This was the best session of the day, and highly informative. Eric Siebert is a fast talker, and his slides were packed to the gills with details (and I didn’t even capture 25% of the content). This is the first session that I grabbed a few screenshots from and have incorporated them into my session notes. Please pardon the perspective skew in the photos. If you attended VMworld and have questions about vSphere storage, check out the whole deck. It is a goldmine of good information if you are not well versed in the topic. If you are a storage SME and are a consultant, this is an excellent reference deck to use in customer engagements.
Storage is fundamental to virtualization, and so often it’s not designed properly. And in VDI environments in particular, engineering the right storage solution is no easy task when done on a large scale. Eric covers the top 10 areas in storage that you need to consider to successfully implement a virtualized storage solution. This is just the tip of the iceburg, but a great resource.
If there’s one area of virtualization that can be a resume generating event (RGE), I would say that has to be storage. Unless you like getting pink slips, pay attention to how you design storage!
Why is Storage Critical
Shortage of any one resource prevents higher VM density
Storage is the slowest and most complicated of the four food groups
Available storage resources drive host VM density
#1 File or Block?
File or block and how do I decide?
Storage protocols are a method to get data from host to a storage array
Different protocols are a means to the same end
Each protocol has its own pros and cons
Review the two charts below for a good comparison of protocols
The Speed Race
Bandwidth is not about speed, it is about the size of the highway
Many factors influence performance and IOPS – Bandwidth, cache, disk speed, RAID, bus speeds, etc.
Things to consider:
VMware won’t tell you which protocol to use
vVols will level the playing field between NFS and block
Some applications tend to favor one specific protocol (e.g. VDI for NFS)
VMFS evolves with each release and new features usually first appear in block (e.g. VAAI)
#2 Storage Design Considerations
Must meet the demands: Integration (SRM, VASA, VAAI, etc.); High performance; High Availability; high efficiency
Flash, SSD, and I/O accelerators can enhance performance and eliminate bottlenecks
Avoid using extents to grow VMFS
VAAI (ATS) does not mean unlimited VMs per datastore
Use the PVSCSI controller
Avoid upgrading datastores to VMFS-5. Create new stores and migrate
LUNs don’t have unlimited IOPS
Use jumbo frames for iSCSI/NFS
Always isolate network storage traffic
Always read vendor best practices for your array
Always read vSphere performance best practices papers (vendor recommendations should take priority over generic VMware settings)
#3 Choosing between RDMs and VMFS
RMDs have limited value and more difficult to manage
Use RDMs only for practical reasons (e.g. clustering)
Use VMFS in most cases
VMware testing shows almost no performance benefit for RDMs
This is an old argument and pretty much put to bed…VMFS is the way to go for block
#4 Storage Performance
IOPS and Latency are key performance indicators
Constantly monitor IOPS and latency to spot bottlenecks
Monitor throughput
Baseline your system to know what’s normal and what’s not
Disk latency stats: GAVG, KAVG, DAVG
GAVG less than 20ms
GAVG or KAVG under 1ms
High DAVG indicates problem with storage array
GAVG = DAVG + KAVG
HP announced a vCOPS integration pack for 3PAR which will GA later this year
Use esxtop to monitor storage performance
vSCSIStats is a good CLI utility
Storage Plut-in for vCenter server
#5 Why high available storage is critical
Single point of failure for most environments
When storage fails, all hosts and VMs connected to it fail as well
Protect at multiple levels: adapter, path, controller, disk, power, etc.
vSphere Metro Storage Cluster vs. vCenter SRM
#6 The role of SSDs in a vSphere Environment
SSDs should be a strategic tier, just don’t throw them in
Combine SSD and HDD for balance in performance and cost
SSDs wear out, so monitor closely
#7 Impact of VAAI
Enables integration with array-specific capabilities and intelligence
Fewer commands and I/O sent to the array
vSphere attempts VAAI commands every 16,384 I/Os
#8 Why Thin is In
Almost every VM never uses it’s full disk capacity
Starting thin is easy, staying thin is hard
Saves money
SCSCI UNMAP has been greatly improved (but not automatic) in vSphere 5.5
#9 vSphere Storage Features vs Array Features
Let your array do auto-tiering
Array thin provisioning is more efficient
Lots more on slides that I could write down fast enough..see the slide deck
#10 Benefits for VSAs
Less expensive than a shared storage array
Enables HA, vMotion, etc.
Lower capital cost
Will become more popular in the future, but does have some down sides
The Speed Race
Great post! Need to get hold of this presentation once all the material is released…. =)