Published on February 20, 2014
Controlling performance in the cloud: Taking charge of your hosting environment
Agenda • Introducing: Databarracks and SolidFire • High performance applications in a hosting environment: Historical challenges • The language of performance – IOPS, Queue Depth, IOPS per Disk, IOPS per Application and calculating IOPS • What key applications can now take advantage of flash performance in the cloud? • Q&A www.databarracks.com | 2
Speaker Bio’s • James Watts, Head of Marketing, Databarracks • Stuart Oliver, Senior Service Provider Program Manager, SolidFire • Currently the Senior Service Provider Program Manager at SolidFire where he is responsible for articulating SolidFire's value proposition to hosting service providers globally. • Spent over 19 years in executive I.T. Management, Product Management and Product Marketing roles. • Born in U.K. then emigrated to Canada and now currently lives in Colorado U.S.A. www.databarracks.com | 3
Prize draw – stick around! Stay until the end of today’s webinar, where we’ll be giving away a 256GB Solid State Drive worth £200. Fill out the brief poll later on to be entered into a draw to win a Sphero Ball worth £100. Winners revealed after Q&A! www.databarracks.com | 4
67% Of all workloads will be run in the cloud by 2016
The Cloud Evolution Cloud today Performance Sensitive Apps $$$ Early cloud Applications Med Test / Development Backup / Archive Startups • • • • • $ Cloud 1.0 High Performance QoS / Hard SLAs Massive scale Reliability Security Cloud Evolution IOPS $$ High Oracle / SAP / Private Cloud Hadoop / NoSQL MS Exchange, VDI ERP,CRM Low Cloud 2.0 www.databarracks.com | 6
Why don’t you run performance apps in the cloud? • Current storage architectures delivers inconsistent and variable performance (‘noisy neighbour’ effect) • Inability to efficiently scale performance • Unable to throttle performance independent of capacity • Low levels of transparency (no visibility into systems) • Dedicated storage array costs are prohibitive • Perception of unreliability www.databarracks.com | 7
Eliminating Noisy Neighbours Noisy Neighbour Tier 0 Tier 1 Decreased Performance Tier 2 Tier 3 The Noisy Neighbour Effect Quality of Service in Practice Individual tenant impacts other applications Create fine-grained tiers of performance Unsuitable for performance sensitive apps Application performance is isolated Performance SLAs enforced www.databarracks.com | 8
Enterprise IT lacks storage agility, and is under significant pressure • Deploy new applications and capabilities faster • Provide more agile and scalable infrastructure • Increase application performance and predictability • Enable automation and end-user self-service • Raise operational efficiency and reduce cost www.databarracks.com | 9
Traditional enterprise storage falls short of these requirements x Deploy new applications and capabilities faster x Provide more agile and scalable infrastructure x Increase application performance and predictability x Enable automation and end-user self-service x Raise operational efficiency and reduce cost www.databarracks.com | 10
Enterprises are looking to service providers that can solve these problems x Deploy new applications and capabilities faster x Provide more agile and scalable infrastructure x Increase application performance and predictability x Enable automation and end-user self-service x Raise operational efficiency and reduce cost www.databarracks.com | 11
Why is understanding application IOPS Important? The way an application operates will determine the workload it generates Caching and disk reads and writes all have an effect on how the application creates load on a system In legacy multi-tenant clouds the workloads can be very random with many different applications generating completely random I/O’s on the system. The random I/O’s cause very unpredictable performance to the tenants using the shared cloud platform. The bigger the cloud implementation the bigger the problem. One of the primary reasons that the cloud has been primarily used for test/dev or less critical applications is the unpredictability of storage resource availability I.T. simple can’t risk running key business applications in these highly unpredictable cloud environments. www.databarracks.com | 12
Understanding the Language of Performance Key Concepts – IOPS (Input/Output Operations Per Second) • I/O (Input / Output) – The process of writing data to the disk or reading data from the disk • The specific number of IOPS possible in any system will vary greatly depending on architecture www.databarracks.com | 13
