Tag Archives: next

What’s .NEXT 2016 – Enhanced & Adaptive Compression

Posted on by
Reply

There are so many “under the cover” capabilities of the Acropolis Distributed Storage Fabric (ADSF) which have been designed and built not for short term marketing “checkboxes” but with a long term vision in mind.

As a result, Nutanix has been able to continually innovate and stay ahead of the HCI market while building a next generation platform (including the Acropolis Hypervisor, AHV) for the enterprise cloud.

Nutanix is also 100% software defined which makes adding new features and enhancing existing features possible even for hardware which is several years old.

As a result of the forward looking development of ADSF, it has allowed Nutanix to lead in the SDS space with features like Compression, Deduplication and Erasure Coding (EC-X).

In-line Compression is recommended for most workloads including business critical applications such as Oracle, SQL and Exchange and typically provides not only excellent capacity savings but an increased effective SSD capacity which results in higher performance. Compressing data on the capacity tier (not just flash tier) also helps improve performance and lowers the cost per GB of storage.

As of the next release, the compression functionality has been enhanced to support compressed and uncompressed slices in the same extent groups which for those of you not familiar with ADSF, an “Extent Group” is a group of “Extents” in which data is stored.

In previous generations of ADSF, regardless of if ADSF got good compression or not – all the data for a virtual disk (vdisk) residing in a container with compression enabled will have all of its data compressed. This can causes unnecessary overheads especially in cases where compression savings are minimal, such as for already compressed data such as Video or image files (e.g.: JPG).

This is one reason why it’s important that data reduction features such as compression (and Dedupe/Erasure Coding) can be turned off for workloads where benefits are minimal.

Previously in ADSF, compressed and uncompressed data was not supported within the same extent group which resulted in the cluster (Curator) having the added overhead of moving extents from one extent group to another even for data with low/no compression benefits.

This unnecessary overhead has now been removed which means less background tasks (overheads) resulting in lower CPU utilization by the Nutanix Controller VM (CVM) and better overall compression performance.

Secondly, Nutanix will be moving to the LZ4 group of algorithms which has two variants, LZ4 and LZ4H. LZ4H is really exciting because it gets nearly as much compression as Zlib while having a similar CPU cost but can decompress at the speed of LZ4. LZ4 by itself is marginally superior to Snappy in the common case, but the LZ4H makes this a very attractive choice.

This allows ADSF to do tiered compression – so cold data compressed with LZ4 can be further compressed with LZ4H giving higher compression ratios.

Also some good news for existing customers, this enhanced compression will be included in the next major AOS update which can be deployed via One-Click upgrade without any downtime or the requirement to reformat the drives, that’s true software defined storage.

Stay tuned for an upcoming blog showing the before and after compression savings on the same dataset.

Summary:

The upcoming releases of Acropolis OS (AOS) will provide:

  1. Higher compression savings
  2. Lower CVM overheads
  3. Dramatically reduced background file system maintenance tasks
  4. Enhanced compression will be included in the next major AOS one click upgrade!

Related .NEXT 2016 Posts

What’s .NEXT 2016 – Acropolis File Services (AFS)

Posted on by
Reply

At .NEXT 2015 Nutanix announced the Scale out File Server Tech Preview which was supported for AHV environments only. With the imminent release of AOS 4.7 the Scale out File Server has been renamed to Acropolis File Services (AFS) and will now be GA for AHV and ESXi.

AFS provides what I personally refer to as an “invisible” file server experience because it can be setup with just a few clicks in PRISM without the need to deploy operating systems.

AFS provides a highly available and distributed single namespace across 3 or more front end VMs which are automatically deployed and maintained by ADSF. The below shows a mixed cluster of 10 nodes made up of 8 x NX3060 and 2 x NX6035C nodes with the AFS UVMs spread across the cluster.

AFSoverview

Data is then stored on the underlying Acropolis Distributed Storage Fabric (ADSF) in a Container which can be configured with your desired level of resiliency e.g.: RF2 or RF3 as well as data reduction features such as Compression, Deduplication and Erasure coding.

AFS inherits all of the resiliency that ADSF natively provides and supports operational tasks such as one-click rolling upgrades of AOS and hypervisor without impacting the availability of the file services.

Functionality

Backups

Nutanix will provide AFS with native support for local recovery points on the primary storage (cluster) and allow both Async-DR (60 mins) and Sync-DR (0 RPO) to allow data to be backed up to remote cluster.

