in-booth presentation at 2:00 pm speed-up up... · all-flash storage learn why k2 is gartner's...
TRANSCRIPT
Speed-up Virtualized Databases with All-Flash StorageLearn why K2 is Gartner's highest ranked storage for OLTP workloads
In-Booth Presentation at 2:00 PM
Derek SwansonPrincipal Architect
2 | © 2017 Kaminario
275+Employees
Boston HQLocations in
Israel, London, Paris, Beijing &
Seoul
200+ Channel Partners
K2 All-Flash6th Generation of
ProductFully Integrated
HW and SW Platform
23 Patent Families
Software-defined arch, I/O
Optimization, Data Reduction
Kaminario and K2 All Flash Array At A Glance
Kaminario K2 All-Flash: A History of Continuous Innovation
Kaminario Founded
2008
Kaminario ForeSightTM &
ClarityTM
20162010
K2Gen1
2014
K2Gen5
2013
K2Gen4
2017
K2Gen6
2017 Magic QuadrantLeader
K2.N & Flex Product
Announcements
2018
K2.N NVMe
• Superior investment protection and NVMeleadership
• Strongest positioned leader with under $2B market cap
• Significantly differentiated with only scale-up and out SSA
Inaugural Gartner Magic Quadrant for
Solid State Arrays
2014 MQ
Kaminario a “Leader” in 2017 MQ for Solid State
Arrays
2017 MQ
From a ”Visionary” to an Industry “Leader” Kaminario 36 month MQ evolution2017
Gartner Ranks Kaminario K2 #1 in All-Flash Performance, Analytics and OLTP .. And #2 in every other use case and
category among 19 Solid State Arrays
ITS A SWEEP!Kaminario Ranks #1 in Critical Capabilities for SSAs
Use Case Ranking
High Performance Computing #1
OLTP #1
Analytics #1
VDI #2
Server Virtualization #2
Break the Storage I/O bottleneck with Kaminario K2 All-Flash
2X - 5X Better Oracle and SQL Server performance on K2 All-Flash
* Based on actual implementation data
• Manual configuration of RAID groups, Storage Pools, Snapshot Reserves, Hot Spares, etc.
• Considerations of Unbalanced capacities, Controller multi-pathing and failover load
Disk Partition Configuration Complexity
• SQL Server defaults can slow down - a gate built for legacy storage
ESX Defaults Throttle Performance
• Burst of random writes mainly during checkpoint creates bottleneck
• Storage architected for average, not peak
Burst/Peak loads affecting performance
• Sequential I/O gets blended with IO from other VMs, causing massive bottlenecks
• Inefficient scheduling of read vs write
VMware I/O Blender Effect & Mixed Block Workload
82%Storage I/O bottleneck
The symptoms of storage I/O bottleneck on virtualized databases
8 | ©2017 Kaminario
Disk Partition Configuration Complexity
Manually Configured –§ RAID Groups§ Storage Pools§ Snapshot Reserves§ Hot Spares§ Block Size Configurations
Wait States & Contention Created by Manual Placement of -§ OS Files§ SQL Data, Log Files§ Backups
How Kaminario K2 Solves This?
The K2 uses a single automatically tuned pool of storage for the entire storage cluster!
§ No RAID Configuration§ No Storage Pools to set up§ No Hot Spares or Hot Spots§ No Cache Allocations§ No Manual Block Size Configuration§ NO MANUAL CONFIGURATION AT ALL!
9 | ©2017 Kaminario
Handling Peak Load Latency
Architecting for Peak load too costly!
§ Cheaper to insert a small “caching” layer of DRAM or Flash.
§ Must rely on having a small data skew to store in cache.
§ Perpetually gambling that “pre-fetch” algorithms can keep “cache-miss” ratios low enough that bad performance doesn’t cause application failure.
§ Always breaks when hit with really large load spikes.
How Kaminario K2 Solves This?
The K2 is NOT a caching engine, indeed it has no read cache in it at all.
§ The K2 architecture uses no cache.
§ No reliance on pre-fetch algorithms
§ Natively handles 100% data skew rates!
§ Symmetric active-active cluster
§ Fastest array under most unpredictable workloads with multiple applications.
