VM-aware
The Tintri VMstore file system is designed from the ground-up for VMs. It uses virtual machine abstractions — VMs and virtual disks — in place of conventional storage abstractions such as volumes, LUNs, or files. Each I/O request – reads, writes, or metadata operations – map directly to the particular virtual disk on which it occurs. Tintri VMstore directly monitors and controls I/O performance for each virtual disk. Conventional storage must make file- or volume-level decisions about data location and access prioritization, and then map these to the VM through bolt-on features — with their own complex set of storage abstractions — to give them context in a virtualized environment. By operating at the virtual machine and disk level, Tintri VMstore finally provides administrators with the same level of insight, control and automation as CPU, memory and networking resources as general-purpose shared-storage solutions.
Tintri VMstore communicates with the VMware vCenter™ Server API, to learn which virtual machines are active and reside on Tintri storage. The VMstore collects and reports per-VM and per-virtual disk statistics, such as size, I/O throughput, and resource utilization. An administrator can immediately see which virtual machines and virtual disks are responsible for the consumption of storage resources, and what performance the virtual machines receive, without additional layers of complexity. This approach eliminates “false sharing” between virtual machines or virtual disks which reside on the same storage abstraction.
Performance
Tintri VMstore was designed from the ground-up to support the requirements and access patterns of virtual machines. Conventional NFS servers handle a wide range of applications, file sizes, metadata operations, and stream-based reads and writes — and can’t isolate individual VMs. In contrast, Tintri VMstore creates a tight coupling between I/O activity and VM abstractions. Virtualization loads are mostly block-aligned read-and-write access to large files. The Tintri file system efficiently supports multi-gigabyte to multi-terabyte virtual disk files, to efficiently handle many concurrent reads and writes.
Tintri VMstore also automates flash placement to isolate VM performance. Disk access latency and performance is magnified in a virtual environment and can become highly unpredictable. Flash performance, in contrast, is much easier to characterize, is location-independent, and minimizes interaction between request streams from different VMs. Virtual machines can be deployed on the Tintri VMstore without manual placement or guesswork. The Tintri file system automatically uses the combination of flash and disk that best suits the activity of the virtual machine, adjusting it over time to prevent variations in one VM’s I/O patterns from causing disruptions for other VMs.
The Tintri file system minimizes swaps to disk by using a very small block size with automated placement to ensure only active data is kept in flash. The typical mixture of hot-and-cold VM data in virtualized environments often include minimally-active or periodically busy VMs. The Tintri file system adapts to periodic workloads to ensure they are not evicted from flash during idle periods, and provides good read/write performance even for VMs that reside mainly on disk. Tintri eliminates traditional tiering by automating placement and writing directly to flash when appropriate.
High Availability
Dual-controller VM-aware storage. Tintri VMstore T540′s dual-controller redundant architecture delivers the high availability and reliability that enterprise IT environments demand. It eliminates single points of failure in the VMstore appliance architecture, so VMs can operate from the VMstore without disruption. The underlying active-passive dual-controller architecture of the T540 provides two independent and fully redundant sets of computing resources for the Tintri file system. Any hardware failure, either external or internal, that may make one set of resources unavailable causes the other to take over.
MLC FLASH
Tintri VMstore is designed to fully capitalize on the most cost-effective flash technology available. Multi-level cell (MLC) flash – combined with inline deduplication, compression, and a unique flash/disk file system – enables Tintri to cost-effectively run the majority of enterprise applications with flash-level performance.
Tintri’s innovative file system solves MLC flash problems that previously made it unsuitable for enterprise environments:
- MLC flash suffers from high levels of write amplification for random write workloads. Unchecked, this reduces random write throughput by more than 100 times, and dramatically reduces flash lifetime. The Tintri file system eliminates write amplification and delivers excellent system performance and longevity.
- MLC flash produces periodic latency spikes due to internal metadata processes. This delays access for many hundreds of times longer than average. Tintri’s proprietary file system refinements eliminate these spikes.
The Tintri file system integrates flash as a first-class storage medium rather than as a bolt-on cache to fully leverage continued improvements in flash price and performance.
Flash as an intelligent, highly granular resource — combined with inline dedupe and compression technology in a hybrid flash/disk file system architecture — radically alters the economics for VMs.
Enterprise Operations
Tintri’s unique performance capacity gauge simplifies decision-making and planning. Tintri VMstore lets administrators simply and accurately monitor performance utilization. For each VM, the Tintri file system monitors flash, disk and CPU usage over several days, and continuously calculates aggregate system impact. It evaluates factors like available flash capacity, disk I/O and the access patterns of virtual disks over time to ensure each VM gets the resources it needs to run near peak performance. Tintri’s unique approach allows VMstore to adapt to a range of different usage scenarios.
With a file system specifically designed for reliability and fast recovery, Tintri VMstore employs advanced transactional schemes to support non-disruptive software patches. Most patches apply in less than 30 seconds, with no disruption to VM clients or end-users.
Every VMstore appliance features sophisticated SSL-encrypted auto-support and one-button diagnostic data collection. Results are quickly and reliably sent via the cloud, enabling quick diagnosis and problem resolution without relying on e-mail and manual log file collection.