Forget about storage. Nutanix - a revolution in virtualization

Recently, the idea of ​​SDN (Software-defined Networking) is gaining popularity. Against the background of this concept, and perhaps under its influence, the idea of ​​SDS is developing - Software Defined Storage. This direction looks promising and is beginning to actively develop, for example, entering the beta version of VSAN.



')
I would like to talk about a solution that is probably familiar to many, and partly similar to VSAN, I am talking about Nutanix . One of the interesting differences is that VSAN 1.0 Beta and its development started only last year, and Nutanix version 3.5.2 has existed since 2009. I'll tell you more about this decision.

Nutanix offers a virtual computing platform for VMware vSphere, MS Hyper-V, KVM solutions. It combines server computing resources and corporate storage in a single solution. This approach provides several advantages:

• Expected infrastructure costs. Nutanix expands in blocks and you can always tell how much the system will cost 2, 4, 8 times more productive. You do not have to change storage controllers to more powerful ones or purchase licenses when upgrading to an older storage model.
• Step expansion. Starting with 3 nodes, you can re-buy them as needed. Without the need to immediately lay out productive controllers, if you plan a rapid increase in resource requirements.
• Simplify architecture. This solution does not require SAN and storage , which simplifies administration.
• High performance. The use of RAM, SSD drives and local drives allows you to achieve high speed and minimum delays.

Initially, Nutanix emerged as a solution for VDI (Virtual Desktop Infrastructure) . The potential of the platform itself, the development of the product and the increase in functionality quite allow using it for server virtualization.

The five-minute marketing is over, let's see what it is.
The solution is x86 servers with local hard drives memory processors. They are designed as twin servers 4 / 2U, 2 / 2U or one server for 2 units. Servers connect to the 10Gb / s network (only the younger 1Gb / s model) and use this network for all communications. There is absolutely nothing unusual here, the software makes all the magic of this solution. To be honest, you could use any platform, but it is sold only in a SW + HW bundle.



Nodes are available in the configurations below. The minimum number of nodes to order 3 pcs. Within the same cluster it is possible to use different nodes. Buy more nodes one by one.

		Model	CPU	Storage device	Memory	The number of VM	Network
NX-1000 Series	Per Node (4 per Appliance)	NX-1050	2x Intel Sandy Bridge E5-2620, 12 cores / 2.0 GHz	400 GB SSD, 4x 1 TB HDDs	64 GB or 128 GB	up to 50 VM	2x 10 GbE, 2x 1 GbE, 1x 10/100 BASE-T RJ45
NX-3000 Series	Per Node (4 per Appliance)	NX-3050	2x Intel Ivy Bridge E5-2650v2 16 cores / 2.6 GHz	2x 400 GB SSD, 4x 1 TB HDDs	128 GB or 256 GB	up to 100 VM	2x 10 GbE, 2x 1 GbE, 1x 10/100 BASE-T RJ45
		NX-3051		2x 800 GB SSD, 4x 1 TB HDDs
		NX-3060	2x Intel Ivy Bridge E5-2680v2, 20 cores / 2.8 GHz	2x 400 GB SSD, 4x 1 TB HDDs	128 GB, 256 GB or 512 GB	up to 115 VM
		NX-3061		2x 800 GB SSD, 4x 1 TB HDDs
NX-6000 Series	Per Node (2 per Appliance)	NX-6020	2x Intel Ivy Bridge E5-2630v2, 12 cores / 2.6 GHz	1x 800 GB SSD 5x 4 TB HDDs	32 GB, 64 GB, 128 GB or 256 GB	up to 75 VM	2x 10 GbE, 2x 1 GbE, 1x 10/100 BASE-T RJ45
		NX-6050	2x Intel Sandy Bridge E5-2670, 16 cores / 2.6 GHz	2x 400 GB SSD, 4x 4 TB HDDs	128 GB or 256 GB	up to 100 VM
		NX-6060	2x Intel Ivy Bridge E5-2680v2, 20 cores / 2.8 GHz		128 GB, 256 GB or 512 GB	up to 115 VM
		NX-6070	2x Intel Sandy Bridge E5-2690, 16 cores / 2.9 GHz	2x 800 GB SSD, 4x 4 TB HDDs	128 GB or 256 GB	up to 110 VM
		NX-6080	2x Intel Ivy Bridge E5-2690v2, 20 cores / 3.0 GHz		128 GB, 256 GB or 512 GB	up to 115 VM
NX-7000 Series	Single Node (1 per Appliance)	NX-7110	2x Intel Ivy Bridge E5-2680v2, 20 cores / 2.8 GHz	2x 400 GB SSD, 6x 1 TB HDDs	128 GB or 256 GB	Users - GRID K1 vSGA: 8-64, GRID K2 vSGA: 6-48, GRID K2 vDGA: 6; 3x PCIe expansion slots; Up to 2x GRID K1, 3x GRID K2, 1x APEX	2x 10 GbE, 2x 1 GbE, 1x 10/100 BASE-T RJ45

