Category: Data centre

Posted on

Seagate’s new KOS disk drives aim to entice cloud builders

Among the most interesting conversations I had at the storage show SNW (aka Powering the Cloud) in Frankfurt this year was with Seagate’s European cloud initiatives director Joe Fagan, as we talked about the company’s proposed Kinetic Open Storage (KOS) drives.

The disk drive company is trying to move up the stack from what has become commodity hardware by converting its drives into servers. Instead of attaching using a SATA or SAS connector, Kinetic drives will have – a SATA or SAS connector, not an RJ45. But the data flowing inside the connector will be using IP not storage protocols, while the connector remains the same for compatibility purposes.

The aim is to help builders of large-scale infrastructures, such as cloud providers, to build denser, object-based systems by putting the server on the storage, rather than, to paraphrase Fagan, spending the energy on a Xeon or two per server along with a bunch of other hardware. Seagate argues that KOS could eliminate a layer of hardware between applications and storage, so data will flow from the application servers directly to storage rather than, as now, being translated into a variety of protocols before it hits the disk.

Fagan said two cloud builders were interested in the technology.

Behind this is, of course, a bid to grab some of the cash that enterprises and consumers are spending on cloud applications and services.

There are a few ‘howevers’, as you might imagine. Among the first is that every disk drive will need an IP address. This has huge implications for the network infrastructure and for network managers. Suddenly, there will be a lot more IP addresses to deal with, they will have to be subnetted and VLANned – did I mention that Kinetic drives will use IPV4? – and all this assumes you can summon up enough v4 addresses to start with.

Another concern is that mechanical disk drives fail relatively frequently while servers don’t, as of course they have no moving parts. So when a drive fails – and in large-scale deployments they surely will – you have to throw away the internal server too. Could be expensive.

And finally, there’s also a huge amount of inertia in the shape of today’s installed systems and the expertise needed to manage and operate them.

Is that enough to halt the initiative? Seagate clearly hopes not, and hopes too that other drive makers will come on board and develop their own versions in order to help validate the concept. It has provided APIs to help app developers exploit the concept.

As ever, time will tell. But will you find these drives in a server near you any time soon? Don’t hold your breath.

Posted on

Seagate launches new solid-state disks (SSD)

Seagate, the biggest maker of hard disks, recently launched a new range of solid state disk drives, as it aims to align itself better with current buying trends.

In particular, the company’s new 600 SSD is aimed at laptop users who want to speed their boot and data access times. This is Seagate’s first foray into this market segment.

Claiming a 4x boot time improvement, Seagate said that SSD-stored data is safer if the laptop is dropped. From my own experience over the last five years of using using SSDs in laptops, I can confirm both this, and that their lower power consumption helps to improve battery life too.

The 600 SSD is available with up to 480GB and in multiple heights including 5mm, which the company says makes it “ideal for most ultra-thin devices as well as standard laptop systems”. The drive features up to 480GB of capacity and comes in a 2.5 form factor. It’s compatible with the latest 6Gbps SATA interface.

The other new SSD systems are aimed at enterprises. The most interesting of these is the X8 Accelerator, which is the result of Seagate’s investment in Virident, a direct competitor with Fusion-io, probably the best-known maker of directly-attached SSDs for servers. The Seagate product is also a PCIe card with a claimed IOPS of up to 1.1 million. The X8 offers up to 2.2TB in a half-height, half-length card.

Of the two other new drives, the 2.5-inch 480GB 600 Pro SSD and the 1200 Pro SSD, the first is targeted at cloud system builders, data centres, cloud service providers, content delivery networks, and virtualised environments, and is claimed to consume less power and so need less cooling. It consumes 2.8W, variable according to workload, which Seagate reckons is “the industry’s highest IOPS/watt”.

Up the performance scale is the 800GB 1200 Pro SSD, which is aimed at those needing high throughput. It attaches using dual-port 12Gbps SAS connectors and “uses algorithms that optimize performance for frequently accessed data by prioritizing which storage operations, reads or writes, occur first and optimizing where it is stored.”

Seagate said it buys its raw flash memory from Samsung and Toshiba but holds patents for its controller and system management technologies.

Posted on

Hard disks and flash storage will co-exist – for the moment

When it comes to personal storage, flash is now the default technology. It’s in your phone, tablet, camera, and increasingly in your laptop too. Is this about to change?

