Storage Hardware Acceleration

Hardware Acceleration helps ESXi to offload the Storage and Virtual Machine operations to Storage Hardware, thus consuming less CPU, Memory and Storage Bandwidth. It’s supported by Blocked storage devices, SAN, iSCSI and NAS devices

Can view the status in the Hardware Acceleration column of Devices View and datastore view in vSphere Client

vSphere uses ESXi extensions referred to as Storage API’s – Array Integration (formerly known as VAAI). With the release of vSphere 5, these extensions are implemented as T10 SCSI based commands else the ESXi reverts to using the VAAI Plug in’s

With Hardware Acceleration, host can get hardware assistance on:

–          Storage vMotion

–          Cloning/Deploying VM’s from Template

–          VMFS file locking, Atomic Test Set (ATS) (no use of SCSI Reservations)

–          Writes to Thin and Thick provisioned disks

–          Creating/Cloning thick disks on NAS devices

Hardware Acceleration on NAS Devices is implemented through vendor-specific NAS plug-ins and no claim rules are required

–          File clone; this is similar to VMFS block cloning, except NAS clones entire files instead of file segments

–          Create Thick disks

–          Accurately report Space Utilization for Virtual Machines

Hardware Acceleration Considerations

The VMFS data mover does not leverage hardware offloads and instead uses software data movement when one of the following occurs:

–          The source and destination VMFS datastores have different block sizes.

–          The source file type is RDM and the destination file type is non-RDM (regular file).

–          The source VMDK type is eagerzeroedthick and the destination VMDK type is thin.

–          The source or destination VMDK is in sparse or hosted format.

–          The source virtual machine has a snapshot.

–          The logical address and transfer length in the requested operation are not aligned to the minimum alignment required by the storage device. All datastores created with the vSphere Client are aligned automatically.

–          The VMFS has multiple LUNs or extents, and they are on different arrays.

Some quick Command lines:

Display Hardware Acceleration Plug-Ins and Filter

esxcli –server=server_name storage core plugin list –plugin-class=value (Filter or VAAI)

Verify Hardware Acceleration Support Status

esxcli –server=server_name storage core device list -d=device_ID

Verify Hardware Acceleration Support Details

esxcli –server=server_name storage core device vaai status get -d=device_ID

List Hardware Acceleration Claim Rules

esxcli –server=server_name storage core claimrule list –claimrule-class=value (Filter or VAAI)

Add Hardware Acceleration Claim Rules, only for Storage Devices that do not support T10 SCSI Commands

esxcli –server=server_name storage core claimrule add –claimrule-class=Filter –plugin=VAAI_FILTER

esxcli –server=server_name storage core claimrule add –claimrule-class=VAAI

esxcli –server=server_name storage core claimrule load –claimrule-class=Filter

esxcli –server=server_name storage core claimrule load –claimrule-class=VAAI

esxcli –server=server_name storage core claimrule run –claimrule-class=Filter

Verify Hardware Acceleration Status for NAS, will require you to install the NAS Plug-ins using VIB’s

esxcli –server=server_name storage nfs list

**END**

Advertisements

Oracle Design Best Practices on vSphere

Let me start with some general considerations while deploying Oracle on vSphere

  • Do not install any software components that aren’t necessary, such as Office Suites, Graphics, Sound & Video Programs and Instant Messaging services
  • It is also recommended to disable some unnecessary foreground and background host processes
  • For Linux, anacron, apmd, atd, autofs, cups, cupsconfig, gpm, isdn, iptables, kudzu, netfs, and portmap
  • For Windows, alerter, automatic updates, clip book, error reporting, help and support, indexing, messenger, netmeeting, remote desktop, and system restore services
  • For Linux installs, the database administrator (DBA) should request that the system administrator compile a monolithic kernel, which will only load the necessary features

CPU:

  • Oracle databases are not usually heavy CPU consumers and therefore are not characterized as CPU-bound applications
  • As always start with less number of vCPU’s and then increase the count depending on the workload
  • Enable Hyper Threading for Intel Core i7 Processors
  • Use %RUN, %RDY & %CSTP metrics to determine CPU performance

Memory:

  • Set memory reservations equal to the size of the Oracle SGA, atleast for Production systems
  • Acceptable to introduce more aggressive over-commitment in non-production environments such as development, test, and QA etc
  • Set the Virtual Machine to Automatic selection of CPU/MMU Virtualization option
  • Use Large Memory pages, enabled by default on ESX 3.5 and later

Network:

  • Oracle is not heavy when it comes to network utilization
  • Use separate virtual switches, with each switch connected to its own physical network adapter to avoid contention between the ESX service console (applicable for 4.1 and earlier), the VMkernel, and virtual machines (especially virtual machines running heavy networking workloads).
  • To establish a network connection between two virtual machines that reside on the same ESX/ESXi host, connect both virtual machines to the same virtual switch. If the virtual machines are connected to different virtual switches, traffic will go through wire and incur unnecessary CPU and network overhead
  • Use the VMXNET network adapter for optimal performance
  • For IP based storage, enable Jumbo frames end to end

