Nutanix Tech Nte Virtualizing Micrsft SQL Server n Cnverged Infrastructure The increase in virtualizatin f critical applicatins such as Micrsft SQL Server has given rise t the cncept f hypercnvergence in the data center. This tech nte reviews the perfrmance advantages f and best practices fr cnverged infrastructure fr Micrsft SQL Server. 2014 All Rights Reserved, Nutanix Crpratin
SQL Server n Cnverged Infrastructure If there is ne technlgy trend that is revlutinizing the enterprise datacenter as much as flash, it is hypercnvergence. This is where strage and cmpute (CPU and RAM) are prvided in a single mdular building blck, cnnected tgether using standard Ethernet netwrks. This tech nte will review key aspects f Micrsft SQL Server perfrmance and the benefits f using the cnverged Nutanix Virtual Cmputing Platfrm fr critical SQL Server wrklads. The Nutanix Virtual Cmputing Platfrm is a purpse-built infrastructure slutin fr virtualizatin and clud envirnments. It brings tgether the many benefits and ecnmics f web-scale architectures frm cmpanies such as Ggle, Facebk, and Amazn t the enterprise, thrugh its Nutanix Distributed File System (NDFS). The Nutanix slutin includes highly dense strage and server cmpute (CPU and memry) in a single platfrm building blck. Each building blck is based n industry-standard and high perfrming Intel prcessr/server technlgy, and delivers a unified, scale ut, shared-nthing architecture with n single pint f failure (SPOF). The Nutanix platfrm desn t rely n traditinal SAN r NAS strage, r n expensive strage netwrk intercnnects. What sets Nutanix apart frm ther strage slutins is its uncmprmising simplicity. This simplicity is nt just demnstrated in ease f deplyment (30 minutes) and peratins, such as when dynamically increasing the system size, but als in the ease f designing and architecting successful slutins even fr business critical-applicatins such as Micrsft SQL Server. The Nutanix platfrm desn t suffer frm the prblems r cmplicatins f traditinal strage envirnments, eliminating the difficulties encuntered when designing infrastructure fr SQL in a traditinal architecture. There are n LUNs t manage, n RAID t cnfigure, n FC switches r zning r masking t cnfigure, n registered state change ntificatins (RSCN) t track, and n cmplicated strage multipathing t setup. All strage management is VM-centric, dealing with virtual disks. Strage IO frm a virtual disk is seen as what it is, and ptimized by NDFS. There is ne shared pl f strage acrss a distributed Virtual Cmputing Platfrm that includes flash-based SSDs fr high perfrmance, and lw-latency and high-capacity HDDs fr affrdable capacity. The different types f strage devices in the strage pl are autmatically tiered using intelligent algrithms t make sure the mst frequently used data is available in memry r in flash. Figure 1 shws an verview f the Nutanix Virtual Cmputing Platfrm Architecture, including each hypervisr hst (VMware ESXi r Micrsft Hyper-V), Micrsft SQL Server VMs (User VMs), Strage Cntrller VM (Nutanix Cntrller VM), and its lcal disks. Each Cntrller VM is directly cnnected t the lcal strage cntrller and its assciated disks. By using lcal strage cntrllers n each ESXi hst, access t data thrugh NDFS is lcalized. It desn t always require data t be transferred ver the netwrk, thereby imprving latency. NDFS ensures that writes are replicated, distributing data within the platfrm fr resiliency. The lcal strage cntrller n each hst ensures that strage perfrmance 2014 All Rights Reserved, Nutanix Crpratin
as well as strage capacity increases when additinal ndes are added t the Nutanix Virtual Cmputing Platfrm. Figure 1: The Nutanix Virtual Cmputing Platfrm easily scales withut the cnstraints f traditinal strage. While the strage is lcal t each nde in the distributed scale ut architecture, it appears t the hypervisr as shared strage and therefre integrates perfectly with the virtualizatin layer. In the case f VMware vsphere, this includes VMware DRS, VMware High Availability, and VMware Fault Tlerance. The cmbinatin f SSD and HDD lcal strage in additin t intelligent autmated tiering balances bth cst and perfrmance, while NDFS resiliency techniques eliminate the perfrmance penalties assciated with RAID. The lcalizatin f data allws fr perfrmance and QS t be