Druva bets on the demise of the (big) datacentre

Antony Adshead | No Comments
| More

Druva has added disaster recovery capability to its existing Phoenix product. But its what they don't plan to do with Phoenix that's perhaps more interesting.


Phoenix allows globally deduplicated backup with the cloud as a target, from physical and virtual machine environments in ROBO-size deployments as well as from VMware environments up to about 200TB, which it classes as "small datacentre".


Phoenix already has cloud-based archiving, with Amazon Glacier as its repository.


It recently added cloud-based disaster recovery in which VMs and data are converted to Amazon Machine Images (AMIs) and kept in a virtual private cloud that allows failover and failback while maintaining essential configurations from the likes of Active Directory, name servers, DSCP etc.


This is, like the rest of Phoenix, a ROBO and small datacentre-class product. So, the question is begged, why not build this out to large datacentre capability with physical backup and other targets than the cloud?


For Druva CEO Jaspreet Singh the approach taken is informed by predictions that the future of the large datacentre is limited by the rise of the cloud.


He said: "We don't plan to offer backup to local targets for big datacentres because that market is already well-covered. And we don't see a future with large datacentres in it."


He added: "We see there being a majority of data kept in the cloud with smaller datacentres as local storage to provide high availability."


And for Singh that's the prediction that informs the company' strategy.


"Veeam rode a wave," said Singh. "And that was virtualisation. With the on-premises datacentre wave there's not much differentiation we can offer, so the wave we're riding is the cloud. We're placing bets that we will have enough operational experience in the cloud to be the number one in cloud backup in the future."


Object storage vs NAS . . . in the same product

Antony Adshead | No Comments
| More

It has been an interesting month in object storage, with developments that push the technology along but also provided an interesting footnote regarding file access that illustrate an interesting stage of the evolution of object storage.


On the one hand HPE and Scality announced a partnership that will see the object storage provider becoming the technology of choice for HPE in that space. The two companies have had a global distribution agreement since October 2014 but that was with HPE's server division. Now HPE's server and storage divisions have embraced Scality as their "standard offering" in object storage, according to Scality CEO Jerome Lecat.


Lecat couldn't comment on the fate of HPE's existing StoreAll object storage products but said: "What I know for sure is that for large capacity file and object storage deployments will now be fulfilled by HPE with Scality software on HPE servers."


Lecat was also keen to stress that there is a strategic imperative to offer object and file access with its products and said more than 50% of Scality's deployments are scale-out file access, and that "most of our competition is with [EMC's scale-out NAS product] Isilon."


Scality does this by providing NFS, Fuse (Linux) and SMB/CIFS file access to its object stores.


It has offered this for some time, but also this week EMC announced an update to its Elastic Cloud Storage (ECS) software-defined object storage platform in which a new feature is the inclusion of NFS access.


What's interesting is that a bleeding edge set of products (software-defined, aimed at cloud use cases and very large volumes of files) should have as a major addition of functionality a protocol that dates back to the mid-80s.


It demonstrates if not an obstacle then a speed bump in the adoption cycle of object storage and that many customers want the great scalability it offers but do not have apps built with object connectivity.


Lecat suggested this is the case.


"The choice [between object and file access] is down to the application. If you have an application that is a completely object-based application then you don't use file access. But if the application expects Windows, for example, then you need file access," said Lecat.


Is there a performance penalty when using file access instead of object?


It depends, said Lecat.


"In file access there are a number of metadata operations that are not there in object storage," he said. "There can be up to a 20% performance penalty, for example if there are a lot of metadata operations and directory movement. But if it's for streaming files, for example, performance will be almost identical."

NetApp buys Solidfire but it has been a long and tortured road to all-flash

Antony Adshead | No Comments
| More

In case you missed it, the big storage vendor news during the holiday was NetApp's acquisition of all-flash array maker SolidFire for $870 million.


The move brings to a conclusion several years of vacillation and false starts in NetApp's stance towards flash storage and gives the company a solid all-flash presence in the market.


This is something it has arguably lacked. The history of NetApp's relationship to flash is one that could be described as cool or cautious. Or alternatively, it could be described as confused and reactive.


We'll look more at that below, but what the Solidfire deal gives NetApp is a well-regarded set of all-flash storage arrays.


Solidfire started out with all-flash targeted at cloud service providers and is Fibre Channel and iSCSI block storage. With cloud in mind, it provided automation and multi-tenancy with the ability for administrators to assign storage volumes with different characteristics to different customers.

More recently, Solidfire added advanced storage features such as replication and other data protection features to appeal more to the enterprise market.


Its four all-flash arrays are the SF2405, SF4805 and SF9605, that all offer 50,000 IOPS per 1U node with raw capacities of 2.4TB, 4.8TB and 9.6TB, respectively. The SF9010 also has 9.6TB raw capacity but with eight-core CPUs offers 75,000 IOPS per node.


With data reduction - data deduplication, compression and thin provisioning - effective capacity is upped by between 3x and 9x.



NetApp's tortured journey to all-flash


So, NetApp has gained a solid offering of all-flash arrays that are suited to cloud provider and enterprise all-flash use cases.


But, its journey to this point has been interesting to say the least. Was NetApp just cautious, and hence a bit late to the all-flash market? Or has there been a history of confusion of vision and reacting slowly to the market?


