Speaker: Chad Sakac, Vaughn Stewart
This was a really great session! Had two of the best sessions on the last day of VMworld.
- Each protocol has different configuration considerations
- Majority of customers use block protocols (iSCSI, FC, FCoE)
- NetApp: NFS 50% usage, block the other 50% from autosupport data
- Best flexibility come from a combination of VMFS and NFS
- Key Point 1: Leverage Key documents
- VMware technical Resource Center
- FC SAN Config Guide
- iSCSI SAN Config Guide
- Best practices for NFS storage
- Key partner documents – Best practices
- Key Point 2: Setup Multipathing Right
- vSphere Pluggable Storage Architecture (SATP) has several components
- Don’t be inclined to make changes to the defaults..makes it more complicated and adds risk to the design. Don’t change the claim rules or I/O defaults.
- PSP – Path Selection Policy
- Fixed – Used commonly on active-active arrays
- MRU – Default for many active-passive arrays
- Round Robin – Default in vSphere 5.1 for EMC VNX/VMAX.
- MPP
- ALUA – Asymmetric Logical Unit Access. Common on mid-range arrays like NetApp and EMC VNX, and many other brands. Not true active/active for all paths and all LUNs.
- Active – Optimized
- Active – Non-optimized
- Standby
- Dead – APD – Target/array toally dead
- “Gone away” – PDL – Can reach the array, but device such as LUN went away
- Multi-Pathing with NFS
- Significantlly different multi-pathing architecture than block protocols
- NFSv3 is very basic in terms of understanding of multi-pathing
- Must rely on switching technology for link aggregration
- Single TCP connection from the ESXi for data and control information
- Active/Passive path today until a future release of vSphere with NFS4
- Use vendor specific vCenter plug-ins to enhance NFS support/configuration
- Microsoft Cluster Service
- Unsupported Storage Configuration – FCoE, iSCSI, NFS, Round Robin PSP, NPIV
- Vendor support: 3rd party MPPs or Guest connected storage
- Use iSCSI in guest – works very, very well (storage partners support this)
- vSphere 5.1 has expanded support – up to 5 node cluster support
- NFS Best practices
- Use vCenter plug-ins, always! Automates configuration and tweaks
- You can use FQDNs now and it will work
- NetApp Cluster-Mode requires one IP per datastore
- Jumbo Frames?
- Recommendation is to NOT use jumbo frames. Adds more complexity, and performance increase is very marginal.
- Stick with standard block sizes.
- Optimize I/O
- Misalignment of filesystems results in additional work on storage conroller to satisfy IO request
- Affects VMFS and NFS datastores
- Align the guest partitions! Automated in fresh installs of Windows Server 2008/Win7 and later
- Linux will likely not align partitions. Must manually align paritions.
- EMC UBerAlign – Free tool
- NetApp – Data Ontap 8.1.1 – VSC plug-in Migrate and Optimize
- Leverage Plug-Ins (VAAI and VASA)
- 5.1 changes: vCenter Client plug-ins, NFS assists, block assists. See Chad’s Blog post
- Keep It Simple
- Use large capacity datastores
- Avoid extents
- Avoid RDMs
- Array end – Use thin volumes and LUNs
- Storage DRS
- Use it! Even if in manual mode, it will make recommendations
