Upload
uthram
View
223
Download
6
Embed Size (px)
Citation preview
1 © Copyright 2012 EMC Corporation. All rights reserved.
VNX and Microsoft Best practice
Niko Dukić Technology Consultant | MidTier Specialist [email protected]
2 © Copyright 2012 EMC Corporation. All rights reserved.
Agenda
EMC and Microsoft Alliance
VNX general best practice
VNX and Microsoft integrations
VNX best practice for Microsoft applications
– Exchange Server
– SQL Server
– Sharepoint Server
3 © Copyright 2012 EMC Corporation. All rights reserved.
Top 5 Reasons Why EMC for Microsoft
1) EMC and Microsoft have been engaged as strategic partners for over 16 years
2) EMC has over 300 published technical solutions and another 100 data sheets on Microsoft technologies
3) EMC is Microsoft’s #3 Enterprise Partner (out of 32) for total revenue involving Microsoft products
4) EMC has over 2500 Microsoft Certified Professionals (MCPs)
5) Microsoft is EMC’s 5th largest customer with over 400 CLARiiONs and 40 Symmetrix installed worldwide
4 © Copyright 2012 EMC Corporation. All rights reserved.
VNX General Best practice
5 © Copyright 2012 EMC Corporation. All rights reserved.
VNX General Best Practice
Distribute load over available hardware resources
Design for 70 percent utilization (activity level) for hardware resources
Avoid mixing response time sensitive I/O with large block I/O and high load sequential I/O
Best practices target “80%” of customer solutions
Understanding workloads is key
– Not just IOPS; what kind of IOPS?
6 © Copyright 2012 EMC Corporation. All rights reserved.
VNX General Best Practice: Memory
Install software and enablers before setting read and write cache sizes
Set read cache to roughly 10% and give remaining memory to write cache
Leave cache page size at 8 KiB for majority of configurations. Only adjust if predominant workload I/O size fits 2, 4, or 16 KiB
Start with low watermark of 60% and high watermark of 80%. Lowering watermarks might help reduce frequent forced flushing
7 © Copyright 2012 EMC Corporation. All rights reserved.
VNX General Best Practice: Drive placement
Spread drives across all available buses
Avoid placing drives in the DPE/DAE‐OS enclosure that will participate in a RAID group outside the DPE‐DAE‐OS
The guideline is to prevent a RG rebuild or verify process in the event of a rare power outage. This is one area where guidelines will be broken; not following the guideline does not effect support in any way
Avoid placing more than 8 flash drives per bus, when used for FAST Cache
8 © Copyright 2012 EMC Corporation. All rights reserved.
Resource Limits • Front‐end Host Port Limits
Single Port 760 MiB/s 1000 MiB/s 1200 MiB/s 920 MiB/s
All Ports (Single Card) 1600 MiB/s (VNX5100‐5500) 3000 MiB/s (VNX5700‐7500) 1500 MiB/s 1600 MiB/s (VNX5100‐5500) 2400 MiB/s (VNX5700‐7500) 1300 MiB/s
Module 8 Gb FC 10 Gb iSCSI NEW 10 Gb iSCSI (303‐164‐104D) 10Gb FCoE
9 © Copyright 2012 EMC Corporation. All rights reserved.
Resource Limits • VNX File Limits
Module 1 GbE 10 GbE Optical NEW 10 GbE Optical (303‐195‐101B)
NEW 10 GbE Copper
Single Port 110 MiB/s 1100 MiB/s 1200 MiB/s 1200 MiB/s
All Ports (Single Card) 440 MiB/s 1500 MiB/s 2400 MiB/s 2400 MiB/s
(303‐164‐104D) Data Mover bandwidth limited to dual 8 Gb FC attach ▪ Per Data Mover ~1500 MiB/s ▪ Scale environment by adding active Data Movers, or use VG2/VG8 with 4x FC ports via RPQ
10 © Copyright 2012 EMC Corporation. All rights reserved.