Go back to the turn of the decade and NetApp was in a similar position to the other major storage vendors with regard to flash. In 2009 it offered a flash read cache with its FAS arrays and in 2010 offered flash drives with them too. In 2011 it offered them with its E-Series arrays as well.


That was the early days, but soon things got hotter in the all-flash market. In April 2012 EMC announced it would buy Israeli all-flash startup Xtremio and in August that year IBM announced it would buy solid state pioneer Texas Memory Systems.


This was when it started to look like a big six storage player needed to have all-flash arrays in its portfolio and to get that they had to buy one or build one or retro-fit spinning disk storage controllers for solid state and be clear about the performance limitations.


By late 2012 the elephant in the room for NetApp was, what would be its next move with regard to all-flash?


In Dublin in November CTO Jay Kidd told me NetApp had no plans to get into all-flash. He said the market wasn't that big and that there were better places to put flash, such as at the server, with PCIe flash. All of which was in keeping with the NetApp attitude to flash hitherto.


Cue the deafening sound of back-pedalling, rowing back or whatever you want to call it. Within a couple of weeks the NetApp CTO's office scheduled an interview with my US counterpart in which he went a long way to reversing the previous stance, saying it would be a milestone year for all-flash and that NetApp could buy or build an all-flash product.


Then, in early 2013, NetApp brought out its first all-flash array, albeit an E-Series box - designed for spinning disk - with flash drives fitted.


But at the same time FlashRay was unveiled - a dedicated flash environment based on NetApp's Data Ontap operating system. Finally, it looked like NetApp would be building its flash future on in-house developed intellectual property.


By the end of 2014 the big six array makers had mostly decided on an approach to all-flash. The year before, EMC and IBM had got products they had acquired onto the market and Hitachi Data Systems (HDS) had built an all-flash module for its VSP arrays. Meanwhile, HP had brought all-flash to the market in its 3PAR arrays while Dell and Fujitsu were quite clear their approach would be to retrofit flash to spinning disk architectures, and that hybrid flash was their preferred approach.


By this time NetApp offered FAS filers as all-flash and had the E-Series too, but only a single-node selected customer beta array had emerged from the FlashRay venture with its Mar OS. Still, as we went through 2015 it looked like NetApp had settled on a three-way flash strategy - flash in FAS and E-Series plus FlashRay.


But things didn't look good for FlashRay. By early 2015 its founder Brian Pawlowski had left NetApp and still there was no product on general release. Bear in mind this was now two years after IBM and EMC had brought all-flash, ground-up designed products to market.


Fast forward to December 2015 and NetApp announces its purchase of Solidfire and the scrapping of FlashRay.


Now, NetApp has a good offering of performance all-flash arrays that it can offer alongside the less-costly all-flash FAS and E-Series as well as the hybrid options those bring.


Better late than never, I suppose. But what NetApp to falter over flash? What lay behind the tardy decision to invest in and ultimate axing of FlashRay?


I can only suppose there was a background set of attitudes - and potentially conflicting ones - towards flash storage that hamstrung NetApp.


Early on, NetApp made it clear that all-flash was not really on the agenda for the company, with cache and server flash the preferred approaches. As we saw, by late 2012 NetApp execs downplayed flash but were clearly under pressure not to write off their potential involvement with it.


Then as the market hotted up NetApp was pulled along and FlashRay was conceived. But why so late? Why such apparent lack of commitment? Was there a background resistance to flash internally? Even in late 2014 we can still see some NetApp execs saying flash just isn't that big a deal.


Perhaps it was a miscalculation over the likely timescale over which flash will be important to the world of storage. Perhaps there are or were different voices that competed over the nature and extent to which NetApp should be involved with flash storage.


Whatever the background reasons, at least now NetApp has a clear strategy with regard to flash, albeit two years later than its main competitors.


Spectra Logic Black Pearl object storage + SMR disk + tape = Arctic Blue

Antony Adshead | No Comments
| More

When Spectra Logic launched its DPE archive-class NAS in September it was a play at a space often targeted now by object storage products. So, I asked whether Spectra Logic had considered object storage. It's an obvious choice for large amounts of unstructured data.


The answer to the question came in October when Spectra Logic married its Black Pearl object storage head with a back end that can comprise a mixture of 8TB Shingled Magnetic Recording (SMR) drives (the ones that overlap data tracks to pack more in) that can be powered down with tape.


The idea is that Arctic Blue performs the role of an archive layer, but with tiers that allow for varying read and write times.


On the one hand the different classes of media - disk and tape - are better suited to reads and writes respectively, so Spectra Logic claims the same throughput (around 1GBps) for reads and writes. Normally, with just disk, reads would be much fast than writes but tape's fast write speeds evens the two out.


Also, policies can be set so that stored data is stored and migrated to different tiers: Spinning SMR disk, powered-down SMR disk, and tape. Claimed access times are less than 1 millisecond for powered-up disk drives and about 30 seconds for powered down SMR drives.


Arctic Blue puts up to 96 8TB SMR HDDs in a 4U box. Four of these can be clustered to provide a full rack of 6.4PB of disk. To this can be added tape capacity up to exabytes in scale.


Spectra Logic says the cost of storage on Arctic Blue is around 10c per GB, which it says compares to 7c for a tape library. Its ideal user, it says, is a customer that is probably at less than 3PB of capacity and is happy to pay the extra 3c per GB for the features - ie tiering - that a pure tape library can't offer.

