infos from agents for servers and DSA for storage.

A RESTful API is provided to integrate the DMS service with deployed and famous data center tools.

An other way to see this IT Infrastructure approach is "Disag. (hardware resources) and Compose" a new logical computer different from classic virtualization or container mode. In fact these 2 services can reside above SCI to offer application independence and mobility.

A DriveScale cluster groups diskless stateless servers and JBOD thanks to the DSA. This key elements has 2 12Gb four-lane SAS interfaces and 2 10Gb Ethernet interfaces. Power supplies are redundant as well. 80 disk drives are supported and could deliver a throughput of 80GB per second. For servers, DriveScale supports 2U, 1U, 1/2U and 1/4U models hosting an Agent.

In term of licensing, one adapter per JBOD costs $10,000, node licenses is at $2,000 per logical node per year, drive $25 per year to reach approximately $106,000 the first year for 30 nodes and 300 disks and thereafter $76,000.

Key partners for DriveScale are Hortonworks, Cloudera, MapR, DellEMC, HPE, Arista, Cisco, SuperMicro, Quanta, Promise, Sanmina and Foxconn with several resellers such WWY, DellEMC or Promark.

This effort is a new example of vendors trying to adopt a cloud model and limit on-premise erosion. The industry has already addressed this movement with hybrid cloud as they realized that dollars are going elsewhere especially to the big 3 cloud service providers pockets. Again this is the direction of the industry. DriveScale will benefit from infrastructure vendors who try to delay this certain move. The company is leading this industry SCI wave with a pure independent solution. Clearly 2018 will be an interesting year to watch...

January 23, 2018

Hedvig leads the SDS pack

Always a pleasure to visit Hedvig and meet Avinash Lakshman and his team, we had a new opportunity during the last IT Press Tour in December. Obvious leader i the SDSD space, the team has presented a new iteration of the product fully hardware and cloud agnostic.

The product, Hedvig Distributed Storage Platform aka DSP today in 3.0, was illustrated with a multi-cloud and cloud orchestration approach, validating again that software makes the difference. Thanks to the HPE investment, support and active participation of Milan Shetti, CTO datacenter infrastructure at HPE, as technical advisor, the company has made significant progress and market penetration. HPE has finally selected the right player after several tentatives that never took off.

Hedvig means independence and intelligence as finally the solution is not linked or limited to any interface - file, block and object - supporting any workload, any application, open to virtualization and containers, any hardware and working today the top 3 cloud service providers AWS, GCP and Azure. Clearly the message is around universal usage, providing a converging platform between primary and secondary storage across multiple applications.

Again like others players, the company provides a multi-cloud model surfing on an market hype. Two effects, "me two" so the company can be selected for that support and at the same time no differentiator against direct competition. But again you have to be multi-cloud, hybrid and cloud agnostic nowadays.

Hedvig defines itself as a HyperScale SDS, I should say as well, multi-protocol SDS.

In France BNP Paribas has chosen the product and designed a geo distributed data center across Paris and London.

To refresh readers on the product, we need to say the the philosophy has scalability, agility and resiliency in mind. Three components define the platform:

  • Storage Service as a software installed in Linux x86 systems, glued together. They formed the scale-out back-end.
  • Storage Proxy is the front-end exposing virtual disks to the applications tier with block or file protocol.
  • And APIs to integrate service close to the application, this is the object interface place.
Virtual disk is a core concept here delivering what many other SDS solutions don't and can't provide especially encryption, deduplication and compression.

DSP 3.0 introduced last summer provides three new features:
  1. FlashFabric with the ability to scale flash across private and public clouds and deliver optimized caching supporting PCIe, NVMe and 3D XPoint into s single platform,
  2. Encrypt360 for end-to-end encryption, deduplication done first, based on 256 bits AES per disk and accelerated via Intel processors,
  3. and new and improved CloudScale plugins for partners software like Veritas OST, VMware or Red Hat for container imaging.
We anticipate some great news from Hedvig in 2018 being present on the hot list of many players and of course customers.