Nutanix is all in one solution building block for virtualization. It allows you to virtualize your workloads without requiring a SAN. This approach allows for many benefits, such as buy what you need, when you need it and a reduction in complexity around architecture and operations. I see Nutanix as perfect fit for VDI and Cloud workloads. Where there is uncertainty in the workload and large scale is needed, Nutanix can make a great fit.
Below is how their storage works inside of their 2U building blocks that contain 4 separate nodes. The Name of operating system Nutanix runs on is called HOT(Heat Optimized Tiering). The controller VM is the magic sauce of the operation. All the I\O flows through the controller VM. As data is written to the Fusion-IO card and then is serialized. When the data is cold , it will be laid out to disk in a nice clean format.
The SATA SSD is for ESXi, the Controller VM and VM Swap, nothing else gets to live here.
Write IO - Data is always written locally. Data is replicated on other nodes for high availability. Replica's are spread aacross the cluster for high performance.
Read IO - All data is read locally, Remote access of data only if the data does not exist locally
NFS
Nutanix was built around iSCSI but now have the ability to run NFS as well. Nutanix calls their implementation of NFS, NDFS. It’s a really crazy bit of abstraction. Each VM is basically getting its own iSCSI LUN. Of course you don’t see any of that under the covers stuff. With their release of NDFS they have re-written the standard NFS code to turn allow for better performance. NFS was built to house thousands and thousands of files, today with virtualization it’s almost the opposite. With VMware we have large files, known as VMDK’s that containerize all the smaller files. Nutantix has tuned their implementation of NFS to work with few files large files instead of the old paradigm. NDFS can localize NFS client-server communication giving it the shortest path on the IO highway at all times.
Nutanix supports VAAI (vStorage APIs for Array Integration )and VCAI (View Composer Array Integration) and provide thier own command line interface called nCLI.
Some other articles on Nutanix
How Sales To Government Help Storage Start-Up Nutanix
Nutanix announces more vSphere integration features
Check out all the inter-workings in this Video from Tech Fied day
Nutanix Presents at Tech Field Day 8 from Stephen Foskett on Vimeo.
I am very surprised and confused at the same time by the claims made here.
First of all cache in fusion-io is always going to be lot less than total storage, so how is this solution going to work for true random reads ?
Also your local disk storage can never come close to SAN storage in terms of capacity, so you can not fully replace SAN with local storage.
In case of cloud deployments the use case scenarios are very limited today but this architecture won’t be enough for true enterprise applications if they have to run in cloud.
This is nothing but a solution that will reduce latency in some limited number of scenarios. So why is it worth so much money ?
Hi Chandra
Thanks for visiting my blog, I will try and answer your questions one at time.
Reads – Stargate will move data back into PCI-SSD tier based on access patterns. If that data is only being it once in a blue moon I don’t think this will matter much. Do you have a specific use case? The Controller VM(CVM) that acts as the Storage Processor in traditional terms also has a read cache, we call it a Cache extent internally. If you have a really big read workload you can actually add more memory to the CVM and it adjust the Cache extent automatically. There a way to tell your cache hits/misses on the system.
Storage – Nutanix’s bread and butter is software. I think you see some impressive things on the capacity front from Nutanix. Today we offer both inline and post-process(async) compression to go along with the raw 20TB over 4 nodes. The capacity will be going up. We don’t have worry about RAID either and the long rebuild time associated.
Cloud & Enterprise Ent – We aren’t everything to everyone today. We’ve have had 3 major releases in the course of one year, unheard of with existing vendors. Lots of Enterprise apps don’t belong in the cloud. Regular Server virt gets labeled as private cloud and it isn’t the same. Nutanix is really could lots of VM’s and can levitate pain points due “overcrowding of resources”. Most server virt is great over time but then as Ops keeps adding VM’s due to business pressure, the performance drops off. With Nutanix you get a predictable, repeatable model for every node you add. I think most Enterprise’s would love to scale the IO and number of Storage controllers every time they add more compute.
I really like the views posted on cloud in this article – http://gigaom.com/cloud/the-cloud-backlash-could-be-deep/
Today every node you add is going additional 5,000 IOPS of capacity(20,000 IOPS per block), very few apps need more than that. We aren’t trying to cater to Moster VM’s because for the most part the live in the land of make believe.
Cost – We are 30 % cheaper than our competitors on list price. To my knowledge we have never lost a deal to pricing when comparing the same about of storage and compute power.