UPDATE: HGST has deployed Active Archive units

Antony Adshead | No Comments
| More

Last week HGST business development manager Manfred Berger told me that the company had not closed a sale on its Active Archive petabyte-scale cold storage arrays since launch in April, but had 160 prospects. That blog post is here.


But, soon after publication HGST's press representatives got in touch to say it had made deployments. It said there had been Active Archive deployments globally, including in Europe. Here is what HGST said in a statement:


"As a public company, we cannot disclose the number of shipments. However, what I can tell you is that we have deployments globally, including Europe."


I asked what type of deployments these were. Were they proof-of-concept trials at customers or the result of actual sales? Unfortunately, HGST's press representatives were unable to elaborate, they said, due to HGST being a public company. 







HPE hopes to put a cuckoo in the nest of EMC VMAX customers

Antony Adshead | No Comments
| More

Hewlett Packard Enterprise (HPE) has made a bold play to convince EMC VMAX and Oracle users that they should migrate to HPE's 3PAR all-flash arrays.


It should be called "Project Cuckoo" as it's a clear plan to supplant another vendor's array from the datacentre, but HPE execs seem to have let that naming opportunity slip by.


HPE says customers that implement 3PAR all-flash arrays with their EMC VMAX still in place will gain total cost of ownership (TCO) savings of 50%.


The scheme hinges on the use of Oracle Automatic Storage Management (ASM), which allows customers to set more than one storage array to support the Oracle installation. The two arrays can co-exist alongside each other and the software automatically utilises the best-performing storage.


HPE is clearly convinced ASM will show 3PAR all-flash to be superior to their old VMAX, which can be configured as an all-flash array, but in most cases probably won't be and is not designed as such. HPE seems to be betting that current VMAX users will give them a chance when it comes to a hardware refresh, rather than, say, turning to EMC's XtremIO ground-up developed all-flash array.


HPE global account pre-sales manager Phil Lewis said: "Our analytics tell us we can save the customer 50% in terms of TCO, based on power, heat, cooling and administration costs."


Elsewhere, HPE also announced it would add the ability for customers to migrate data from IBM XIV arrays to 3PAR arrays. It already offers that capability for EMC and Hitachi Data Systems customers. Also now available are Samsung 3D NAND 400GB drives for 3PAR all-flash arrays.


HPE offers all-flash and hybrid flash arrays in its enterprise-class 20000 series, launched in June.

Despite a rubbish name Mangstor makes super-fast NVMe-based flash arrays

Antony Adshead | No Comments
| More

Top prize this week for rubbish company name goes to Mangstor.


The Texas-based all-flash array provider's CTO Paul Prince told me the "Mang" part alludes to "mango", ie the fruit, and was chosen in a hurry to meet trademark inspection deadlines (while the "stor" bit is obvious). The idea had been to use mango branding but that seems to have fallen by the wayside.


So, it might not have a great name but Mangstor makes bold performance claims for its so-called "NVMe over fabric" and "network-optimised storage".


Mangstor provides its NX6320 storage nodes, which are based on Dell R730 servers with capacity via its own MX6300 2.7TB NVMe PCIe cards, with four of them per box. The company offers four models in capacities of between 8TB and 32TB and claims between 1.5 million and 3 million IOPS read IOPS for 4K blocks (1.1 and 2.5 million IOPS are the write figures).


Mangstor's chief selling point is blistering performance. It claims to provide this by the use on its MX cards of storage controller software that manages flash operational and maintenance functionality in conjunction with a Coherent Logix 100-core CPU.


"It's about being extra-efficient in parallel operations," said Prince. "The software schedules reads and writes so queue and device management are optimised. The 100-core CPU dedicates cores to manage different functions."


Mangstor also applies intelligence higher up the stack, at the network traffic level of the controller, where it optimises traffic in and out of the device.


So, what's the chief advantage of using PCIe cards in place of flash drives with connects such as SAS?


Prince says, "The primary advantage of our cards is performance; high throughput and low latency. You could get cards from Intel, Toshiba, Samsung, but they'd be slower."


That's Mangstor's tilt at the market, with claimed blistering speeds and feeds. Prince admits there are a lot of contenders in the all-flash array space already, but believes Mangstor is, "highly differentiated in terms of storage numbers. Across every vertical there are businesses with some apps where performance is important."


Currently, like many first generation all-flash arrays, replication is not built in but can be effected with a third party product. Mangstor is in partnership with Caringo to offer object storage.

The glacial sales cycle of HGST's cold storage Active Archive

Antony Adshead | No Comments
| More

It's not too often you get to hear the exact progress of a new product in the market but this week at Cloud Expo Europe in Frankfurt, HGST senior business development manager for elastic storage products, Manfred Berger, revealed the company - drive maker turned system seller - had not closed a single sale on its Active Archive petabyte-scale array. (See update here).


Now, HGST is unruffled by that. Active Archive is a big investment. It marries Amplidata Himalaya object storage with HGST 8TB HelioSeal helium-filled SATA hard drives and crams in 42U worth of these for a total capacity of 4.7PB. It claims 15 nines reliability using erasure coding and access times of less than 100 milliseconds for 90% of the time.


So, I asked how many customers HGST had secured for Active Archive. "None", said Berger, "but I have 160 prospects and one European order since [launch in] April."