I’ve installed solid-state disks in my laptops for the last three or so years simply because it means they fire up very quickly and – more importantly – battery life is extended hugely. My Thinkpad now works happily for four or five hours while I’m using it quite intensively, where three hours used to be about the maximum.

The one downside is the price of the stuff. It remains stubbornly stuck at 10x or more the price per GB of spinning disks. When you’re using a laptop as I do, with most of my data in the cloud somewhere and only a working set kept on the machine, a low-end flash disk is big enough and therefore affordable: 120GB will store Windows and around 50GB of data and applications.

From a company’s point of view, the equation isn’t so different. Clearly, the volumes of data to be stored are bigger but despite the blandishments of those companies selling all-flash storage systems, many companies are not seeing the benefits. That’s according to one storage systems vendor which recently announced the results of an industry survey.

Caveat: industry surveys are almost always skewed because of sample size and/or the types of questions asked, so the results need to be taken with a pinch – maybe more – of salt.

Tegile Systems reckons that 99 percent of SME and enterprise users who are turning to solid state storage will overpay. They’re buying more than they need, the survey finds, at least according to the press release, which wastes no time by mentioning in its second paragraph that the company penning the release just happens to have the solution. So shameless!

Despite that, I think Tegile is onto something. Companies are less sensitive to the price per GB than they are to the price / performance ratio, usually expressed in IOPS, which is where solid-state delivers in spades. It’s much quicker than spinning disks at returning information to the processor, and it’s cheaper to run in terms of its demands on power and cooling.

Where the over-payment bit comes in is this (from the release): “More than 60% of those surveyed reported that these applications need only between 1,000 and 100,000 IOPS. Paying for an array built to deliver 1,000,000 IOPS to service an application that only needs 100,000 IOPS makes no sense when a hybrid array can service the same workload for a fraction of the cost.”

In other words, replacing spinning disks with flash means you’ve got more performance than you need, a claim justified by the assertion that only a small proportion of the data is being worked on at any one time. So, the logic goes, you store that hot data on flash for good performance but the rest can live on spinning disks, which are much cheaper to buy. In other words, don’t replace all your disks with flash, just a small proportion, depending on the size of your working data set.

It’s a so-called hybrid solution. And of course Tegile recommends you buy its tuned-up, all-in-one hybrid arrays which saves you the trouble of building your own.

Tegile is not alone in the field, with Pure Storage having recently launched in Europe. Pure uses ordinary consumer-grade disks, which should make it even cheaper although price comparisons are invariably difficult due to the ‘how long is a piece of string?’ problem.

There are other vendors too but I’ll leave you to find out who they are.

From a consumer point of view though, where’s the beef? There’s a good chance you’re already using a hybrid system if you use a recent desktop or laptop, as a number of hard disk manufacturers have taken to front-ending their mechanisms with flash to make them feel more responsive from a performance perspective.

Hard disks are not going away as the price per GB is falling just as quickly as it is for flash, although its characteristics are different. There will though come a time when flash disk capacities are big enough for ordinary use – just like my laptop – and everyone will get super-fast load times and longer battery life.

Assuming that laptops and desktops survive at all. But that’s another story for another time.

Posted on

Technology highlights 2013

I’ve been shamefully neglecting this blog recently, yet a lot of interesting new technologies and ideas have come my way. So by way of making amends, here’s quick round-up of the highlights.

Nivio
This is a company that delivers a virtual desktop service with a difference. Virtual desktops have been a persistent topic of conversation among IT managers for years, yet delivery has always been some way off. Bit like fusion energy only not as explosive.

The problem is that, unless you’re serving desktops to people who do a single task all day, which describes call centre workers but not most people, people expect a certain level of performance and customisation from their desktops. If you’re going to take a desktop computer away from someone who uses it intensively as a tool, you’d better make sure that the replacement technology is just as interactive.

Desktops provided by terminal services have tended to be slow and a bit clunky – and there’s no denying that Nivio’s virtual desktop service, which I’ve tried, isn’t quite as snappy as having 3.4GHz of raw compute power under your fingertips.

On the other hand, there’s a load of upsides. From an IT perspective, you don’t need to provide the frankly huge amounts of bandwidth needed to service multiple desktops. You don’t care what the end user wants to access the service with – so if you’re allowing people to bring and use their own devices into work, this will work with anything, needing only a browser to work. I’ve seen a Windows desktop running on an iPhone – scary…

And you don’t need to buy applications. The service provides them all for you from its standard set of over 40 applications – and if you need one the company doesn’t currently offer, they’ll supply it. Nivio also handles data migration, patching, and the back-end hardware.

