- A new VM-centric NFS storage with a purpose-built FS to accommodate VMs.
- General availability of their product called Tintri VMstore T440.
- A marketing machine running at full steam.
Now, as a long time storage and virtualization fan, I always wondered why no one has ever come up with a real VM-centric storage, I mean, NetApp, EMC and others made a big deal out of their VMware integration points, but no other product in the market can be considered really VM-aware like Tintri is claiming its products to be.
What’s in the box?
VMstore T440 is currently the sole product coming from Tintri, it’s a single 4U box that contains 9 SSD (MLC, 160GB each), 15 HDD (SATA, 1TB each) and can be attached to your network using 1 or 10Gb ethernet.
Each box presents a single 8.5 Tb datastore via NFS to the ESX hosts and their management interface completely integrates with vCenter completing their seamless approach.
They address scalability using an approach similar to ESX, each storage block has fixed capacity and fixed throughput/performances, you scale out with a new box when you reach the maximum capacity of one, this approach still has limitations especially from the HA side, which is currently not provided at the storage level (however it’s planned in the next release), but in a sense it lines up with the ESX host paradigm.
Their VM-aware approach
Tintri proud itself for being the first VM-aware storage, they built VMstore from ground up using VMs and VMDKs as basic concepts, they don’t have legacy concepts like raid groups and LUNs, they simply consider VMDKs and VMs as atomic components on which they built all their storage stack. They’re able to make VM-level snapshots without snapshotting the entire volume, from a QoS perspective they’re able to pin the single VM (or even the single VMDK) to SSD storage or just leave the system to operate storage tiering at its discretion.
In the white paper that is available on the Tintri website, they claim to have resolved the MLC SSD write amplification problem with their filesystem, you probably remember my article on how SSDs work and one of the main point was that SSD controllers must relate to the non-SSD optimized filesystem write patterns which is one of the main concerns for write amplification, this is the first time that a company claims to have built a FS with write amplification in mind (at least to my knowledge) and I think that we’ll have others jump on this bandwagon (NetApp WAFL is the first that comes to mind).
During SNW 2011 I spent some time chatting with Chris Bennett, VP of Marketing for Tintri and I made this short interview with him:
I’m sure that Tintri will be one of the hottest startups in 2011 and will be probably a hot acquisition target for many (I sense EMC coming out first on this), we’ll see if they will be able to deliver and start growing or if someone bigger will buy them out of the market.