"The sales cycle is longer than with hard drives and it takes at least six months from introduction to purchase order. It's a very long sales cycle - with customers that want an Amazon, but in-house - with organisations that may be very interested but who have to wait until budget cycles make buying possible," added Berger.


Given the budget cycles of likely customers - often university departments - it's no surprise sales cycles are slow. The projects Active Archive aims at are far from trivial, being cases where organisations must build a cloud in-house for legal and compliance reasons and can't entrust data to the public cloud, such as in genome research or where universities want to build Dropbox-type services for PhD papers, theses etc.


* In other news HGST has entered a partnership that will see it sell Point Storage Manager, a software tool that monitors existing NAS assets and identifies "lazy files", ie ones that have not been used for some time and can transfer them to cloud storage or Active Archive. It then leaves a stub on the NAS filer that pulls the full file from Active Archive, the cloud etc when it is accessed.

Rozo brings scale-out NAS plus erasure coding

Antony Adshead | No Comments
| More

I spoke with French startup Rozo this week.


It's USP is that it marries the scalability of scale-out NAS - ie virtual unlimited and with potentially billions of file names - with erasure coding and targets high performance computing (HPC) use cases and analytics, especially with large volumes of large unstructured files.


There seems to be a rumbling battle going on between scale-out NAS and object storage for use cases that involve lots of unstructured data. Both approaches have the ability to deal with very large amounts of files and can usually grow performance and capacity independently.


But usually erasure coding as a data protection method is the preserve of object storage. In this case Rozo applies it to scale-out NAS.


According to Philippe Nicolas, systems advisor with Rozo, erasure coding allows Rozo to sustain high levels of I/O for all data even during disk failures. At the same time it allows clusters to be built more cheaply than most scale-out NAS, which uses replication as its key data protection method.


He said: "Replication is very good for small files. It's fast and there's no coding, just a copy. But with large files replication takes time and there's a CPU impact."


That's because replication requires the customer to provide at least double the storage capacity for those replicated copies. Erasure coding, at least in the Mojette Transform (Mojette is a famous bean in the Nantes area) algorithm, requires only a 1.5x capacity overhead.


According to Nicolas, Rozo is the only scale-out NAS supplier to offer erasure coding for all data. EMC's Isilon uses it only for data above 128Kb.


Customers can build Rozo clusters by installing the software on commodity servers. Storage comes from internal server storage or can be on a JBOD or even the user's existing shared storage array.


The Mojette Transform algorithm encodes all data upon ingest but also maintains metadata for rapid index-related functionality.


Rozo can be downloaded and tried for free in its community edition without optimised erasure coding, while the Advanced commercial version includes it.


Rozo's target markets are high performance computing (HPC) media and entertainment, genomics and specifically, said Nicolas, use cases where you find the Lustre parallel file system, Qumulo scale-out NAS or EMC's Isilon.

VMware's hybrid cloud vision falls short in storage

Antony Adshead | No Comments
| More

VMware's vision of the near future of IT revolves heavily around the hybrid cloud in which customers run both on-premise and off-premises cloud.

Listening to VMware execs at VMworld Europe this week the company's efforts towards this are very much a key message and source of slogans such as "One cloud, one app, any device".

It is perhaps so far, largely just talk, however, and one area that is certainly the case is in storage.

While VMware's Virtual SAN (VSAN) has reached version 6.1 and now supports synchronous replication between site clusters, NVME and RAM storage there is little of the oft-talked of "seamlessness" between in-house VSAN deployments and the cloud.

But is it reasonable to expect such cross-platform functionality? Well, storage hardware vendors such as EMC and NetApp have for at least a year offered the cloud as a tier, also with much talk of hybrid cloud functionality.

EMC offers its EMC Hybrid Cloud Solution, while NetApp offers Cloud Ontap. Both come with partnerships with cloud providers and aim at data portability between on- and off-premise datacentres.

More recently, Avere launched its CloudFusion cloud NAS appliance. This works only in the cloud and tiers between classes of Amazon storage. It's not too much of a stretch to see how in-house and cloud versions of such software could work together.

These types of thing illustrate what might be possible when it comes to seamless (or nearly so) working in hybrid cloud storage.

It shouldn't be too long before VMware comes up with a take of its own, but storage manager Gaetan Castelein said there is nothing in the VMware portfolio of that nature at present and was tight-lipped about plans.

He said: "We don't have an explicit product that can be deployed in the cloud that can tier storage. You can have a connection point between VSAN and vCloudAir to allow replication and providers can offer block and object storage but the customer has no visibility into what that underlying storage is."







Dell and EMC's Big Splash point to hyper-convergence

Antony Adshead | No Comments
| More

On the eve of VMworld Europe it's like witnessing the IT/storage world's equivalent of the Big Splash. Yes, I'm talking about Dell's proposed $67 billion purchase of storage giant EMC.

The aforementioned collision between a proto-earth and a planet the size of Mars is thought to be responsible for the water-bearing, life-supporting sphere we now live on. Will Dell's EMC provide the same?

In storage terms there are likely to be some casualties, largely on the Dell side of things.

EMC has not only been the biggest of the big six storage makers but has been at the forefront of (often acquisition-led) innovation/productisation wave in storage with an early starts in all-flash (XtremIO) and out-there flash acquisitions (DSSD) as well as holding a wide ranging stable, from enterprise class downwards in storage arrays, that can be hybrid and all-flash (VMAX and VNX), Fibre Channel and iSCSI.

