Assure Performance of Applications Using Scale-out Storage
By Scott Cleland, Sr. Director Corporate Marketing
November 18, 2019
The promise of a scale-out storage system (block or file) is that application performance (IOPs or bandwidth) scales “near” linearly. For instance, IOPs and throughout scale almost linearly as you add nodes in a Dell EMC Isilon cluster while latency remains around 1 msec. Isilon is the Dell EMC scale-out NAS storage platform and an ideal system for bandwidth intensive workloads requiring access to unstructured data. As your dataset grows the ability to scale linearly becomes critical.
On the other hand, scale-out block storage based on Dell EMC VxFlex OS is ideal for virtualized applications such as Exchange, SQL Server, Epic, Oracle and SharePoint. It is well suited for databases that require high write performance, high availability, quick recovery and low cost of storage. It is also useful for applications such as MongoDB or Cassandra that would run on bare-metal servers.
VirtualWisdom Value-add for Dell EMC Scale-out Block Storage (VxFlex OS)
VxFlex OS is the Dell EMC scale-out block storage service that can scale from 3 nodes to 1024 nodes in a single cluster. It enables customers to create a resilient server SAN or a Hyper Converged Infrastructure (HCI) solution on x86 servers. Dell EMC offers VxFlex Ready Nodes, the VxFlex appliance and the VxFlex integrated rack (previously called VxRack FLEX) a fully engineered rack-scale HCI system.
In the VxFlex architecture, the Storage Data Client (SDC) is the component which is installed on servers that consume storage. In contrast, the Storage Data Server (SDS) is installed on nodes which contribute their local storage to the system which creates a storage pool from which volumes are carved out and made available to the SDC. Operation of the cluster is supervised by the Metadata Manager (MDM).
The Dell EMC VxFlex portfolio includes the VxFlex Manager which provides alerting and monitoring on node hardware and enables proactive alerting and automated technical support. VirtualWisdom which currently monitors compute, networking, storage and storage wire data introduced in VirtualWisdom 5.4 the ability to monitor VxFlex OS. Since VirtualWisdom is designed to handle big-data it scales easily to support the largest VxFlex OS deployments of 1024 nodes without resorting to aggregating data to 5, 10 or 15-min intervals as would be the case with other monitoring platforms on the market.
VirtualWisdom 6.1 introduced an Investigation that can track an increase in cache evictions from the VxFlex OS data server which could be symptomatic of a heavy writer. This investigation relies on the Machine Learning (ML) based Trend Matcher analytic. This is an example of the run-book style approach to root-cause analysis that is unique to VirtualWisdom. While most tools would be content with generating an alert, VirtualWisdom actually walks you through the steps to get to root-cause.
Another interesting development is the addition of an investigation for monitoring “Application Workload Drift” while relying on the Trend Matcher analytic. This investigation monitors for changes in a workload profile (IOPs, throughput, I/O sizes) which could cause a change in response time for a business-critical application.
VirtualWisdom Value-add for Dell EMC Scale-out File Storage (Isilon)
Isilon is the Dell EMC scale-out file storage platform for high-volume storage, backup and archiving of unstructured data. An Isilon cluster running OneFS 8.2 can scale up to 252 nodes and 58 PB of storage.
VirtualWisdom 6.1 expands its coverage to include Dell EMC Isilon storage with an application centric monitoring approach that spans from cluster to disk. The VirtualWisdom software integration for Isilon collects 100s of Isilon metrics at 10 sec granularity. This integration provides visibility into the Isilon cluster, node and client for the following protocols: HDFS, NFS3/4, FTP and HTTP.
Purpose-built ML based analytics such as Capacity Forecaster offer capacity forecasting for compute, network and storage (Dell EMC VMAX/PowerMax, Dell EMC Isilon and IBM SVC) out-of-the-box. Capacity forecaster is unique in its ability to capture every interval, providing hourly calculations of the max, 99th percentile, 98th percentile etc. In contrast, other monitoring products on the market average over an hour and in doing so they lose the peaks in the data resulting in an opaque and less useful view of a problem situation.
VirtualWisdom 6.1 also introduces the functionality to generate an alarm when a capacity threshold is crossed while factoring in intangibles such as your internal procurement cycles for new hardware.
Differs from the Approach Taken by Native Monitoring Tools Such as InsightIQ®
VirtualWisdom complements Dell EMC InsightIQ by providing visibility beyond the storage array all the way up to compute and using correlation across silos to produce actionable insights. Dell EMC recommends that you do not use a single instance of InsightIQ to monitor a cluster that contains more than 80 nodes and that you monitor no more than eight clusters or 150 nodes at a time. InsightIQ works with averages and polls for performance metrics every 15 sec over the OneFS® API.
In contrast, one instance of VirtualWisdom with its software integration for Isilon can monitor the world’s largest Isilon clusters with over 252 nodes and can retrieve conversation-level performance metrics every 10 seconds. In addition, VirtualWisdom monitors compute (on-premise and in the public cloud), networks, “big iron” block and file storage, Software Defined Storage (SDS) such as VMware vSAN, Dell EMC VxFlex OS and displays relevant insights in a single proactive role-based dashboard.
Unique Benefits of AIOps-based VirtualWisdom for Monitoring Isilon
The software integration for Isilon storage when coupled with wire-data monitoring enabled by the Virtual Instruments Performance Probes gives you visibility into host application conversations at the highest possible granularity in the industry. This gives you a view both within the Isilon cluster as well as a view from the 10GbE link used to transport application conversations to the cluster.