For customers who employ 3rd party backup tools, AFS can also be simply backed up as an SMB share which is a common capability amongst backup vendors such as Commvault and Netbackup.

The below shows a high level of what a 3rd party backup solution looks like with AFS.

AFSbackup2

Quotas

AFS also allows administrators to set quotas to help with capacity management especially in environments with multi-tenant or departmental deployments to avoid users monopolising capacity in the environment.

Patching/Upgrades

Acropolis File Server can be upgraded and patched separately to AOS and the underlying hypervisor. This ensures that the version of AFS is not dependant on the AOS or hypervisor versions which also makes QA easier and minimizes the chance of bugs since the AFS layer is abstracted from the AOS and hypervisor.

This is similar to how the AOS version is not dependant on a hypervisor version, ensuring maximum flexibility and stability for customers. This means as new features/improvements are added, AFS can be upgraded via PRISM without worrying about interoperability and dependancies.

Patches and upgrades are one-click, rolling, non-disruptive upgrades the same as AOS.

Scaling

As the file serving workload increases, Acropolis File Server can be scaled out by simply adding instances to balance the workload across. If the Nutanix cluster has more nodes than AFS instances, this can be done quickly and easily through prism.

If the cluster has for example 4 nodes and 4 AFS instances are already deployed, then to scale the performance of the AFS environment the UVMs vCPU/vRAM can be scaled up OR additional nodes can be added to the cluster and AFS instances scaled out.

When one or more additional AFS instances (UVM) are added, the workload is automatically balanced across all UVMs in the environment. ADSF will also automatically balance the new and existing file server data across the ADSF cluster to ensure even capacity utilization across nodes as well as consistent performance and linear scaling.

So in short, AFS provides both scale up and scale out options.

Interoperability with Storage Only nodes

Acropolis File Server is fully supported on environments using storage only nodes. As the storage nodes provide a Nutanix CVM and underlying storage to ADSF, the available capacity and performance is made available to AFS just like it is to any other VM. The only requirement is 3 or more Compute+Storage nodes in a cluster to support the minimum 3 AFS UVMs.

AFS deployment examples

Acropolis File Services can be deployed on existing Nutanix clusters which allows file data to be co-located on the same storage pool with existing data from virtual machines as well as with physical or virtual servers utilising Acropolis Block Services (ABS).

AFS_ExistingCluster

Acropolis File Services can be deployed on dedicated clusters such as storage heavy and storage only nodes for environments which do not have virtual machines, or for very large environments while be centrally managed along with other Nutanix clusters via PRISM Central.

AFS_DedicatedCluster

Multi-tenancy

AFS also allows multiple seperate instances to be deployed in the same Nutanix cluster to service different security zones, tenants or use cases. The following shows an example of a 4 node Nutanix cluster with two instances of AFS. The first has 4 AFS instances (UVMs) and the second has just 3 instances. Each instance can have different data reduction (Compression, Dedupe,EC-X) settings and be scaled independently.
AFSMultipleFileServers

Summary:

  • AFS supports multiple hypervisors and is deployed in mins from PRISM
  • Can be scaled both up and out to support more users, capacity and/or performance
  • Interoperable with all OEMs and node types including storage only
  • Supports non-disruptive one-click rolling upgrades
  • Supports multiple AFS instances on the one cluster for multi-tenancy and security zone support
  • Has native local recovery point support as well as remote backup (Sync and Async) support
  • All data is protected by the underlying ADSF
  • Supports all ADSF data reduction technologies including Compression, Dedupe and Erasure Coding.
  • Eliminates the requirement for a silo for File sharing
  • Capacity available to AFS is automatically expanded as nodes are added to the cluster.

Related .NEXT 2016 Posts

What’s .NEXT 2016 – Acropolis Block Services (ABS)

Posted on by
Reply

Acropolis Block Services or ABS (not to be confused with Anti-lock Braking Systems), is an extension of the In-Guest iSCSI Nutanix announced at .NEXT 2015.

The original goal of the In-Guest iSCSI was to enable support for applications like MS Exchange which are not supported on NFS and applications such as SQL clustering for quorum drives, and this has been very successful. However customers have been telling us for a number of years they want to make Nutanix the standard platform for their datacenters, however they have not been able to realise this vision due to a number of reasons including:

  • The desire/requirement to re-use existing servers
  • Applications which are not virtual (for many reasons, mostly political)
  • Performance / Scalability of externally connected servers
  • Complexity including operational considerations of external iSCSI

Let’s discuss each of these topics and how ABS solves these challenges.