Drive Type • Match the appropriate drive type to the expected workload:
Workload Examples Extreme performance; best performance for transactional random workloads
General performance tier
Well‐behaved streaming, aging data and archive purposes, and backups
Drive Type Flash SAS NL‐SAS
11 © Copyright 2012 EMC Corporation. All rights reserved.
Drive Type and IOPS • Rules of thumb – These are a conservative starting point for sizing, not the absolute maximums – IOPS assumes small block random with good response time – Drives are capable of a sustained IOPS workload based on drive type, as follows:
IOPS 90 150 180 3500
Drive Type NL‐SAS SAS 10K SAS 15K Flash
12 © Copyright 2012 EMC Corporation. All rights reserved.
RAID Level • For best performance from the least number of drives, match the appropriate RAID level with the expected workload:
Workload Examples Heavy transactional with high (greater than 25 percent) random write component, and you plan to stay on HDD Medium‐high performance, general‐purpose workloads, and sequential NL‐SAS, read‐biased workloads, archiving; additional RAID protection
RAID Level RAID 1/0 RAID 5 RAID 6
13 © Copyright 2012 EMC Corporation. All rights reserved.
Traditional RAID Drive Count • Traditional RGs might be selected for specific data placement and/or full stripe writes
• Certain drive counts are more likely to enable this behavior, though these optimizations can also occur with non‐preferred drive counts • With a predominance of large‐block sequential operations the following applies:
RAID Level RAID 5 RAID 6 RAID 1/0
RAID group count 4+1 (256 KiB full‐stripe) 8+1 (512 KiB full‐stripe) 8+2 (512 KiB full‐stripe) 4+4 (256 KiB full‐stripe, no parity calculations) 1+1 1 for VNX File volumes, and stripe 4 volumes
14 © Copyright 2012 EMC Corporation. All rights reserved.
Storage Configuration
• Use the default horizontal positioning method – With current architecture, there are now almost no advantages to using vertical positioning – Storage pools do not use vertical provisioning and should not be forced to do so • Use large element size only when workload is large‐block random read – Only 4+1 RAID 5 supported – 512 KiB element = 2 MiB full‐stripe write • Full‐stripe writes are not expected; informational only
15 © Copyright 2012 EMC Corporation. All rights reserved.
Storage Pool Failure Domains • Minimize failure domains – Don’t t mix primary and replica, table spaces and redo logs DB1 Tables DB1 Logs
DB2 Logs Home Dir/User Backup
DB2 Tables HomeDirs/Users
16 © Copyright 2012 EMC Corporation. All rights reserved.
Storage Pool Tiers • Use RAID 5 with a preferred count of 4+1 for the best performance versus capacity balance – Using 8+1 improves capacity utilization at the expense of slightly lower availability
• Use RAID 6 for NL‐SAS tier – Options for preferred drive count are 6+2 and 14+2 where 14+2 provides the highest capacity utilization option for a pool, at the expense of slightly lower availability • Consider the following rule of thumb for tier construction:
Tier Name Extreme Performance Performance Capacity
Drive Type Flash SAS NL‐SAS
RAID Level 4+1, RAID 5 4+1 or 8+1, RAID 5 6+2 or 14+2, RAID 6
Avoid a drive count that results in a small remainder when divided by the preferred drive count specified
17 © Copyright 2012 EMC Corporation. All rights reserved.
Single Storage Pool • Single Pool? – Modest bandwidth loads can cohabitate if working sets don’t overlap on the same drives – That’s hard to determine; so if in doubt, separate
5/20/75 Mix This design can work with a 3 tier design if the DB has a welldefined working set that
DB1 Tables DB2 Tables Backup
DB: 8 KiB Random Backup: 256 KiB sequential
inhabits primarily top two tiers. Typically used in smaller systems. Note: if backup is hitting high rates during busy DB hours, NLSAS drives on the same bus as SAS can slow down access.
18 © Copyright 2012 EMC Corporation. All rights reserved.