Dell doesn't come close in terms of leading edge-ness, and doesn't rival EMC in the enterprise storage space. In the mid-range its Compellent arrays (with Fibre Channel and iSCSI access) compete with VNX, and their future must be surrounded by question marks.

Dell's other key block storage offerings are distinctly entry-level (the iSCSI EqualLogic), and could survive because of that. In file access it's a similar picture with Dell's Windows NAS up against EMC's Isilon scale-out NAS.

Across other key areas EMC is characterised by having its own product offerings while Dell resells someone else's kit - eg, ScaleIO in software-defined storage vs EVO:RAIL; Atmos and ViPR/ECS in object storage vs Scality; in hyper-converged ScaleIO Node vs Nutanix.

But of course the deal must be viewed from a much wider perspective than storage alone. We live in an age where hyper-convergence is a key trend, where combined server, storage and networking are the direction in which the datacentre is moving, dragged along by the web-scale pioneers, Google, Facebook et al.

And it's at this level the giant deal looks to have its key synergy, with Dell adding EMC's storage expertise and VMware's virtualization glue* to its existing market leadership (well, almost) to provide the basis for a hyper-converged infrastructure giant.

Now all it needs is to add networking.

(* Interesting footnote question is can or will Dell's VMware live with Microsoft Hyper-V in its tent via existing relationships on the Dell server front?)

SolidFire puts flash for object (and file) storage on the roadmap

Antony Adshead | No Comments
| More

All-flash array maker SolidFire plans to apply the speed of flash to object storage. So said CEO Dave Wright this week.


We were discussing the prospects for the all-flash datacentre, following IDC's recent publication of a white paper which set 80TB to 90TB of mixed workload data as a feasible threshold for organisations to go all-flash.


Wright generally concurred with this assessment, but added that other areas such as object and file storage would also become candidates for flash treatment. Albeit, as the cost per GB came down.


Wright said: "We're looking forward, as the cost per GB decreases, for more workloads to be run on SolidFire. These include file and object workloads that are primarily on disk right now. You can read into this what we're thinking about, though there's no timing here but we want to be there when this occurs."


He added: "There will be some object storage that can make use of flash. Primarily this will be things like content distribution, for example in e-commerce firms that needs to serve billions of small images. You need object rather than block or file storage for this and traditional deep and cheap ways of supplying object storage are not good for real-time access to data."


"Also, there is a case for flash and object storage for enterprise content repositories such as Sharepoint where there's not a need for intensive I/O but a need for rapid access and regular changes."

StorPool's Linux-based software-defined storage dodges VMware and Hyper-V, for now

Antony Adshead | No Comments
| More

Software-defined storage products seem to be coming thick and fast at the moment.

This week I spoke with another, the Bulgaria-based StorPool, which offers storage software that can be deployed on commodity servers to provide SAN-like block storage from distributed disk but which can sit alongside app processing tasks to provide hyper-converged storage.

StorPool requires at least three instances of server hardware be deployed. These can be for storage only, or - by using only 5% to 10% of CPU performance for storage tasks - can co-reside with compute. Software also has to be deployed on the client side to manage storage consumption.

With an upgrade this week StorPool now claims performance of 250,000 random read IOPS and 4,200 MBps sequential reads on three servers, with these figures increasing as hardware nodes are added with scale-out capability.

Currently however, StorPool is limited to deployment in hardware running Linux operating systems (OS), from where it can be provisioned for Linux-friendly virtualisation platforms such as KVM, Xen or Docker.

CEO Boyan Ivanov said the ability to deploy StorPool into Windows servers is currently under development and will be available within months.

For the moment the Linux-only abilities of StorPool are something to which the company is cutting its cloth to suit. And so the bulk of its target customers are service providers which already run a lot of Linux - among whom it claims deployments of up to several hundred TB - although it also aims at devops use cases in wider verticals.

As Ivanov puts it: "We are a young vendor so we have to do one thing very well."

So, for now it's not targeting the enterprise market with its preferences for VMware, to which StorPool is not natively compatible, though it can run in VMware "with performance degradation," says Ivanov.

Disk types accommodated can be HDD, flash or a combination of the two and storage features include synchronous replication, snapshots and thin provisioning.

Write-back cache enables write operations to be acknowledged by the storage system, but stored in memory and then flushed to the hard disk at a later stage.

Like other software-defined storage products we've looked at recently, StorPool protects data by replicating between a minimum of three hardware instances.

Will 3D XPoint kill flash and leave disk (and tape) standing?

Antony Adshead | No Comments
| More

With the recent announcement of 3D XPoint memory by Intel and Micron, there are, inevitably, more questions than answers.


On the one hand there is the big question for computing architectures at a general level. IE, what will become of the current paradigm of CPU plus RAM and (tiers of) storage if a new medium 1,000x faster and with adequate endurance arises?


On the face of things 3D XPoint will make RAM redundant in compute architectures.


But what will it mean for enterprise storage hardware architectures?


3D XPoint will deal a heavy, possibly fatal, blow to existing NAND flash products. With the Intel-Micron brainchild's 1,000x performance premium and equal or better lifetime endurance, NAND flash's days are numbered. But we knew that anyway (and there are other possible successors).