prvided per hst, s nisy VMs dn t greatly impact the perfrmance f their neighbrs. This allws fr large mixed wrklad vsphere clusters that are mre efficient frm a capacity and perfrmance standpint and mre resilient t failure. At the time this reprt was published, the typical perfrmance frm the midrange mdel with fur ndes ccupying 2RU f space is up t a cmbined 100K 4K Randm Read IOs, 50K 4K Randm Write IOs, and 1.4GB/s sequential write and 3GB/s sequential read thrughput. Users Service Interactin Operatinal Intelligence Data Integratin Desktps Applicatins/Data Servers External Surces MSSQL VMs SCALE SCALE Figure 2: Micrsft SQL Server n Nutanix Virtual Cmputing can run alng side different virtualizatin wrklads. 2014 All Rights Reserved, Nutanix Crpratin
The simplified nature f the Nutanix Virtual Cmputing Platfrm architecture and NDFS simplifies the strage layut. Figure 3 illustrates an example layut, which is standard in a Virtual Cmputing Platfrm envirnment. It is cmprised f a single NFS datastre (r in the case f Hyper-V a single SMB 3.0 share) and single strage pl. There is n need t wrry abut multiple LUNs r assciated queue depths. The next few steps are shwn fr VMware vsphere, but the equivalent can be applied fr Micrsft Hyper-V. Figure 3: SQL Server VM Disk Layut n the Nutanix Virtual Cmputing Platfrm The simplified strage layut and lcalizatin f data t each nde where the VM is executing prvides a number f perfrmance benefits t each type f SQL database. Nutanix Features and Benefits SQL OLTP Transactinal Database SQL OLAP Analytical Database Lcalized I/O fr lw latency peratins Flash fr index and key database files Handles bth randm and sequential wrklads with ease High perfrmance queries and reprts with lcalized I/O Abundant sequential read and write thrughput Scales with ease t accmmdate grwth Table 1: Nutanix perfrmance benefits fr OLTP and OLAP SQL databases. T demnstrate the perfrmance f the Virtual Cmputing Platfrm fr applicatins such as Micrsft SQL Server, a number f SQLIO benchmarks were perfrmed as part f the SQL n Nutanix Best Practices White Paper. The fllwing figures 4 thrugh 8 shw the results frm the benchmark tests: 2014 All Rights Reserved, Nutanix Crpratin
Figure 4: SQL Server SQLIO Single VM Randm IOPS by Blck Size Figure 5: SQL Server SQLIO Single VM Thrughput by Blck Size Figure 6: SQL Server SQLIO Multiple VM IOPS Scalability Figure 7: SQL Server SQLIO Multiple VM Thrughput Scalability 2014 All Rights Reserved, Nutanix Crpratin
The Nutanix Virtual Cmputing Platfrm can c-exist with existing strage investments and fflad wrklads frm existing strage platfrms, freeing up bth capacity and perfrmance until the existing envirnment is due fr refresh. It is easy t migrate int the Virtual Cmputing Platfrm. The perfrmance capability and linear scalability, when cmbined with the uncmprmising simplicity f the Nutanix platfrm, makes it a very gd ptin fr delivering n SQL Server Database as a Service initiatives. Nutanix has created a series f dcuments n virtualizing Micrsft SQL Server. These reprts can be fund at www.nutanix.cm under the resurces sectin: VMware vsphere n Nutanix Best Practices White Paper SQL Server Best Practices White Paper and Reference Architecture Nutanix Bible: Technlgy Guide 2014 All Rights Reserved, Nutanix Crpratin
Appendix: MS SQL Server n Nutanix Best Practice Checklist MS SQL Server n Nutanix best practices can be summarized int the fllwing high-level items. NOTE: The majrity f best practice cnfiguratin and ptimizatin benefits cme at the MS SQL Server level. General Perfrm a current state analysis t identify wrklads and sizing Spend time upfrnt t architect a slutin that meets bth current and future needs Design t deliver cnsistent perfrmance, reliability, and scale Dn t undersize, dn t versize right size Start with a PC, test, ptimize, iterate, and scale Cre Cmpnents MSSQL Perfrmance and Scalability Utilize multiple drives fr TempDB Lg/Data and Database Lg/Data Start with a minimum f 2 drives fr small envirnments r 4 fr larger envirnments Lk fr PAGEIOLATCH_XX cntentin and scale number f drives as necessary Utilize a 64KB NTFS allcatin unit size fr MSSQL drives Enabled lcked pages in memry fr MSSQL Server service accunt (NOTE: if this setting is used the VM s memry must be lcked, nly applies with memry > 8GB) TempDB Data Files Set TempDB size between 1 and 10% f instance database sizes If number f cres < 8 # f cres = # f data files If number f cres > 8 Use 8 data files t being with Lk fr cntentin fr in-memry allcatin (PAGELATCH_XX) and scale 4 files at a time until cntentin is eliminated Database Data files Size apprpriately and enable AUTOGROW respective t database grwth D nt AUTOSHRINK data and lg files 2014 All Rights Reserved, Nutanix Crpratin