All you need to do is hand over $35 per month per user.

Quantum
The company best known for its tape backup products launched a new range of tape libraries.

The DXi6800 is, says Quantum’s Stéphane Estevez, three times more scalable than any other such device, allowing you to scale from 13TB to 156TB. Aimed at mid-sized as well as large enterprises, it includes an array of disks that you effectively switch on with the purchase of a new licence. Until then, they’re dormant, not spinning. “We are taking a risk of shipping more disks than the customer is paying for – but we know customer storage is always growing. You unlock the extra storage when you need it,” said Estevez.

It can handle up to 16TB/hour which, is, reckons the company, four times faster than EMC’s DD670 – its main competitor – and all data is encrypted and protected by an electronic certificate so you can’t simply swap it into another Quantum library. And the management tools mean that you can manage multiple devices across datacentres.

Storage Fusion
If ever you wanted to know at a deep level how efficient your storage systems are, especially when it comes to virtual machine management, then Storage Fusion reckons it has the answers in the form of its storage analysis software, Storage Fusion Analyze.

I spoke to Peter White, Storage Fusion’s operations director, who reckoned that companies are wasting storage capacity by not over-provisioning enough, and by leaving old snapshots and storage allocated to servers that no longer exist.

“Larger enterprise environments have the most reclaimable storage because they’re uncontrolled,” White said, “while smaller systems are better controlled.”

Because the company’s software has analysed large volumes of storage, White was in a position to talk about trends in storage usage.

For example, most companies have 25% capacity headroom, he said. “Customers need that level of comfort zone. Partners and end users say that the reason is because the purchasing process to get disk from purchase order to installation can take weeks or even months, so there’s a buffer built in. Best practice is around that level but you could go higher.”

You also get what White called system losses, due to formatting inefficiencies and OS storage. “And generally processes are often broken when it comes to decommissioning – without processes, there’s an assumption of infinite supply which leads to infinite demand and a lot of wastage.”

The sister product, Storage Fusion Virtualize “allows us to shine a torch into VMware environments,” White said. “It can see how VM storage is being used and consumed. It offers the same fast analysis, with no agents needed.”

Typical customers include not so much enterprises as systems integrators, service providers and consultants.

“We are complementary to main storage management tools such as those from NetApp and EMC,” White said. “Vendors take a global licence, and end users can buy via our partners – they can buy report packs to run it monthly or quarterly, for example.”

Solidfire
Another product aimed at service providers, SolidFire steps aside from the usual pitch for all solid-state disks (SSD). Yes solid-state is very fast when compared to spinning media but the company claims to be offering the ability to deliver a guarantee not just of uptime but of performance.

If you’re a provider of storage services in the cloud, one of your main problems, said the company’s Jay Prassl, is the noisy neighbour, the one tenant in a multi-tenant environment who sucks up all the storage performance with a single database call. This leaves the rest of the provider’s customers suffering from a poor response, leading to trouble tickets and support calls, so adding to the provider’s costs.

The aim, said Prassl, is to help service providers offer guarantees to enterprises they currently cannot offer because the technology hasn’t – until now – allowed it. “The cloud provider’s goal is to compute all the customer’s workload but high-performance loads can’t be deployed in the cloud right now,” he said.

So the company has built SSD technology that, because of the way that data is distributed across multiple solid-state devices – I hesitate to call them disks because they’re not – offers predictable latency.

“Some companies manage this by keeping few people on a single box but it’s a huge problem when you have hundreds or thousands of tenants,” Prassl said. “So service providers can now write a service level agreement (SLA) around performance, and they couldn’t do that before.”

Key to this is the automated way that the system distributes the data around the company’s eponymous storage systems, according to Prassl. It then sets a level of IOPS that a particular volume can achieve, and the service provider can then offer a performance SLA around it. “What we do for every volume is dictate a minimum, maximum and a burst level of performance,” he said. “It’s not a bolt-on but an architecture at the core of our work.”

Posted on

2012: the tech year in view (part 2)

Here’s part 2 of my round-up of some of the more interesting news stories that came my way in 2012. Part 1 was published on 28 December 2012.

