Virtualization Technology News and Information
What DataCore thinks on a few topics

Quoting TechWorld

Techworld had the opportunity to meet with DataCore's Roni Putra, VP SW engineering recently. Here are his responses to a few questions.

TW: What do you think of file virtualisation like the way Acopia is doing it?

It's a form of NAS virtualisation. Why not do away with the NAS altogether and have block storage directly?

TW: What do you think of the idea of combining NAS and SAN storage?

We are thinking of doing a NAS/SAN combination, in software of course; hardware is a commodity. Windows NAS downsides include performance and its inability to scale. That's a NetApp edge. With a SANmelody engine inside a NAS we've seen performance and scalability better than a NetApp box. The underlying block engine is embedded into the Windows box and has caching.

Another NAS downside today - usually you don't have high reliability. If the NAS head goes down you lose your files. If we do NAS we'll do it with high-availability front and back ends.

We don't like hardware. It's complicated and behind the curve.

For DataCore any such NAS+SAN combination product would be a software one. It has no interest in developing hardware products.

TW: What do you think about VMware directly interacting with a storage resource to automate storage provisioning for virtual servers?

We are having a lot of success with VMware pull-down and thin provisioning. VMware can already use Microsoft's Virtual Disk services (VDS) in Windows to do some of this. Also SMI-S has a generic storage API. It was supposed to be open but suppliers like EMC and HDS have extended it with proprietary extensions so it is no longer absolutely standard.

VDS hasn't taken off for a couple of reasons. Microsoft added complexity which wasn't needed. They came out with VDS and left it to lie. They didn't push it. No-one uses it. There isn't a usable GUI. The end user isn't interested in typing command lines.

We are talking to VMware (about an API-type protocol for storage provisioning). It's a rather complex relationship. They see storage as a hardware entity. We're not hardware. They are a software group too – like us.

We can receive commands from VMware today but we would be doing it with our proprietary interface. VMware has a protocol for starting snapshots off. We support it today. If they extended it then it would be no big deal to support it.

VMware is getting bigger in itself and becoming more like Microsoft. The troops on the ground are okay. It's from a corporate standpoint that things are hard. It's the same with Microsoft.

We complement VMware very well.

It seems that the cachet for DataCore of getting a VMware certification will be quite a hurdle. Roni Putra talked about how VMware corporate people find it relatively straightforward to certify hardware product partners but not software ones. The VMware people ask: "What hardware do you run on?" and when DataCore replies: "Any," they find it hard to take on board as they have a method for certifying hardware product suppliers but not such a standard way of certifying software suppliers.

Also, continues Roni, “VMware had a reluctance to embrace storage virtualisation because of a poor relationship with a previous storage virtualisation vendor.” He wouldn't say which one, though.

Storage tiering

Turning to storage tiering and block storage virtualisation, there can either be as many virtualised pools of storage as there are storage tiers or one pool with the tiering invisible to applications and with file movement between tiers happening in the background according to settable policies.

TW: What do you think about storage tiering inside a virtualised pool of storage? Should it be visible?

We have the capability (for background tiering behind a single virtualised storage pool) with SANmotion; not a product but a capability. We haven't done this yet. We will be doing it when we bring out continuous protected recovering (CPR).

We track sequentially all write updates. So you can do away with mirrors and RAID 5, etc., because, if you can access the sequential stream fast enough you can rebuild lost data on the fly; say inside 60 seconds. If you can do that it doesn't matter how the data comes through. You avoid RAID rebuilds (when a disk fails).

We are working on it, currently working on an algorithm to speed access. At year-end we'll come up with a new thin provisioning module to address this. Snapshots and CDP (continuous data protection) all occur in flight. You will truly be able to roll back to five minutes ago instantly. We'll implement it in Traveler 3.0.

We formally launched Traveler six months ago. It captures the I/O stream and enables you to choose how long you want to roll back to. It's absolutely simple I/O. With Traveler 3.0 you'll be able to archive data for months and years. We'll make the formats public so people can recover data without DataCore and also create their own schemes on top of it.

The thing is a business continuity and recovery solution. It's better than NetApp Snap and Snap because it is a standard product.

TW: What does DataCore think of filesystems?

Life would be simpler for us if filesystems didn't exist. Intel had the notion of virtual I/O a few years ago. You write directly to the disk with no filesystem in the way.

There is a Bill Gates' directive that all Microsoft software applications run through the NTFS filesystem. SQL Server could be faster if it bypassed the filesystem.

TW: Do you think vendors like 3PAR and Pillar will survive the rise of iSCSI block storage?

We're surprised that 3PAR continues to exist. Their kit, although good, is very expensive. How does it succeed against EMC and HDS? People pay for peace of mind I suppose.

TW: Could you describe DataCore's company status and your thoughts on an IPO?

DataCore is a privately-held company. It's 65 percent employee-owned with the rest being investors (but not VC-type investors anxious to crystallise their holdings). We have no venture capitalists of any note. There are 80-100 employees and we are worldwide in operation. We don't give out annual revenue or unit sales figures.

We've been thinking of an IPO for eight years. Investment people come to us and try and interest us in a reverse IPO whereby we'd buy a failed technology company and thus gain publicly-traded status without the expense of an IPO, much like what FalconStor did a few years ago. They went public backwards, so to speak.

Such a reverse-IPO did not seem of interest. DataCore will carry on adding storage management applications to its core block virtualisation product, like the business continuity, recovery and thin provisioning features, and do its bit to thin the middleware layer between storage providing/provisioning resource and storage-using applications.

Read the original, here.

Published Thursday, June 14, 2007 5:39 AM by David Marshall
Filed under:
There are no comments for this post.
To post a comment, you must be a registered user. Registration is free and easy! Sign up now!
<June 2007>