acrolinx

Author: b-ahibbard

articles/azure-netapp-files/performance-virtual-machine-sku.md

@@ -1,30 +1,30 @@
 ---
-title: Azure virtual machine SKUs best practices for Azure NetApp Files | Microsoft Docs
-description: Describes Azure NetApp Files best practices about Azure virtual machine stocking keeping units (SKUs), including differences within and between SKUs.
+title: Azure virtual machine stock-keeping units (SKUs) best practices for Azure NetApp Files | Microsoft Docs
+description: Describes Azure NetApp Files best practices about Azure virtual machine stocking-keeping units (SKUs), including differences within and between SKUs.
 services: azure-netapp-files
 author: b-hchen
 ms.service: azure-netapp-files
 ms.topic: conceptual
 ms.date: 07/02/2021
 ms.author: anfdocs
 ---
-# Azure virtual machine SKUs best practices for Azure NetApp Files
+# Azure virtual machine stock-keeping unit best practices for Azure NetApp Files
 
-This article describes Azure NetApp Files best practices about Azure virtual machine stock keeping units (SKUs), including differences within and between SKUs.  
+This article describes Azure NetApp Files best practices about Azure virtual machine stock-keeping units (SKUs), including differences within and between SKUs.  
 
 ## SKU selection considerations
 
 Storage performance involves more than the speed of the storage itself. The processor speed and architecture have a lot to do with the overall experience from any particular compute node. As part of the selection process for a given SKU, you should consider the following factors:
 
 * AMD or Intel: For example, SAS uses a math kernel library designed specifically for Intel processors. In this case, Intel SKUs are preferred over AMD SKU.
-* The F2, E_v3, and D_v3 machine types are each based on more than one chipset. In using Azure Dedicated Hosts, you might select specific models (Broadwell, Cascade Lake, or Skylake when selecting the E type for example). Otherwise, the chipset selection is non-deterministic. If you are deploying an HPC cluster and a consistent experience across the inventory is important, then you can consider single Azure Dedicated Hosts or go with single chipset SKUs such as the E_v4 or D_v4.
+* The F2, E_v3, and D_v3 machine types are each based on more than one chipset. In using Azure Dedicated Hosts, you might select specific models (Broadwell, Cascade Lake, or Skylake when selecting the E type for example). Otherwise, the chipset selection is nondeterministic. If you're deploying an HPC cluster and a consistent experience across the inventory is important, then you can consider single Azure Dedicated Hosts or go with single chipset SKUs such as the E_v4 or D_v4.
 * Performance variability with network-attached storage (NAS) has been observed in testing with both the Intel Broadwell based SKUs and the AMD EPYC™ 7551  based SKUs. Two issues have been observed:
     * When the accelerated network interface is inappropriately mapped to a sub optimal NUMA Node, read performance decreases significantly.  Although mapping the accelerated networking interface to a specific NUMA node is beneficial on newer SKUs, it must be considered a requirement on SKUs with these chipsets (Lv2|E_v3|D_v3).
-    * Virtual machines running on the Lv2, or either E_v3  or D_v3 running on a Broadwell chipset are more susceptible to resource contention than when running on other SKUs. When testing using multiple virtual machines running within a single Azure Dedicated Host, running network-based storage workload from one virtual machine has been seen to decrease the performance of network-based storage workloads running from a second virtual machine. The decrease is more pronounced when any of the virtual machines on the node have not had their accelerated network interface/NUMA node optimally mapped. Keep in mind that the E_v3 and D_V3 may between them land on Haswell, Broadwell, Cascade Lake, or Skylake. 
+    * Virtual machines running on the Lv2, or either E_v3  or D_v3 running on a Broadwell chipset are more susceptible to resource contention than when running on other SKUs. When testing using multiple virtual machines running within a single Azure Dedicated Host, running network-based storage workload from one virtual machine has been seen to decrease the performance of network-based storage workloads running from a second virtual machine. The decrease is more pronounced when any of the virtual machines on the node haven't had their accelerated network interface/NUMA node optimally mapped. Keep in mind that the E_v3 and D_V3 may between them land on Haswell, Broadwell, Cascade Lake, or Skylake. 
 
 For the most consistent performance when selecting virtual machines, select from SKUs with a single type of chipset – newer SKUs are preferred over the older models where available. Keep in mind that, aside from using a dedicated host, predicting correctly which type of hardware the E_v3 or D_v3 virtual machines land on is unlikely. When using the E_v3 or D_v3 SKU:
 
-* When a virtual machine is turned off, de-allocated, and then turned on again, the virtual machine is likely to change hosts and as such hardware models.
+* When a virtual machine is turned off, deallocated, and then turned on again, the virtual machine is likely to change hosts and as such hardware models.
 * When applications are deployed across multiple virtual machines, expect the virtual machines to run on heterogenous hardware.
 
 ## Differences within and between SKUs