Storage Pool Virtual LUN Type • Storage Pool LUNs can be created as either thick (fully allocated) or thin (virtually provisioned)
LUN Type Usage Considerations
Thin • Maximize ease‐of‐use and capacity utilization
• 8 KiB tracking granularity increases metadata workload
• •
Planning to implement VNX Snapshots and/or block compression Storage efficiency requirements outweigh performance requirements
overhead; not all tracking is 8 KiB; contiguous space allocations are tracked as single entry in table
• Adding a Flash tier to the pool can improve performance as Thin LUN metadata is promoted to the Flash tier
Thick • •
Highest level of pool‐based performance Use only Thick LUNs for VNX File;
• Thick LUN performance is comparable to traditional LUN
instead use a thin‐enabled file system performance • 1 GiB tracking granularity for virtual provisioning
19 © Copyright 2012 EMC Corporation. All rights reserved.
VNX and SQL Best practice
20 © Copyright 2012 EMC Corporation. All rights reserved.
Key Perfmon Counters
Perfmon Counter Description
Disk Reads/sec
Disk Writes/sec
Disk Transfers/sec
Measures: Number of IOPs (Read, Write, Total)
Average Disk Bytes/Read
Average Disk Bytes/Write Measures: Size of I/Os
Average Disk sec/Read
Average Disk sec/Write
Measures: Disk latency
1 - 5 milliseconds (ms) for Log (ideally 1 ms or less on average)
5 - 20 ms for Database Files (OLTP) (Ideally 10 ms or less on average)
Current Disk Queue Length
Measures: # of outstanding I/Os waiting to be serviced
High queue depths + high latencies = performance problem!
High queue depth + low latencies = active and efficient system.
Disk Read Bytes/sec
Disk Write Bytes/sec Measures storage: bandwidth
21 © Copyright 2012 EMC Corporation. All rights reserved.
A Day in the life of SQL Server… SQL Server Storage I/O
Plan for user load peaks, not systematic peaks…
22 © Copyright 2012 EMC Corporation. All rights reserved.
VNX and SQL Best Practice
Understand the applications using SQL Server – SQL Server is just the database!
Plan for performance and capacity – Choose the appropriate disk type and RAID protection
– Size first for performance, than capacity
Isolate files for maximum performance – Database Files: RAID 5
– Log Files: RAID 1/0
– Data Warehouse: RAID 5 or 6
– Group workloads of similar I/O together
Use SQLIOsim.exe to validate all changes to storage – No need for SQL server to be installed
23 © Copyright 2012 EMC Corporation. All rights reserved.
VNX and SQL Best Practice
Consider log performance – Sequential workload and can cause contention if put on the same drives as user db
– Creating another log file will not help as it writes in single file only
– Consider recovery scenarios – plan for recovery, not backup
Consider tempdb performance – Global resource to all tables in instance and can be very io intensive
– Can cause disk contention if put on the same drives as user db
– Start with reasonable size, 1 to 10 % of total size
– Set auto increment to a minimum of 10 % to reduce fragmentation
Divide filegroups among different spindles – Filegroups can be accessed in parallel
24 © Copyright 2012 EMC Corporation. All rights reserved.
VNX and SQL Best Practice
Balance database and log LUNs between SPs
Enable FAST cache only on highly active databases with OLTP workloads
– Disable FAST cache for logs
– Disable FAST cache for sequential workloads if possible
Consider FAST VP on large databases – Control when FAST VP will move data not to interfere with high usage time
Align disk windows partitions
25 © Copyright 2012 EMC Corporation. All rights reserved.
Flash in Cache and SAS in Pool
• VNX 5700 – (20) SAS drives in
pool – (4) Flash drives in
FAST Cache
• SAS Only – 1448 TPS, 0.8s Ravg
• SAS + FAST Cache – 5778 TPS, 1.95s Ravg
Recommendation – Pools YES – FAST Cache YES – FAST VP YES
Virtualizing SQL Server 2008 Using EMC VNX Series and VMware vSphere 4.1 http://powerlink.emc.com/km/live1/en_US/Offering_Basics/White_Paper/h8188-virtualizing-sql-vnx-
vsphere-psg.pdf
~ 4x performance improvement in TPS with only x2 increase in response time
Microsoft SQL Server
>2x performance improvement in TPS with no increase in response time
26 © Copyright 2012 EMC Corporation. All rights reserved.