Datacentre infrastructure
NextIO impressed with its network consolidation product, vNet. This device virtualises the I/O of all the data to and from servers in a rack, so that they can share the bandwidth resource which is allocated according to need. It means that one adapter can look like multiple virtual adapters for sharing between both physical and virtual servers, with each virtual adapter looking like a physical adapter to each server. The main beneficiaries, according to the company, are cloud providers, who can then add more servers quickly and easily without having to physically reconfigure their systems and cables. According to the company, a typical virtualisation host can be integrated into the datacentre in minutes as opposed to hours.

In the same part of the forest, the longer-established Xsigo launched a new management layer for its Data Center Fabric appliances, its connectivity virtualisation products. This allows you to see all I/O traffic across all the servers, any protocol, and with a granularity that ranges from specific ports to entire servers.

Nutanix came up with a twist on virtualisation by cramming all the pieces you need for a virtualisation infrastructure into a single box. The result, says the company, is a converged virtualisation appliance that allows you to build a datacentre with no need for separate storage systems. “Our mission is to make virtualisation simple by eliminating the need for network storage,” reckons the company. Its all-in-one appliances mean faster setup and reduced hardware expenditure, the company claims. However, like any do-it-all device, its desirability depends on how much you value the ability to customise over ease of use and setup. Most tend to prefer separates so they can pick and choose.

Cooling servers is a major problem: it costs money and wastes energy that could be more usefully employed doing computing. This is why Iceotope has developed a server that’s entirely enclosed and filled with an inert liquid: 3M Novec 7300. This convects heat away from heat-generating components and is, according to chemical giant 3M, environmentally friendly and thermally stable. The fluid needs no pumping, instead using convection currents to transport heat and dump it to a water-filled radiator. The water is pumped but, Iceotope says, you need only a 72W pump for a 20kW cabinet of servers, a far cry from a typical 1:1 ratio of cooling energy to compute power when using air as the transmission medium.

Networking
Vello Systems launched its Data Center Gateway incorporating VellOS, its operating system designed for software-defined networking (SDN) – probably the biggest revolution in network technology over the last decade. The box is among the first SDN products – as opposed to vapourware – to emerge. The OS can manage not just Vello’s own products but other SDN compliant systems too.

Cloud computing
One of the highlights of my cloud computing year was a visit to Lille, to see one of OVH‘s datacentres. One of France’s biggest cloud providers, OVH is unusual in that it builds everything itself from standard components. You’ll find no HP, IBM or Dell servers here, just bare Supermicro motherboards in open trays, cooled by fresh air. The motivation, says the company comes from thefact there are no external investors and a high level of technical and engineering expertise at the top. Effectively, the company does it this way because it has the resources to do so, and “because we are techies and it’s one of our strong values.” The claimed benefit is lower costs for its customers.

I had an interesting discussion with Martino Corbelli, the chief customer officer at Star, a UK-based cloud services provider. He said that the UK’s mid-market firms are getting stuck in bad relationships with cloud services vendors because they lack both the management and negotiation skills required to handle issues and the budget to cover the cost of swapping out.

“The industry for managed services and cloud is full of examples of people who over promise and under deliver and don’t meet expectations,” he said, reckoning that discussions with potential customers now revolve more around business issues than technology. “Now it’s about the peer-to-peer relationship,” he said. “Can you trust them, are you on the same wavelength, do you feel that your CFO can call their CFO and talk to them as equals?”

We also saw the launch of new cloud computing providers and services from mid-market specialist Dedipower, CloudBees with a Java-based platform service, and Doyenz with a disaster recovery service aimed at smaller businesses.

Storage
Coraid boasted of attracting over 1,500 customers for its unique ATA-over-Ethernet (AoE) storage products. This means that storage is using native Ethernet rather than storage-specific protocols. Coraid reckons this reduces protocol overheads and so is three to five times faster than iSCSI. The company makes a range of storage systems but, although AoE is an open standard, no other company is designing and selling products with it.

WhipTail joined the growing list of vendors selling all-flash storage systems with its Accela products. Solid-state gives you huge performance advantages but the raw storage (as opposed to the surrounding support infrastructure) costs ten times as much compared to spinning disks, so the value proposition is that the added performance allows you to make more money.

Eventually, the bulk of storage will be solid-state, as the price comes down, with disk relegated to storing backups, archives and low-priority data, but that time has yet to come. It’s a delicate balancing operation for companies such as WhipTail and Violin Memory: they don’t want to be too far ahead of the mass market and don’t want to miss the boat when flash storage becomes mainstream.