Nutanix Launches Fat Cluster Nodes, Hits $200 Million Run Rate
Driven by the need to support "monster VMs" for big database and email servers, hyperconverged cluster maker Nutanix is beefing up its appliance server nodes with heftier machines that pack more compute and storage. The news comes hot on the heels of the general availability of the 4.0 release of its Nutanix Operating System, a new set of software to link Nutanix clusters to public clouds, and the disclosure that its business has more than doubled.
The new NX-8150 node is a dual-socket system using Intel's "Ivy Bridge" Xeon E5 v2 processors. It comes in two flavors, Sachin Chheda, director of solutions marketing, explains to EnterpriseTech. The machine equipped with two ten-core E5-2690 v2 processors has a 3 GHz clock speed on the core and is aimed at database workloads such as Oracle 11g and 12c or Microsoft SQL Server 2014, while the variant using the two-twelve cores E5-2697 v2 chips running at 2.7 GHz will be aimed more at email servers and application servers where having more threads is more important than having a higher clock speed. The nodes currently have 16 GB memory sticks, yielding a maximum main memory of 384 GB, but Nutanix will be adding support for 32 GB sticks soon to double up the memory.
If these database and infrastructure workloads get hefty enough, Nutanix may be pushed by customers to move on up to Xeon E7 processors, with four or even eight sockets per node and many times more memory capacity, but thus far it has been able to make do with two-socket machines in its clusters to meet customer demand.
The NX-8150 comes with four solid-state drives, with capacities ranging from 400 GB to 1.6 TB, and they are Intel S3700 drives. The node can also hold up to twenty 2.5-inch 1 TB SATA drives; Nutanix has chosen Seagate Constellation 2 drives for this machine. It comes with a number of different networking configurations, including two or four 10 Gb/sec Ethernet ports as well as four Gigabit Ethernet ports plus one slower port for linking into the system management network.
In benchmark tests, the NX-8150 is able to host around 3,500 Exchange Server seats, about six times that of its NX-3000 series predecessor. The company did not have benchmark tests available for database workloads at press time.
Nutanix is taking orders now for the NX-8150 and will start shipments in September. With two ten-core E5-2690 v2 processors, 128 GB of memory, and the full complement of flash and disk storage and the Nutanix Operating system thrown on, this node has a list price of $100,000. That is quite a bit of dough to spend on a two-socket server, and it just goes to show you how much value add is in that software as far as Nutanix and its customers are concerned.
The Nutanix Operating System creates a distributed virtual storage area network, known as the Nutanix Distributed File System, across the nodes in the cluster, and it can be run atop VMware's ESXi hypervisor on the server nodes or Microsoft's Hyper-V or Red Hat's KVM. Nutanix got its start on ESXi two years ago when it dropped out of stealth, but Chheda says that adoption for Hyper-V is growing fast as Nutanix scales up its business and that a lot of companies are implementing multiple hypervisors for different lines of business or departments. The preference for any particular hypervisor varies by industry and geography, with banking and finance customers tending to prefer something that leans a little more open source, for instance.
The NDFS software can scale across hundreds of nodes and does not have an upper limit in terms of its scalability, says Chheda, but there are limits to which it has been tested. Some larger customers have single clusters sharing resources that span 32 nodes and beyond, and some have hundreds of nodes. But often, the clusters are partitioned based on the clustering limits of the underlying hypervisor and its management console. (At the moment, VMware's vSAN virtual SAN scales up to 32 nodes, and can probably go a lot further than that.)
The Nutanix operating and file system hybrid uses the Cassandra NoSQL datastore created by Facebook to manage metadata describing the contents of the file system and uses Apache Zookeeper to keep track of the configuration information for each node in the cluster; the consistency of the data is managed by the Paxos algorithm, which was not invented by Google but which was certainly perfected by the search engine giant to create its distributed data stores, which do not just span clusters but multiple physically distributed datacenters. The company uses a MapReduce framework to parallelize data deduplication and compression across the cluster and to spread data around the cluster, including tiering data to the flash or disk. Nutanix uses the Snappy compression algorithm, also used by Google, and compresses data at the sub-block level instead of at the LUN level for more efficiency, and it also allows data chunk sizes up to 128 KB to increase the efficiency of the compression. The compression can software can be run inline or post-process, depending on the performance requirements. (Inline compression eats CPU and adds latency. Post-process compression still eats CPU, but you can run it when the system is mostly idle.)
The deduplication software also runs inline or post process, with inline dedupe running against main memory and flash memory. As each new chunk of data comes into the cluster, it is tagged with a SHA-1 hash, a function that is accelerated by units on the Xeon E5 chip. Disk data is deduped in a post process fashion, and on workloads where this is a lot of duplicate information on virtual machines (such as with virtual desktop infrastructure) the effective capacity of the storage can be increased by as much as a factor of 10 to 1. To save further on capacity requirements, the Nutanix software also does thin provisioning, which means telling a hypervisor and its guest virtual machine partitions that they can have whatever amount of capacity they want to make them happy when they install, but only actually giving them what they need as they are running.
So while NDFS is homegrown, it stands on some pretty strong open source, parallel, and hyperscale shoulders. The Nutanix Prism Central software, which is part of the NOS 4.0 release, allows for multiple Nutanix clusters to be managed from a single pane of glass and integrate with VMware vCenter, Microsoft System Center, or Red Hat Enterprise Virtualization. And with Cloud Connect, which is also part of the 4.0 release, virtual machines running on Nutanix clusters can be backed up and recovered on public clouds such as Amazon Web Services, all managed from within Prism.
Nutanix now has over 800 customers worldwide, and Chheda says that 29 of them have spent more than $1 million on systems. The company has over 10,000 nodes sold thus far in ten quarters of shipping, with a hockey stick slope. In May 2013, Nutanix had an annual run rate of $80 million, and as it exited its fiscal 2014 year the company had an annualized revenue run rate of $200 million. If this growth rate accelerates, it won't be too long before Nutanix breaks through $1 billion in sales.