VNX and Exchange Best practice
27 © Copyright 2012 EMC Corporation. All rights reserved.
VNX and Exchange Best Practice
New storage schema to reduce disk requests
Io size is larger
Messages per day Exch2007 IOPS Exch2010 IOPS (standalone database)
Exch2010 IOPS for protected database
25 0.11 0.040 0.030
50 0.18 0.060 0.050
100 0.32 0.120 0.100
Version IO Size
Exchange 2003 4 kB
Exchange 2007 8 kB
Exchange 2010 32 kB (BDM is 256 kB)
28 © Copyright 2012 EMC Corporation. All rights reserved.
VNX and Exchange Best Practice
Background Database Maintenance (BDM) is online defragmentation and online database scanning
BDM produces a large sequential 256 KB read IO
BDM is enabled by default and runs 24x7 on both active and passive database copies
29 © Copyright 2012 EMC Corporation. All rights reserved.
Disk type selection
The types of disks that are appropriate for your Exchange Server 2010 deployment depend on a number of factors, including your users’ mailbox size limit and IOPS requirements.
SAS disks provide high capacity with moderate IO speed, which makes them highly suitable for Exchange Server 2010 environments with high IOPS per user requirements.
NL SAS disks are a good fit for the less demanding IO requirements of Exchange Server 2010. NL SAS disks are typically the best choice for larger mailbox sizes and average to heavy IO profiles.
Flash drives are not the best choice for Exchange data, but they can be appropriate when using FAST Cache
30 © Copyright 2012 EMC Corporation. All rights reserved.
RAID Type selection
Any RAID type can be used, provided there are enough disks to handle the IO and storage capacity requirements.
RAID 1/0 is the best choice for Exchange Server 2010, especially if NL/SAS drives are used.
RAID 5 is most appropriate in environments with low IO requirements and where large mailboxes are being deployed.
RAID 6 can be used but has much more impact on write requests
31 © Copyright 2012 EMC Corporation. All rights reserved.
VNX and Exchange Best Practice
isolate the Microsoft Exchange Server database and log workload from other IO-intensive applications and workloads
always calculate disk IO requirements before calculating capacity requirements.
Select appropriate disk types that meet your IO and capacity requirements.
do not place both database and log files for the same database on the same physical disks
Deploy each DAG copy on its own set of physical disks
Spread the load as evenly as possible across storage array resources
32 © Copyright 2012 EMC Corporation. All rights reserved.
VNX and Exchange Best Practice
Format Windows NTFS volumes used for Exchange databases and logs with an allocation unit size of 64 KB
The use of FAST Cache on VNX systems neither benefits nor detracts from Exchange performance because of the Exchange IO pattern
If FAST Cache is used segregate database volumes from log volumes
Enable FAST Cache only for database volumes
It is not recommended to use FAST Cache for logs
The use of FAST VP with Exchange Server 2010 on VNX is not recommended (BDM activity affects storage tiering priorities)
33 © Copyright 2012 EMC Corporation. All rights reserved.
• FAST Cache does not particularly benefit Exchange 2010
– Exchange has little to no data locality
• FAST VP is not presently a good choice for Exchange 2010
– BDM skews the data movement, which is on a 24 hour cycle
– Exchange allocates data space in a way that works against the FAST VP algorithms
• Use the SOAPTool utility to evenly distribute data on Pool LUNs • Improves Jetstress performance
Recommendation
– Pools YES, with SOAP Tool Utility
– FAST Cache Neutral
– FAST VP NO, BDM causes non-optimized relocations
VNX and Exchange Best Practice
34 © Copyright 2012 EMC Corporation. All rights reserved.
VNX and Exchange Best Practice
Either RAID groups or storage pools can be used for Exchange Server 2010 on VNX.
– With RAID 5 do not stripe metaLUNs across multiple RAID groups (reduces Exchange performance)
– Use homogeneous Pools or RAID Groups
When using storage pools, the following guidelines apply: – Design and grow storage pools by using the appropriate multiplier for best
performance (R1/0 4+4, R5 4+1, R6 6+2).