Re-using existing servers

As it’s uncommon for customers to be at the exact right time in the refresh cycle for servers and storage to replace all infrastructure at once, ABS allows customers to either get started with Nutanix by deploying some nodes/blocks, or to scale the existing environment/s while being able to use the Acropolis Distributed Storage Fabric (ADSF) to provide storage to existing HCI workloads and non HCI workloads.

A couple of key advantages of ABS compared to the existing In-Guest iSCSI support and traditional SAN/NAS is:

  • ABS load balances and optimizes paths so MPIO and ALUA are not needed
  • New storage is automatically added without requiring client-side changes

The downside to using ABS as a stop gap until the refresh cycle for the compute hardware is that is does add complexity which I discuss in this article from July 2015.

Scaling Hyper-converged solutions – Compute only

However, if the goal is to maximise the return on investment (ROI) of existing infrastructure, ABS is in my opinion a better option than having another silo of storage to install/configure and manage as it:

  • ABS load balances and optimizes paths so MPIO and ALUA are not needed
  • New storage is automatically added without requiring client-side changes
  • Removes the requirement for another silo.
  • Increases performance/capacity/resiliency of an existing cluster
  • Allows customers to standardize their infrastructure
  • Gives customers flexibility to quickly add/remove nodes from a cluster/s to meet requirements.

Scalability:

ABS ensures linear and automated scalability by creating virtual targets to ensure performance is not limited by iSCSI limitation of one session per initiator and target. This means a single LUN (or Volume Group in Nutanix speak) can be serviced by the multiple virtual targets which are spread across all Nutanix CVMs. This ensures multiple network threads are used which also mitigates against network threads being a bottleneck.

By default 32 virtual targets are used to ensure optimal performance for even the largest and most I/O intensive workloads.

This process is also transparent to the administrator and application to avoid any complexity in implementation and ongoing support.

The following diagram shows how the data services IP sits in front of the virtual targets (which are on each CVM) and the vDisks are spread across all controllers for maximum performance.

ABSvirtualtargets

At .NEXT 2015 Nutanix announced support to scale storage seperate to compute using “Storage Only” nodes and this capability is fully compatible with ABS. This ensures capacity and performance can be scaled separately to compute for maximum flexibility.

ABSnoiSCSIMPIO

Resiliency:

If a vDisks active CVM goes offline due to failure or planned maintenance, any active sessions against that CVM are disconnected, which triggers a re-logon from the iSCSI client. The re-logon occurs through the external data services IP, which redirects the session to a healthy CVM.

This means things like One-Click rolling AOS upgrades can still be performed as they are with native Nutanix environments.

ABSCVMfailure

Functionality:

ABS supports SCSI-3 persistent reservations for shared storage-based Windows clusters, which are commonly used with Microsoft SQL Server and clustered file servers.

As of Acropolis OS (AOS) 4.7, ABS will be supported with physical servers or virtual machines. Support for connecting ESXi via iSCSI is expected to follow in a future release.

ABS supports several use cases, including:

  • iSCSI for Microsoft Exchange Server.
  • Shared storage for Linux-based clusters
  • Windows Server Failover Clustering (WSFC).
  • SCSI-3 persistent reservations for shared storage-based Windows clusters
  • Shared storage for Oracle RAC environments.
  • Bare-metal environments.

ABSoverview

ABS enables server hardware separate from the Nutanix environment to consume the Acropolis DSF resources, so you can leverage existing server hardware investments against Nutanix storage resources. Workloads not targeted for virtualization can also use the DSF.

Supported Client OS & Qualified Applications

  • RHEL 6+
  • Windows 2008 R2 & Windows 2012 R2
  • Oracle RAC
  • Microsoft SQL Server
  • Microsoft Exchange Server

Summary:

Whether you have applications that require shared storage access or environments with separate storage and compute needs, Acropolis Block Services (ABS) simplifies deployment and highlights the dynamic scale out, extreme performance, and high availability of the Nutanix platform. ABS automatically load balances iSCSI clients to take advantage of all resources in the cluster, and failure events are managed seamlessly. The same upgrade, snapshot, and asynchronous replication workflows that customers leverage today work consistently whether you are using VMs or VGs. By enabling VM, file, and block services, Nutanix offers a single platform to consolidate workloads and ease administration, thus reducing risk and enabling organizations to simplify their infrastructure.

Related .NEXT 2016 Posts