It's the timing of that eclipse that is currently un-knowable, being dependent on the cost and availability of 3D XPoint. And at present we know nothing certain of either except that the first products are likely to ship in 2016. But will they be ready for mainstream enterprise adoption?


On both counts it's doubtful, for some time at least. We will see 3D XPoint in the much more voluminous consumer market at first. And in fact it's the fabrication plant economies of scale that come with consumer product adoption that help bring the price down to levels acceptable to the relatively minority use case of enterprise storage.


So, for some time 3D XPoint will likely be an exotic and costly but rapid storage medium, and so available in small quantities to enterprise storage and server hardware makers. In that case, what will be it's role?


Probably more or less that of RAM right now. As a fast access cache or tier - in server or array - on top of NAND flash and possibly spinning disk.


But that phase will pass and as the cost comes down and the ubiquity of 3D XPoint, its clones or functional equivalents rises so the process and economies of silicon production probably mean flash will suffer a relatively sudden death.


Ironically, that might leave spinning disk -- and tape! -- still standing after flash has died, with a niche but valuable role at the glacial end of archival storage.

Infinidat stakes claim to the future of hyperscale storage in the enterprise

Antony Adshead | No Comments
| More

Enterprises need hyperscale or webscale storage. They look at the likes of Google and Facebook and need to do what they're doing; processing and storing huge amounts of data with transactions and analysis on the fly. But they lack storage products to do it, and existing enterprise storage isn't up to the job.


Those are the views of Infinidat, which came out of stealth this year and which believes enterprises are trying to solve the problems of the future datacentre with storage architectures from the past.


Infinidat CTO Brian Carmody said: "What we're seeing is a competitive advantage by a small number of technically advanced companies - Facebook, Google etc - who are trying to do what they do with incumbent technology. The experience of these apex big data companies demonstrate a failure by the storage industry."


What Infinidat offers is high capacity (2PB in a 42U rack), high performance (up to 750,000 IOPS, throughput of 12GBps), highly available (99.99999% - seven nines), which it calls mainframe levels of performance and availability but built for enterprise users to handle webscale operations.


It does all this by breaking the mould for enterprise storage in a number of ways. its Infinibox comes with three controllers (in contrast to the standard dual setup) in an active-active-active architecture. These contain DRAM and flash, with all active data held in these two layers, while below that are huge amounts - 480 nearline-SAS HDDs - of spinning disk.


It spreads the workload across all three nodes and out to the massive number of drives. To do that, Inifindat had to throw out the existing Linux SAS drivers and re-write the way the controller nodes handle data to the storage media.


"Each node has a multipath connection to 480 spindles and that is many times more concurrent SAS connections than Linux can handle," said Carmody.


For now, Infinibox is block access only, but NFS file access is planned "shortly". Mainframe access will be available by the end of the year and object "shortly after that", said Carmody.


To achieve such levels of claimed availability Infinidat breaks I/O down into 64kb objects it calls sections with all host I/O being dealt with to and from cache. This is then spread out across the HDDs in a 16-way stripe with parity, based on the RAID 6 model but dubbed Infiniraid. Data is given an "activity vector score" to rank it on whether hot or cold, sequential or random etc. This helps with pre- and de-staging data between tiers and also for rebuilds.


"At hyperscale n+1 is not sufficient," said Carmody. "We have three of everything. It's what hyperscale in the enterprise will look like."


That's an intriguing claim. Watch this space.







Quobyte apes Google storage model and plans a tilt at NetApp and EMC

Antony Adshead | No Comments
| More

Does the world need another software-defined storage/storage virtualisation product? There are plenty to choose from already, such as DataCore, Nexenta, Open-E as well as software storage products from big hitters like VMware with its VSAN.


That's what I thought before speaking to Germany-based startup Quobyte, but its case is quite compelling. It's aim is to provide Google-like webscale server-based grids of hugely scalable storage that are manageable by relatively few IT staff.


Quobyte aims to do this - and to attack the customer bases of the likes of NetApp and EMC Isilon - by decoupling hardware from the software intelligence that runs the storage grid, said founder and CEO Bjoern Kolbeck, who is a former Google employee and contributor to the EU-funded Xtreem OS Linux grid environment.


"At Google we saw how large scale infrastructures worked and it inspired us to take Xtreem FS [the file system from Xtreem OS] and build in ideas around operations and automation from Google."


"Google has datacentres in which a very small team look after storage and don't need to interact with the software team. It's a very nice operational model that scales; if you add more data you don't need to add more people."


Kolbeck explained that in the Quobyte view its software-defined storage system needs to provide fault tolerance of "split brain" situations, so that, for example traffic can be automatically re-routed if a broken top-of-rack switch where that might cause database writes to differ between instances. The same fault tolerance also comes into play for day-to-day maintenance. "If a server is down that should not be an exception," said Kolbeck.


Underlying this is the nub of what Quobyte is all about - namely quorum-based replication. IE, there are three copies of replicated data and there must always be a majority, so two out of three copies and it's OK to shut down one copy.


Quobyte's quorum-based triple-copy replication is based on the Paxos lease algorithm, which sees replicated files communicate and decide on a master copy and use that master to ensure data integrity between them. "Then, if that master fails the nodes elect a new master," said Kolbeck.


