2
ACCELERATING SQL ANALYTICS Solution Brief Active Data Analytics Maintain I/O Latency Performance at Large Scale Insights and data driven decisions are at the core of business competitiveness and agility. Microsoft understands this and has worked tirelessly to ensure that SQL Server becomes the platform of choice for business analytics and the pivot source for critical decision making. Real-time insights are now available with in-memory technologies for finite transactional or analytics workloads and, for larger workloads, allowed to scale into real-time enabled storage systems. In traditional storage architectures, I/O latencies will cumulatively increase as usage goes higher. This means that at real world workloads, due to high usage of the host CPU cores, latency will be much longer than at lower workloads and become less efficient as their workloads increase, severely impacting performance. Vexata’s VX-100 scalable systems are designed for parallel and distributed flash access to maintain very low latencies all the way up the bandwidth curve. Figure 1: Vexata SQL Server Analytics Read Performance Consistent Performance Delivers Consistent Results I/O latency spikes impact analytics as well as OLTP workloads. With many sequential reads running in parallel (MAXDOP) and the high performance of modern CPU cores, waiting on a single I/O request to return significantly longer than the average can suspend whole processes and drive CXPACKET, one parallel thread waiting for another to finish, to climb. Vexata addresses these requirements with an architecture that fully decouples the data and control planes, distributing the workload to provide cut-through I/O handling with very low latency and predictability. Vexata VX-100 scalable system provides massive scaling, enabling high bandwidth ingest rates that provide data loads in seconds rather than hours or days. The outcome of this architecture is a system that provides consistent latencies across the entire bandwidth curve, as shown below (Vexata systems produce a 99.9992% sub 1ms latency). Figure 2: Vexata SQL Server Analytics Bandwidth, IOPS & Latency Figure 3: Vexata <1ms Latency Histogram

ACCELERATING SQL ANALTICS - Vexata · Active Data Analytics Maintain I/O Latency Performance at Large Scale Insights and data driven decisions are at the core of business competitiveness

  • Upload
    others

  • View
    0

  • Download
    0

Embed Size (px)

Citation preview

Page 1: ACCELERATING SQL ANALTICS - Vexata · Active Data Analytics Maintain I/O Latency Performance at Large Scale Insights and data driven decisions are at the core of business competitiveness

ACCELERATINGSQL ANALYTICS Solution Brief

Active Data Analytics

Maintain I/O Latency Performance at Large Scale

Insights and data driven decisions are at the core of business competitiveness and agility. Microsoft understands this and has worked tirelessly to ensure that SQL Server becomes the platform of choice for business analytics and the pivot source for critical decision making. Real-time insights are now available with in-memory technologies for finite transactional or analytics workloads and, for larger workloads, allowed to scale into real-time enabled storage systems.

In traditional storage architectures, I/O latencies will cumulatively increase as usage goes higher. This means that at real world workloads, due to high usage of the host CPU cores, latency will be much longer than at lower workloads and become less efficient as their workloads increase, severely impacting performance. Vexata’s VX-100 scalable systems are designed for parallel and distributed flash access to maintain very low latencies all the way up the bandwidth curve.

Figure 1: Vexata SQL Server Analytics Read Performance

Consistent Performance Delivers Consistent Results

I/O latency spikes impact analytics as well as OLTP workloads. With many sequential reads running in parallel (MAXDOP) and the high performance of modern CPU cores, waiting on a single I/O request to return significantly longer than the average can suspend whole processes and drive CXPACKET, one parallel thread waiting for another to finish, to climb.

Vexata addresses these requirements with an architecture that fully decouples the data and control planes, distributing the workload to provide cut-through I/O handling with very low latency and predictability. Vexata VX-100 scalable system provides massive scaling, enabling high bandwidth ingest rates that

provide data loads in seconds rather than hours or days. The outcome of this architecture is a system that provides consistent latencies across the entire bandwidth curve, as shown below (Vexata systems produce a 99.9992% sub 1ms latency).

Figure 2: Vexata SQL Server Analytics Bandwidth, IOPS & Latency

Figure 3: Vexata <1ms Latency Histogram

Page 2: ACCELERATING SQL ANALTICS - Vexata · Active Data Analytics Maintain I/O Latency Performance at Large Scale Insights and data driven decisions are at the core of business competitiveness

ACCELERATINGSQL ANALYTICS Solution Brief

Figure 4: Vexata VX-Manager I/O Performance Analytics

Optimize Analytics Infrastructure Performance

CPU cores are at the center of software licensing costs and a significant performance driver, so architectural components should be designed to enable peak productivity of the processing cores. Latency and I/O response times, are a key driver in achieving optimum CPU efficiency, if these cores are idle waiting for data, then cycles can be wasted forcing queries to take longer and reduce the effectiveness of the system. Even with read-ahead processes, modern CPUs can be left waiting when I/O latencies are too high. When CPU utilization is high, I/O wait times consume excessive CPU cycles, producing a false sense of productivity. To fully optimize the host CPU, I/O latencies must be low at all times.

Insight systems need to be optimized for performance workloads in real-time, best practice is to architect these systems to allow for scale as business demands grow and to utilize tools to measure and analyze performance. The Vexata VX-Manager provides complete system monitoring and analytics to identify hot spots and manage a large data warehouse deployment from a single console.

Analytics system scaling requires the need for additional capacity to be added to accommodate larger data sets while simultaneously adding more processing bandwidth to load and analyze the data. A properly architected insight system is built upon a data framework that linearly scales

capacity and bandwidth without impacting I/O wait times. The Vexata Scalable System supports large scale analytics workloads via a system of data blades that are dynamically added on the fly, always maintaining ultra-low latency.

Architecting SQL Insight Systems

Today’s compute infrastructure is built upon high power CPUs that are dlivered with many cores and can quickly consume volumes of data. Intel’s Scalable Xeon systems can consume into the 10’s of Gigabytes per second which could cripple existing all-flash platforms. Vexata systems have been built, tested and qualified to handle larger analytics workloads, providing faster loads, quicker results and more comprehensive insights.

The VX-100 scales to support very large data warehouse and analytics workloads while simultaneously supporting high transaction rate data processing systems. The ability to support both OLTP and OLAP frameworks on the same storage system provide many advantages when consolidating database and analytics infrastructures, without compromising on either I/O processing or bandwidth performance.

Contact Vexata: [email protected]© 2017 Vexata. All Rights Reserved. All third party trademarks are the property of their respective companies or their subsidiaries in the U.S. and /or other countries.

SB-1004-09072017

ABOUT VEXATA: Vexata is the leader in active data management solutions. Vexata’s unique breakthrough enterprise offerings enable transformative performance and scale from database and analytics applications. With unparalleled ability to consume the latest in media like NVMe Flash now with Optane™ SSDs, Vexata systems deploy simply and seamlessly into existing storage environments. Learn more at www.vexata.com

The Vexata VX-100 scalable systems also provides for advanced storage services such as encryption, thin provisioning and space-efficient snapshots and clones, providing a complete suite of storage services to compliment the SQL based data management tools to extract the most value from your SQL analytics systems.