10 | ©2017 Kaminario
ESX Defaults Throttle Performance
ESX and SQL default configurations are designed for slow storage.
§ SQL Server can issue a lot of outstanding IO operations
§ Queues of a particular guest can become full and IO is bottlenecked
§ ESX queue settings are small by design
§ This is a necessary gate for legacy storage, to prevent IO from overwhelming the storage array and breaking applications
How Kaminario K2 Solves This?
Kaminario Best Practices§ K2 PowerCLI scripts discover and apply ESX cluster
setting optimizations to massively increasing performance from default settings.
§ Settings for Windows, Linux and SQL enable database parallelization and greater IO throughput to maximize performance
§ ESX optimization also built into plugin
11 | ©2017 Kaminario
VMware I/O Blender Effect
VMWare’s inherent scrambling up of IO wreaks havoc on legacy arrays
§ An application’s elegant coalesced sequential IOs get blended in with IO from other VMs
§ Cache based systems are overwhelmed by random IO, as IO schedulers cannot efficiently schedule read vs write ops
§ LUN alignment issues magnify this problem 2 to 3 times
§ The performance penalty can be as much as 50%!
How Kaminario K2 Solves This?
“Log Structured Array” architecture is a 100% fix to the IO Blender.
§ Writes put into a Log in DRAM and mirrored, then write is fast-acked back to the host
§ Logs are held in DRAM until full of byte-aligned compressed data (0 padding or fragmentation), then written down in full stripes to all new locations
§ Small block or large block, random or sequential operations handled the same by the K2
§ No hot spots or drive contention exists, and sub millisecond performance is consistent across every type of workload profile
12 | ©2017 Kaminario
Mixed Block Workload Creates Bottleneck
SQL workload is heavily mixed. Fixed block arrays latency fluctuates dramatically depending on:
§ block size§ read vs write§ random vs sequential operations
How Kaminario K2 Solves This?
Patented “Adaptive Variable Block” Algorithm Optimizes Performance
§ The K2 handles blocks in their native size without the need to carve up IO into a pre-determined fixed small block size.
§ Blocks are written contiguously with a single metadata entry, keeping MD overhead low.
§ Blocks are read and rehydrated on byte aligned boundaries with no wasted cycles.
§ Mixed workloads of all types are handled extremely efficiently
Adaptive block size provides optimal performance for any application
A read is processed as a SINGLE IO operation
16K128K
Patented algorithm adapts to specific block size of different application workloads to optimize performance without compromising on IOPS, latency, or bandwidth
Traditional Flash Array
Latency
Array Optimized for Mixed Workloads
Latency
Adaptive block size – optimized for all workloads
Selective deduplication helps speed up virtualized database performance
• Deduplication on databases is limited or non-existent
• Flexibility of turning off deduplication for Databases – no wasted CPU cycles
• 15-20% - Performance benefit on K2 for turning off deduplication
Real-time and Selective
Selective Deduplication
Dedupe
Non- Dedupe
✔X
Non-DB workload
Databases
Offloading compression to hardware frees up CPU to enhance virtualized database performance
• Compression uses CPU resources
• Stronger compression algorithms require more CPU cycles
• K2 makes the perfect tradeoff to maximize compression effectiveness at HW level while optimizing performance by freeing CPU
Summary: Built from the ground-up to provide extreme performance for Virtualized Database environments
Truly Active-Active
controllers
• Active-Active controllers ensure all CPUs serve database R/W needs• Start small, grow by adding additional active-active controllers (nodes)
• Natively adapt to incoming DB block size for concurrent OLTP/OLAP• No tuning required by the database or storage admin for different workloads
Adaptive block size
• Automatic, global distribution of data and metadata across all nodes• Future proof your investment with mixing and matching old and new gen K2s
Unique shared
metadata
• Consistent low latency and high throughput/IOPS for mixed workloads• Keeps OLTP and OLAP workloads highly responsive even under peak demands
Mixed workload
processing
§ Check out our Live Demos§ Meet-The-VMware-Experts§ See how much you can
save with a Custom TCOScan Your Badge to Win a Trip to OrlandoTweet your session comments @KaminarioFlash for additional entries to win!
Want More Information? Visit Our Booth
Thank You
Amazon Echo Dot Raffle