250 likes | 495 Views
Oracle Use and Best Practices for High Performance Cloud Storage. Paul Johnson Jim Stewart Nagendran J. Program Agenda. Cloud Storage Overview & Challenges Traditional Performance Analysis ZFS Storage Appliance Analytics Case Studies. What Is Cloud Storage?. Shared
E N D
Oracle Use and Best Practices for High Performance Cloud Storage Paul Johnson Jim Stewart Nagendran J
Program Agenda • Cloud Storage Overview & Challenges • Traditional Performance Analysis • ZFS Storage Appliance Analytics • Case Studies
What Is Cloud Storage? • Shared • Many users or applications. • Varying or unpredictable workload • Plurality of uses. • Cost effective
Oracle IT Cloud Storage Architecture Goals • Standardize • Pool storage to reduce the number of devices to manage. • Use Ethernet networks to deliver storage wherever possible. • Simplify • Use NAS to allow simple rapid reallocation of unused space. • Manage backups at the storage level rather than at the client level.
Challenges Storing The Cloud • The data has be to actually be stored somewhere. • Performance requirements vary within a given cloud. • IO patterns vary with a given cloud. • Dynamic changes to the cloud change the storage workload. • Storage as a cloud service means that storage administrators often lack access to client systems to troubleshoot issues.
Traditional Performance Problems • For most vendors, network storage is a black box • Why is our application performance so slow? • Which clients are accessing my data? • Which files are being written the most? • Can I observe my performance at a certain date & time?
Traditional Performance Solutions $ $ $ $ $ • Blame the storage! • Buy more storage! • Faster / bigger systems! • Faster / bigger disks! • Won’t necessarily fix the underlying problem • How certain are we that it’s the storage? • Is our issue load-related or architecture-related? $ $
Modern Approach • Plenty of modern, sophisticated instrumentation • vmstat • iostat • kstat • prstat • top • Unfortunately, most are difficult for non-technical users to decipher
Modern Solution • Visual • There needs to be an easy to navigate GUI • Data needs to be graphed in per-second intervals • Real-time • Data should be presented in real-time • Data should be archived for historical analysis
Statistics • Statistics are a collection of data displayed and graphed by ZFS Analytics • Every layer of the stack has available statistics • CPU • Disk • Network • Protocols
Drilling Down into Statistics • Statistics can be drilled down to form ad-hoc queries • Which clients have the highest access rate on a certain day of the week? • Which NFS operations are taking at least 2 milliseconds during high production load? • Which disks are seeing the most amount of activity over the past month?
Navigate Through Time • ‘Arrow’ buttons move back and forth between the past and present • ‘Pause’ suspends the graph, but not data collection • ‘Zoom’ buttons control the time scale of the graph • ‘Clocks’ are pre-defined zooming to by-minute and by-hour • ‘Calendars’ can display an entire day, entire week, or entire month
Worksheets • A worksheet is a user-defined collection of statistics • They are persistently saved and can be shared by users • Handy for crafting just the family of statistics that are relevant for your performance analysis
Phantom Workloads Exposed • A user running a performance test increases the load on the storage in an unexpected way and the storage administrator can drill into the issue in real time to understand where the load is coming from. • File Level Visibility
Phantom Workloads Exposed Where is my IO load coming from?
Phantom Workloads Exposed What is that client doing?
Phantom Workloads Exposed What’s happening inside that file?
Oracle IT Replaced 15k Disks with 7.2k Disks • Using analytics Oracle IT showed that they could replace their 15k rpm Netapp with 7.2k rpm ZFS Appliance by using analytics to show how increased cache hits on the ZFS Appliance took the load peaks that previously required faster disks. • Analytics Drives Savings • About 4x floor space savings • Reduction in power consumption • Overall Lower TCO