Storage:

  • A dedicated LUN for DB VM’s if the applications have a demanding I/O profile
  • Use VMFS for Single Instance Oracle DB
  • Make sure the VMFS is aligned properly, create the VMFS partitions using vCenter
  • Use Oracle Automatic Storage Management, preferably ASM disk groups with equal disk types and geometries. At a minimum, create two ASM disk groups; one for log files, which are sequential in nature; and one for datafiles, which are random in nature
  • Create a primary controller for use with a disk that will host the system software (boot disk) and a separate PVSCSI controller for the disk that will store the Oracle data files

General guidelines for deploying SQL on VMware vSphere

CPU:

  • Start with a thorough understanding of your workload, you can use VMware Capacity Planner to determine workloads
  • If workloads cannot be determined, start with 1 vCPU, single vCPU VM can support high transaction throughput
  • Account for virtualization overheads (8%-15%, depending on the workload)
  • In Windows Server 2003 guests, when using single-processor virtual machines, configure with a UP HAL (hardware abstraction layer) or kernel. Multi-processor virtual machines must be configured with an SMP HAL/kernel. Windows Server 2008 will automatically select the HAL appropriate for the underlying hardware
  • Avoid CPU overcommitment, otherwise Reserve the full capacity
  • Install/Update latest vmware tools
  • Ensure CPU Compatibility is met for vMotion

Memory:

  • Start with a thorough understanding of your workload, you can use VMware Capacity Planner to determine workloads
  • Increase the database buffer cache to reduce or avoid disk I/O and thus improve SQL Server performance
  • Avoid Memory overcommitment, otherwise reservations can be applied to avoid Ballooning and Swapping
  • If you set the SQL Server lock pages in memory parameter, be sure to set the virtual machine’s reservations to match the amount of memory you set in the virtual machine configuration
  • Use of large pages will help in improving performance

Storage:

There are no concrete recommendations for using VMFS or RDM in SQL Server deployments, both have their advantages and disadvantages. Fibre Channel may provide maximum I/O throughput, but iSCSI and NFS may offer a better price-performance ratio VMware test results show that aligning VMFS partitions to 64KB track boundaries results in reduced latency and increased throughput

It is considered a best practice to:

  • RDM is required when using third-party clustering software, storage based backups to disk or use of third party storage Management software
  • Guest OS can be installed on VMFS and SQL DB and Logs can be on RDM’s
  • Maintain a 1:1 mapping between the number of virtual machines and LUNs to avoid any disk I/O contention
  • It’s recommended to have VMDK’s as eagerzeroedthick
  • Aligning VMFS partitions to 64KB track boundaries results in reduced latency and increased throughput Create VMFS partitions from within vCenter. They are aligned by default
  • Setup a minimum of four paths from an ESX/ESXi host to storage Array, this means that each host requires at least 2 HBA ports

Network:

  • Use NIC teaming and segregate network traffic using VLANs
  • Use the VMXNET3 network adapter for optimal performance. The Enhanced VMXNET3 driver also supports jumbo frames and TSO for better network performance
  • Network communications between co-located virtual machines usually outperforms physical 1Gbps network speed so, if possible, place the various virtual machines that make up an application stack on the same ESXi host

Performance Counters of Interest to SQL Administrators

Subsystem esxtop Counters vCenter Counter
CPU %RDY%USED Ready (milliseconds in a 20,000 ms window) Usage
Memory %ACTVSWW/sSWR/s ActiveSwapin RateSwapout Rate
Storage ACTVDAVG/cmdKAVG/cmd CommandsdeviceWriteLatency & deviceReadLatency kernelWriteLatency & kernelReadLatency
Network MbRX/sMbTX/s packetsRxpacketsTx

*The above content is provided with an Assumption that the VMware environment is vSphere 4.x and later

vSphere Design, Get your basics right the first time!!!!!

How to start designing a vSphere environment, what is involved, whom to involve?

What is a Design?

“A streamlined process which helps the various elements in the organization to determine how to assemble and configure a
Virtual Infrastructure which is strong and Flexible. A design should contain all the important information that meets the
Functional Requirements of an Organization.”

The functional requirments unify 3 different facets of the design:

Technical – What to deploy?
Operational – How to deploy and Support?
Organizational – Who will deploy and support?

Why Vmware or any other Virtualization product for that matter?

There should be a strong reason/objective to deploy Virtualization in the organization, some of them are:

  • Datacenter Consolidation which will help saving Datacenter space, power and Cooling costs etc
  • New Application rollout, Exchange 2010 for ex.
  • Disaster Recovery/Business continuity, Deploy new DR/BC Solution using VMware vSphere and SRM
  • Virtual Desktop Infrastructure

Facets of the design:

Technical:

  • What type of servers, blades or Rack Mounted?
  • Type of Physical CPU’s in the server?
  • Type & Quantity of Storage?
  • Kind of Networking etc?