– Use Thick LUNs only for mission-critical jobs; avoid Thin LUNs for > 100 users
Set the storage array page size parameter to 16 KB (only if Exchange is the only or mosst important application).
35 © Copyright 2012 EMC Corporation. All rights reserved.
VNX and Exchange Best Practice
Key requirements for Exchange Server 2010 include the following: – Total number of users (mailboxes) in the Exchange environment
– Number of users per mailbox server
– User IO profile (number of messages sent/received per day)
– Mailbox size limit
– Read/write ratio
– Average message size
– Outlook mode
– Log protection buffer
– Deleted items retention policy
– User concurrency requirements
– If replication is needed, the number of DAGs and database copies required
– Backup and restore requirements (RTO/RPO)
– Third-party software that affects space or IO (for example, Blackberry)
Technet: Exchange 2010 Mailbox Server Role Requirements Calculator
36 © Copyright 2012 EMC Corporation. All rights reserved.
VNX and Sharepoint Best practice
37 © Copyright 2012 EMC Corporation. All rights reserved.
VNX and Sharepoint
• Cost effective Tiered Storage
– Block, file, object
– Externalize BLOBs
• 95% shrinkage of SQL database size
• Integrated data protection provides peace of mind without the complexity
Exchange SQL Server SharePoint
Virtual Server Virtual Server Virtual Server
38 © Copyright 2012 EMC Corporation. All rights reserved.
SharePoint Performance Considerations
Performance for SharePoint highly use case dependent
SharePoint workload for content management – Can be characterized by 60/20/10/10 for Browse/Modify/Search/Upload
– File size, distribution, and concurrent access impact performance
Symptoms vs. Remedy – Symptoms often occur during individual tasks (e.g. file upload by user)
– Sizing however needs to factor in system level workloads for the entire farm
The storage subsystem is rarely the bottleneck – Generally web front end (WFE) CPU represents the bottleneck
– VNX FAST Suite enables very efficient configurations (majority NL-SAS, 2-5% SSDs)
– Search functions are the workloads that happen to benefit the most
39 © Copyright 2012 EMC Corporation. All rights reserved.
SharePoint Performance Considerations
SQL Database not efficient at handling files IO – Each write represents an update of the table and the log
– Non-externalized storage formats use the SQL buffer pool when the data pages are accessed
– Large files and high concurrency can lead to buffer overruns compounding the issue
– Other processes such as index rebuilds can also compete for the same server resources
– RBS: Remote BLOB storage supported by VNX and Metalogix
40 © Copyright 2012 EMC Corporation. All rights reserved.
SharePoint Component
FAST VP FAST Cache
Workload
Search Index - ++
• R:W Ratio of 90:10; 32K block read, 64K write; Small working set.
• Search Index re-uses the same blocks on disk as working space to process list items during indexing. Storage IO quiet between crawls.
• Highly changing, throw-away data.
Search Query + +
• R:W Ratio of 70:30; 32K read, 64K write; Small working set.
• Query improves due to random large block reads/writes being serviced by FAST Cache, but Query storage is large and costly in FAST Cache.
• Highly-read data with small burst write changes.
tempdb + +
• R:W Ratio of 50:50; 8K read, 32K write. • TempDB (database page reuse) is ideally suited to FAST
Cache algorithms. • Same blocks are re-used on disk and performance of
TEMPDB directly affects SharePoint performance request.
Content DB + +
• R:W Ratio of 95:5; 16K (read and write); Working set reduced via RBS.
• Low IOPs requirements does not require FAST Cache. • DB Index operations see an improvement.
BLOB Store - - • Low IOPs requirements, large-block I/O.
VNX and Sharepoint Best Practice
41 © Copyright 2012 EMC Corporation. All rights reserved.
FAST Cache for SharePoint 2010
Enabling Fast Cache on system and search databases (Crawl, Property and tempdb) improves Search and Index performance
– Crawl operations improved by 90%
– User search response time improved by 27%
– Maximum user capacity increased by 10%
Powerful