At a maximum keep belw 80% f disk capacity utilizatin Use multiple data file and drives Lk fr cntentin fr in-memry allcatin (PAGELATCH_XX), if cntentin increase number f files Lk fr I/O subsystem cntentin (PAGEIOLATCH_XX), if cntentin, spread the data files acrss multiple drives Trace flags Implement trace flag 1118 at startup t remve single page allcatins Implement trace flag 834 t enable large pages (fr tier-1 perfrmance) Utilize the MSSQL Server Best Practices Analyzer (BPA) t identify ptential issues Utilize fast file initializatin Scale number f MSSQL VMs vs. number f MSSQL instances per VM Mre memry = higher perfrmance, if seeing memry pressures, increase VM memry Utilize a dedicated disk fr Micrsft Page File Availability In mst cases vsphere HA will prvide an adequate level f availability and uptime fr nn-missin critical/tier-1 applicatins Fr missin critical/tier-1 applicatins: MSSQL 2012: utilize AlwaysOn availability grups (preferred) MSSQL 2008 and prir: utilize lg shipping r clustered MSSQL using MSCS clusters Take cnsistent database snapshts/backups, frequency shuld be derived frm required RPOs Leverage native r third-party tls t manage backups (example: Micrsft System Center Data Prtectin Manager (DPM), etc.) Manageability Standardize, mnitr and maintain Leverage a MSSQL applicatin mnitring slutin (example: System Center, etc.) Create standardized MSSQL VM Templates Utilize cnsistent disk quantities and layut schemes fr MSSQL VMs Jin the MSSQL Server t the dmain and use Active Directry fr authenticatin Leverage Cntained Database Authenticatin (MSSQL 2012) Use named instances fr MSSQL database instances, even when nly planning a single instance per VM 2014 All Rights Reserved, Nutanix Crpratin
Fr named instances, ensure applicatin cmpatibility with dynamic prts, therwise set instance t use a fixed prt VMware vsphere (leverage equivalent best practices in case f Hyper-V) Fllw VMware perfrmance best practices Avid CPU cre versubscriptin (fr tier-1 wrklads) Fr small MSSQL VMs keep vcpus <= t the number f cres per each physical NUMA nde Fr wide MSSQL VMs size vcpus t align with physical NUMA bundaries and leverage vnuma Keep vcpu numbers easily divisible by NUMA nde sizes fr easy scheduling Leave Hyperthreading sharing at the default plicy (Any) Enable High Perfrmance hst pwer plicy Lck MSSQL VM memry (fr tier-1 wrklads) Size MSSQL VM memry using the fllwing calculatin: VM Memry = SQL Server Max Memry + ThreadStack + OS Memry + VM Overhead Threadstack = SQL Max Wrker Threads * 2MB (fr x64) Use paravirtual SCSI Cntrllers and VMXNET3 NICs Use resurce pls with crrect share allcatin Use DRS anti-affinity rules t keep MSSQL VMs apart Nutanix Virtual Cmputing Platfrm Use a single cntainer fr SQL Server Utilize apprpriate mdel based upn cmpute and strage requirements Ideally keep wrking set in SSD and database size within nde capacity Chse a mdel that can ideally fit the full database n a single nde. NOTE: fr larger databases which cannt fit n a nde, ensure there is ample bandwidth between ndes Utilize higher memry nde mdels fr I/O heavy MSSQL wrklads Create a dedicated cnsistency grup with the MSSQL VMs and applicatins Leverage Applicatin Cnsistent Snapshts n the cnsistency grup t invke VSS when snapshtting Supprting Cmpnents Netwrk Utilize and ptimize QS fr NDFS and database traffic 2014 All Rights Reserved, Nutanix Crpratin
Use lw-latency 10GbE switches Utilize redundant 10GbE uplinks frm each Nutanix nde Ensure adequate thrughput between Nutanix ndes and MSSQL VMs Check fr any pause frames which culd impact replicatin and VM cmmunicatin Active Directry Utilize AD based authenticatin fr MS SQL Server VMs OS and Applicatin Updates Schedule updates t be applied utside business hurs t avid perfrmance impacts Stagger updates in phases 2014 All Rights Reserved, Nutanix Crpratin