Currently, Quobyte lacks features such as synchronous replication that can help it break into the enterprise storage market, but that is planned for later this year. Right now, at least three instances of Quobyte have to be deployed in the same datacentre or metro area for latency reasons, said Kolbeck.


At present Quobyte has implementations aimed at HPC, big data (eg, Hadoop) and OpenStack customer deployments but in the long run it will target NetApp filer users and scale-out storage customers of EMC Isilon, said Kolbeck.


"Asynchronous geo-replication is on the roadmap for enterprise users but we can already be a good fit for virtual machine storage and are running some proofs-of-concept where we are looking to replace NetApp," said Kolbeck.


It'll be interesting to watch the progress of this marrying of software-defined/storage virtualisation and hyperscale storage.

Veeam won't get physical (backup)

Antony Adshead | No Comments
| More

Veeam is a backup software company and pioneer of virtual machine backup. Way back when, while most backup software was firmly rooted in the old world of physical servers and an agent on every box, Veeam and a few others (Quest, PHD Virtual et al) began to offer specialised products aimed at protecting VMs.

Gradually, the mainstream incumbents of the backup world (Symantec, CommVault, HP, EMC, IBM TSM et al) incorporated virtual machine backup into their products. This, one might have imagined, could have adversely affected the fortunes of the VM backup specialists.

But it didn't, and Veeam in particular, roared from nowhere into second place in a TechTarget survey question on customers' backup provider preferences. The mystery then became, why would so many customers, who must have backup products from the incumbents deployed, buy Veeam in such large numbers and add another backup product to their infrastructure?

Speaking to Veeam VP for product strategy Doug Hazelman this week ahead of its VeeamON forum in London he gave his interpretation on that phenomenon. In short, he thinks it's often just too much trouble for enterprises to deploy the new virtualisation-focussed features in their products.

He said: "Many want to have a single vendor environment but the reality is they have more than one, probably 2.5 to 3 on average. Do they like that situation? Everyone thinks it's the Holy Grail to standardise on one backup product but in truth they'd lose a lot of features by doing so."

"Virtualisation and the cloud are the future but in many cases we find customers don't take advantage of the virtualisation features in incumbent products. That's because they'd have to retool their entire existing deployment and so they think, 'Why not just look at best of breed?' Also, lots of companies are just not happy with the backup products they have in place."

It's a plausible explanation, though not one that we can easily test, but it would account for Veeam's good showing in the backup product stakes of late.

The other puzzler for me in backup recently is why server backup software providers do not backup mobile and endpoint devices. Currently you can backup laptops with some of the mainstream backup products but none as far I know extend to smartphones, tablets and the like.

Meanwhile, however, companies such as Druva make a good living by offering endpoint/BYOD backup etc. And it's an area that appears necessary to achieve compliance as data stored anywhere may face a requirement to be retained for legal e-discovery etc.

So, why doesn't Veeam consider it an area it needs to address?

Hazelman's view was this. "If people bring their own stuff into the workplace it's difficult to manage from a backup perspective. Personally, I don't need backup for my own laptop. If I lost it today I wouldn't lose data as it's all also somewhere else. It's a better approach to protect what's on servers than what's on laptops."

I'm a little less than convinced by this. But it'll be interesting to hear the view of others interested in backup and data protection. 

EMC: Open source for the ViPR isn't sauce for the Golden Goose

Antony Adshead | No Comments
| More

This week I spoke to EMC marketing VP Josh Goldstein about XtremIO 4.0.

EMC's flagship all-flash array has seen it move from third place in the all-flash stakes to first, with Gartner in 2013 ranking it at 11.1% market share behind IBM (24.6%) and Pure Storage (17.1%).

By 2014 EMC had achieved pole position with 31.1% market share. Pure Storage was second with 19%, and IBM third with 16% while no other vendor had more than 7%.

That's remarkable progress indeed.

Recently we've also heard much of EMC's forays into open source, with the opening of ViPR code to developers in Project CoprHD, as well as continued commitment to software-defined storage.

In a blog post coinciding with EMC's open sourcing of ViPR Manuvir Das, engineering VP with EMC's advanced software division, spoke of how software development has changed, how open source has become mainstream, how its accelerates innovation and how excited EMC is about its "open source strategy".

So caught up in the excitement was I that when I spoke to Goldstein about XtremIO I thought I'd ask him if it too would become open source.

Of course I already knew the answer and Josh's immediate response was laughter. I pressed further and he said, "There is no plan to open source XtremIO. I'm not sure what the business model would be."

It's not like open source business models are a great mystery. There is a core of software code around a project/product and its open for free use and modification to anyone. Meanwhile, commercial distributions take in updates and changes in a more controlled fashion and charge customers for installation, configuration, support etc.

It could quite easily be applied to any product, even an all-flash array product like XtremIO.

But of course it won't be because it is precisely XtremIO's software that is the goose laying the golden egg for EMC and the upgrade to version 4.0 demonstrated further innovation on the fundamentals of the XtremIO XIOS operating system (OS).

In the upgrades announced at EMC World in May the XtremIO headlines were topped by an increase in X-Brick capacity to 40TB and totals for a rack topping 1PB.