Especially I want to highlight the model NX-7110 . It allows you to install Nvidia Grid K1 / K2 cards and is used for high-load VDI . The main line is the NX-3000 series . The NX-6000 series features a large amount of SATA drives.

A standard hypervisor is installed on each node, and the Nutanix software is deployed as a virtual machine. Applays, which is installed on each node, uses part of the server's computing resources and "takes" all the hard drives using the PCI-E Pass-through. Accordingly, the hypervisor does not see the drives.



Further, the Nutanix (VM) nodes are combined into a cluster, the nodes are called Controller VM - CVM. This can be done by a single command on any of the nodes or via the WEB-interface of management. In principle, at this moment the storage system is ready for operation, it remains only to present the place to the hypervisors. The presentation is carried out in different ways and only by “its own” hypervisors, where there is VM Nutanix. In the case of VMware, this is NFS; Hyper-V - SMN 3.0; KVM - iSCSI.



What happens between these two events, how does it work?
In general, this video can answer this question:



In addition to the video, I want to talk about the main features of the product (but not all) and about the principle of operation in a little more detail.

1. Data availability
   To manage storage resources, Nutanix Distributed Filesystem (NDFS) is used - Nutanix distributed file system. It monitors data read and write operations, guarantees data integrity and reserves them to protect against a disk or node failure. With read operations, everything is clear, they occur whenever possible locally. The only exception is moving the virtual machine. In this case, the node to which the virtual machine moved will not have its full copy. And for read operations, it will query the network for data from other nodes and restore the full copy from itself in the background. But to ensure fault tolerance, it is necessary to replicate data to other nodes. How it happens: first, you need to save metadata. To do this, use the service role on each of the nodes (modified by Apache Cassandra, to ensure the consistency of data - the Paxos algorithm).
   To protect the data itself, synchronous replication between nodes is used. Nutanix uses such a thing as Replication Factor (replication factor (RF) - by default it is equal to two. Accordingly, each newly recorded block has two copies: one local, one on another node. At the moment, from the command line, you can change the RF for each VM, setting its value to 3, for example, we will have copies on three nodes.
2. Performance.
   High speed (IOPS and low latency) is primarily due to the fact that most of the data is stored locally. Each hypervisor has its own controller in the form of CVM, which minimizes network transfer delays. In addition, the “hot” blocks are cached in RAM and on SSD. Write operations go to the SSD, for this purpose, the caching area is allocated on the disk - Oplog. Further data is distributed between SATA and SSD. Configuration is possible when sequential write operations skip Oplog and go to SATA drives. Tiring moves data between SSD and SATA drives, as is usually the case based on the frequency of requests to data blocks.
   In addition, it is possible to enable data deduplication, which reduces the amount of occupied space on the SSD and in RAM. Deduplication as well as online compression is already included in the price.

Speaking of licensing and cost: two options are licensed separately - replication and post compression. Replication is worth a separate mention. In addition to the data transfer mechanism itself, Nutanix replication includes a mechanism for automating the launch of a VM on a remote site. In terms of its functionality, of course, it doesn’t compare with VMware Site Recovery Manager. But given that it is licensed per node, and not VM, it can be very interesting for individual scenarios. Performance on the example of VDI in numbers - Estimated Sizing Guidelines for VDI Deployments .

Cost: We cannot disclose information on the actual price, but from the experience of calculations I can say so, as a result, the cost of the solution on Nutanix is ​​much cheaper than the solution of the classical architecture of 4 dual-processor servers + storage (level VNX5200 , 3PAR 7200 ) with the same size SSD / Hdd Place an order within a week from the date of this publication and indicate the key word - “Buzzer” and get a discount of up to 20% (the amount of the discount depends on the total cost of your order).

I want to remind you that you can take this equipment with us in Demo for testing. For all questions about Nutanix, please contact our head of the direction - Vladimir Podymov .