Organizational:

  • Who will manage the whole environment?
  • Who will provision what? Storage, Network etc?
  • Who will support VM’s, their backup’s etc?
  • Who will be responsible for Security Policies?

Operational:

  • How will the Hosts be managed?
  • How will the VM’s be provisioned?
  • How will I Failover to DR Site?
  • How will compliance be verified?

How to go about desigining any Infrastructure?

  • Review Infrastructure documents provided by the client, although it  may not provide the complete but most of the Functional Requirements
  • Interview the IT teams and IT Management teams (everyone and anyone as required) to understand the environment better
  • Top 5 issues at the user level that can be resolved using VMware Platform

 Assemble bits and pieces?

  • It’s always acceptable to remove functional requirements, if at any point, they don’t serve any purpose what’s the point in having them?
  • Set standards and Best practices, but every Best practice recommended by a vendor may not suite your environment, get a deeper look into the Best Practices and ensure they meet your functional requirements
  • Start documenting every bit of the Design, so that the implementation teams find it easy

All in all, “get your basics right” before you jump into any Technical Jargons 🙂

*Most of the information is being taken from Scott Lowe’s Design book which I thought is the best information available so far 🙂

General Best Practices for deploying Exchange 2010 on VMware vSphere

Best practices do not mean  “A hard and fast rule”, but they are rather some sort of indicators which help you deploy the best possible solution 🙂

CPU:

  • Start with lesser number of vCPU’s and increase on demand
  • 2vCPU’s minimum for mailbox (ideal 6 vcPUs), Unified messaging and Client Access roles and maximum is 12 cores (vSphere 4 has a maximum limit of 8vCPU)
  • Ensure the total number of vCPU’s is equal to or less than the total number of cores on the ESX/ESXi host machine
  • Performance Counters of Interest to Exchange Administrators are CPU %RDY and %USED

Memory:

  • 4GB minimum, 10 GB minimum for multiple roles
  • Size it as per workload, if workloads cannot be determined, use the MS Exchange Design best practices
  • No overcommitment, period
  • Reservertion is preferred, but not recommended, may limit vMotion
  • Do not disable the balloon driver (installed with VMware tools)
  • Performance Counters of Interest to Exchange Administrators are Memory %ACTV, SWW/s & SWR/s

Storage:

  •  Use RDM’s if you need in-guest clustering (no sVmotion)
  •  Install Guest OS on the VMFS datastore
  •  Install Log files and Exchange DB on RDM’s
  •  Maintain a 1:1 mapping between the number of virtual machines and LUNs to avoid any disk I/O contention
  •  Microsoft does not currently support NFS for the Mailbox Server role (clustered or standalone)
  • Performance Counters of Interest to Exchange Administrators are ACTV, DAVG/cmd & KAVG/cmd

Network:

  • Allocate Seperate network adapters/networks for VMotion, VMware FT logging traffic & Management traffic
  • VMXNET3 Network Adapter (available if VMware Tools are installed)
  • Do not use paravirtualized SCSI driver when a VM generates less than 2000 IOPS. For more information see VMware KB 1017652. However this issue has been fixed with 4.1 and later
  • Use VST (for VLAN tagging) which is most commonly used
  • Enable Jumbo Frames
  • Performance Counters of Interest to Exchange Administrators are MbRX/s & MbTX/s

General Recommendations:

  • Use smaller VM’s for eg. a mailbox server with 2 vCPU and 8GB RAM (vMotion can be much faster)
  • Size the VM building blocks considering License costs, more VM’s may mean more Licenses
  • During peak utilization ensure mailbox role is designed not to exceed 70% utilization
  • If deploying multiple roles on the server, the mailbox role should be designed not to exceed 35%
  • Typical guidline is to use n+1 HA/DRS clusters
  • Typical ESX/ESXi host can be of 16 cores (4×4 pCore) , 128GB of RAM, 2 HBA’s and 4 Gigabit Network Adapters
  • Run on a DRS cluster for balanced workloads

HA & DRS Solutions

Local site recovery:

  • Either use Vmware HA for VM level failover or use VMware HA+DAG (Database Availability Groups) for VM and Database level failover
  • vMotion, HA and DRS are not supported for MSCS nodes

Remote Site Availability Options:

  • SRM with DAV
  • Third party Software Replication

Backup & Restore Options:

  • Traditional LAN based backups, Agents installed within the Guest OS
  • Use vDR for Other Exchange Serve roles for ex. Client Access or HUB Transport etc
  • Array based backup Solutions

Also, worth visiting the below links:

Microsoft Exchange Server Profile Analyzer

Exchange 2010 Mailbox Server Role Requirements Calculator to calculate processor, memory, and storage

Mailbox Server Storage Design

VMware White papers on Exchange

*The above content is provided with an Assumption that the VMware environment is vSphere 4.x and later