Understanding the Language of Performance Key Concepts – IOPS per disk type Device Type IOPS 7,200 rpm SATA drives HDD ~75-100 IOPS 10,000 rpm SATA drives HDD ~125-150 IOPS 10,000 rpm SAS drives HDD ~140 IOPS 15,000 rpm SAS drives HDD ~175-210 IOPS Good Quality SSD SSD ~ 5,000 IOPS www.databarracks.com | 14
Understanding the Language of Performance Key Concepts – IOPS per Application Use Case Low Performance Instance Med Performance Instance High Performance Instance (IOPS) (IOPS) (IOPS) Est. Read % MS SQL Server VM 1000 3000 10000 80 MySQL Server VM Oracle 11G SAP Hadoop Mongo DB Cassandra 1000 1000 1000 1000 1000 1000 3000 3000 3000 3000 3000 3000 10000 10000 10000 10000 10000 10000 80 80 80 80 80 80 Web Server VM (windows) 100 300 700 95 Web Server VM (linux) VDI (Windows) VDI (Linux) Dev Server VM Application VM 100 20 20 100 500 300 50 50 300 1000 700 150 150 1000 2000 95 40 40 50 90 www.databarracks.com | 15
Understanding the Language of Performance Key Concepts – Queue Depth • A spinning disk can only serve a single IO at a time, and any seek between IOs adds significant latency. • In cloud environments where multiple applications or virtual machines share disks, unpredictable queue IO to a single head can easily result in orders of magnitude variance in latency, from 5ms with no contention to 50ms or more on a http://www.tomshardware.com/reviews/ssdgaming-performance,2991-3.html busy disk. www.databarracks.com | 16
Understanding the Language of Performance http://www.tomshardware.com/reviews/ssdgaming-performance,2991-3.html Key Concepts – Queue Depth Queue depth refers to the number of outstanding access operations. In the picture above, each solid line represents one disk operation, which can be either a read or write. Because three operations overlap in the same period, there’s a queue depth of three. www.databarracks.com | 17
Understanding the Language of Performance Key Concepts – Queue Depth • Compared to the single-IO bottleneck of disk, SSDs have eight to 16 channels to serve IOs in parallel, and each IO is completed quickly. • So even at a high queue depth, the variance in latency for an individual IO is low. • http://www.tomshardware.com/reviews/ssdgaming-performance,2991-3.html This is the primary reason that you cannot guarantee storage performance on a spinning disk architecture. www.databarracks.com | 18
SSD is ideal for: • Performance oriented relational databases (MSSQL or MySQL) • Large VDI (Virtual Desktop Infrastructure) implementations • Large virtualized corporate (private) cloud environments • Intensive analytics applications • Heavy SAP/Oracle/MongoDB/Hadoop/ERP/CRM implementations • Online Gaming and multimedia configurations www.databarracks.com | 19
5 essential questions when choosing performance storage • Can you guarantee storage quality of service for my critical server? • Can you guarantee IOPS to my critical server? • If needed, can I add IOPS without changing my volume size or migrating my data? • Can you guarantee that I will not have a noisy neighbour problem? • Can you offer me an SLA on storage performance? www.databarracks.com | 20
Summary Recommendations • Save money: You don’t need to overprovision for the high watermark or worst case scenario anymore • Take the time to understand what the IOPS requirements of applications • Don’t buy expensive flash appliances when you can simply provision only what you need from Databarracks www.databarracks.com | 21
Questions? Stay until the end for entry into the prize draw!
Winner Announcements SSD Winner is: Sphero Winner is: www.databarracks.com | 23
... taking charge of your hosting environment ... Sport and Performance Psychology: Controlling Anxiety, ... Controlling Your Cloud ...
Taking charge of your hosting environment . ... Performance is decoupled from ... Sign up for Controlling your performance in the cloud: taking charge of ...
Can't make the webinar? We can send you the full recording after it has ended. If you would like to receive a recording of the webinar, simply fill in your ...
Cloud Servers are Cool ... Differentiating data collection for cloud environment ... Controlling performance in the cloud: taking charge of your hosting ...
Latest Resources Risk Register ... Controlling Performance in the cloud: taking charge of your hosting environment ... Latest News We need your votes!
Webinar: Controlling Performance in the cloud: taking charge of your hosting environment WATCH VIDEO Are we backing up too much? View now «
Q&A with Stuart Oliver from SolidFire, explaining why SSD is heralding the next phase of cloud computing. Channel: Databarracks. Cloud Servers $5
Throttling in a cloud computing environment. ... If your cloud is a PaaS, they will be hosting ... have to scale up your resources and charge the user ...