But the other key announcements all built on XIOS's block storage technology, which sees data deduplicated and compressed on ingest. Building on that fundamental characteristic, EMC announced:

  • Integration of XtremIO and its RecoverPoint data protection software with leverage of the same dedupe and compression as in the XIOS engine to speed data transmission over the wire.
  • Copy data management with, for example, versions of Oracle, SQL Server and SAP databases generated over a lifecycle able to reside on XtremIO by use of space-efficient copies based, again, on that dedupe and compression algorithm in XIOS.

So, really, it's not like EMC couldn't open source XtremIO like it has done with ViPR. It's just that XtremIO's software is a goose that's laying a golden all-flash egg for EMC, while ViPR can show no similar market share figures and has been open sourced in an apparent attempt to breathe life into it.

It's obvious really, but EMC could never come out and say this.

 

PS: Goldstein also said there is no plan to make XtremIO available as a software-only product for customers to install on their chosen X86 hardware.

"It's a choice," said Goldstein. "Our customers like pre-packaged arrays. They're standard X86 servers and CPUs anyway and all the value is in the software so there are no big savings to be made by buying your own hardware."

 

ViPR and CoprHD: EMC forced to swim with the open source tide

Antony Adshead | No Comments
| More

EMC's move to make its ViPR software-defined storage platform open source (as Project CoprHD) appears a bold one, but I can't help thinking it looks like the giant from Hopkinton, MA, has been been subject to forces too strong to resist.

 

ViPR - which was launched two years ago at EMC World in 2013 - allows customers to build pools of web-scale storage from heterogenous storage media; EMC or third-party storage vendor arrays or even commodity hardware. ViPR uses its own Object Data Services that can be accessed via REST APIs including Amazon S3 or HDFS to enable analytics of data under its management.

 

At the time we noted that ViPR's success would be pinned on, as Freeform Dynamics' Tony Lock, put it: ". . . getting third parties involved. That could make or break it. It'll take political will and diplomacy."

 

And that diplomatic success has proved to be limited, with native support from other storage vendors restricted to NetApp and HDS and other makers' arrays accessible only via the OpenStack Cinder plugin.

 

In some ways, support from other vendors is the first hurdle, but not necessarily the most important one. Key of course are actual ViPR customers, and those seem to have been limited to short footnotes in EMC press releases.

 

Meanwhile, of course, open source in the datacentre has come on leaps and bounds, in particular via OpenStack's cloud infrastructure platform and its storage modules. Open source in general has improved its image from beard-and-sandals to near mainstream in the enterprise.

 

No more proof of that is EMC's conversion to open source, and its keenness to parade its credentials in that respect.

 

Open source development gains from the involvement of the community and its feedback, in terms of practical improvement to the product and in terms of buy-in and credibility.

 

Clearly then, EMC has viewed the relative rise of open source as a threat. It initiated ViPR as an EMC product that would tie together a storage and analytics infrastructure over heterogeneous storage.

 

Two years later EMC has had to bow to that threat, or seize the opportunity, depending on how you wish to spin it, and now ViPR has spawned Project CoprHD.

 

The Massachusetts storage giant will continue to sell ViPR as a commercial project, but now it will hope its open source alter ego will help its develop and gain respectability in a way it has failed to so far.

 

 

Software-defined storage: Not for us says Nimble Storage CEO

Antony Adshead | No Comments
| More

Separating software from hardware is an emerging trend in data storage. Under the phrase software-defined storage it has been at the forefront of vendor hype in recent years.

The appeal is that it potentially allows customers to build their own storage arrays by deploying storage software - which is after all where the main smarts in a storage system reside - on commodity server hardware.

So, it was interesting this week to come across one array maker - hybrid flash specialist Nimble Storage - which said it would not go down the road of offering a software version of its product.

In some ways that goes counter to a rising trend.

Vendors that offer storage software range from hardcore hardware pushers such as HP with its StoreVirtual VSA, virtualisation giant VMware with its Virtual Storage Appliance and VSAN, suppliers that have made their name with storage software such as Nexenta and DataCore, and more recently startup flash array providers like SolidFire that now offers its Element X operating system (OS) as a virtual appliance. Even hardware giant EMC offers software version of its VNX and Celerra products for lab use.

But Nimble Storage says it won't go down that road. CEO Suresh Vasudevan told me this week that while its engineers use software versions of the Nimble OS in test and dev, the company would not offer a software-defined storage product to customers.

Basically, he said there's nothing in it for Nimble. His argument went like this:

Vasudevan said: "If, for example, a storage system costs $100 and the hardware from China is $40 of that, here's how the rest is spent: $28 on sales and marketing, $12 on engineering and R&D, $5 on company admin. So, if I sell just software and the cost of the hardware is the same or possibly more then we would have to sell software at much less. But I still have to pay the same amount to sales and marketing people and to engineers, so it's really not clear there's a benefit."

Vasudevan also said that the software-only hyper-converged model - where server and storage reside together on the same box - is likely to lead to increased customer costs due to the need to protect numerous discrete compute/storage instances.

"When you have software on commodity servers the belief is it will lower costs, but in fact it often leads to overprovisioning. That's because the way people protect data on nodes without redundancy features such as dual controllers is to mirror data, often with triple replication," said Vasudevan.

So, that's why one array supplier will not go down the route of software-defined storage. I guess the argument works, for them, but it could be argued that they could at give the customer that choice. The most compelling part of Vasudevan's case is probably that the hardware will cost the same or more to a customer, but for the largest organisations out there that may not be the case.