VSAN gets mature and recently proved to be the leader of the HCI market.
Here is the list of new features that comes with VSAN 6.2:
- Deduplication and Compression
- Erasure coding
- Quality of Service (QoS) using IOPS limits
- New performance service and improved health service
Disk space efficiency
Deduplication and Compression features are presented in VSAN 6.2 to reduce amount of consumed storage by as much as 7x. This is best-case scenario and average ratio, I guess, will be around 2.5-3X.
Both technologies are available in All-Flash VSAN configuration only and are available as a single feature. You can't enable Deduplication without enable Compression and vice versa.
By default, Deduplication and Compression are disabled. To enable it ESXi servers will need to reformat all disk groups which can be pretty lengthy process, depending on the VSAN size. But don't worry - this procces doesn't require any downtime of your VMs. In worst case scenario some redundancy reduction will be incurred, e.g. like in the screenshot below
First, the ESXi server will dedupe data during de-staging phase when data is copied from the cache tier to capacity tier. The VSAN uses 4K blocks of data when looking for identical blocks. The data is deduped only within a single disk group.
Once data is destined and deduplicated ESXi host will try to compress 4K blocks of data, but it will only do so if the compressed block of data is less than 2K, otherwise the data will be left uncompressed
Note - The same rule is applied to compression of memory pages during the RAM contention. I am wondering if VMware simply re-used the compression algorithm. :)
RAID-5/6 which goes under name Erasure Encoding
That's the feature a lot of VSAN fans have been waiting for.
Prior to VSAN 6.2 the only option to provide data redundancy was mirroring which is quite expensive. RAID-5/6 , on the contrary is significatnly cheaper, but comes at performance cost, particularly, due to write overheads when parity data has to be recalculated every time any block of the stripe is altered. Since Erasure Coding is available in All-Flash configuration only the performance impact can be safely neglected for most of the environments.
FT=1 setting will use RAID5 and FT=2 will implement RAID6.
Once data is destined and deduplicated ESXi host will try to compress 4K blocks of data, but it will only do so if the compressed block of data is less than 2K, otherwise the data will be left uncompressed
Note - The same rule is applied to compression of memory pages during the RAM contention. I am wondering if VMware simply re-used the compression algorithm. :)
RAID-5/6 which goes under name Erasure Encoding
That's the feature a lot of VSAN fans have been waiting for.
Prior to VSAN 6.2 the only option to provide data redundancy was mirroring which is quite expensive. RAID-5/6 , on the contrary is significatnly cheaper, but comes at performance cost, particularly, due to write overheads when parity data has to be recalculated every time any block of the stripe is altered. Since Erasure Coding is available in All-Flash configuration only the performance impact can be safely neglected for most of the environments.
FT=1 setting will use RAID5 and FT=2 will implement RAID6.
Another new feature introduced in VSAN 6.2 is Swap Efficiency. With this setting enabled all Virtual Machnie swap files will be created as sparse object. That is, the disk space won't be actually used for swap files until memory contention happens and ESXi host starts to swap out RAM pages into swap files. Along with Deduplication and Compression this swap efficiency technology provides very significant disk space savings.
Quality of Service (QoS) using IOPS limits
Now VSAN gets closer to traditional VMFS datastores in terms of storage QoS. The first releases of VSAN didn't allow IOPS limits/shares for VMs residing on VSAN datastores since the SIOC wasn't compatible with VSAN. With VSAN 6.2 administrators will have an option to limit IOPS per VM to be able to control "noisy neighbours".
QoS for VSAN 6.2 doesn't make any distinction between Reads and Writes.
It ought to be noticed that all IOs are normalised to 32KB block size. That means that if you, for instance, set a IOPS limit to 100 and block size of the IOs will be up to 32KB that VM will be able to get 100 IOPS. However, with block size of 64KB the VM will get only 50 IOPS.
New performance service and improved health service
VSAN 6.2 delivers new graphs and data to provide performance information at the cluster, host, virtual machine and virtual disk levels.
Interestingly, that performance data isn't stored in vCenter database. It is stored in as a separate VSAN object and has no dependecy on the vCenter.
Here are some examples of new types of graphs.
For instance, on the cluster level you can compare Virtual Machine IOPS and Throughput stats with Virtual SAN Backend stats. The latter shows the 'raw' amount of IOPS produced to deliver Virtual Machine IOPS, e.g. with Failures to tolerate set to 1 for each VM Write IO the VSAN backend has to produce 2 Write IOs to mirror data to two different hosts.
On the host level you can see Disk Group and Disk detailed stats in addition to VM IOPS and Throughput graphs mentioned above. So you can see how well-balanced your disks groups if you have more than one per host.
You can also see the analysis of used capacity now.
Quality of Service (QoS) using IOPS limits
Now VSAN gets closer to traditional VMFS datastores in terms of storage QoS. The first releases of VSAN didn't allow IOPS limits/shares for VMs residing on VSAN datastores since the SIOC wasn't compatible with VSAN. With VSAN 6.2 administrators will have an option to limit IOPS per VM to be able to control "noisy neighbours".
QoS for VSAN 6.2 doesn't make any distinction between Reads and Writes.
It ought to be noticed that all IOs are normalised to 32KB block size. That means that if you, for instance, set a IOPS limit to 100 and block size of the IOs will be up to 32KB that VM will be able to get 100 IOPS. However, with block size of 64KB the VM will get only 50 IOPS.
New performance service and improved health service
VSAN 6.2 delivers new graphs and data to provide performance information at the cluster, host, virtual machine and virtual disk levels.
Interestingly, that performance data isn't stored in vCenter database. It is stored in as a separate VSAN object and has no dependecy on the vCenter.
Here are some examples of new types of graphs.
For instance, on the cluster level you can compare Virtual Machine IOPS and Throughput stats with Virtual SAN Backend stats. The latter shows the 'raw' amount of IOPS produced to deliver Virtual Machine IOPS, e.g. with Failures to tolerate set to 1 for each VM Write IO the VSAN backend has to produce 2 Write IOs to mirror data to two different hosts.
On the host level you can see Disk Group and Disk detailed stats in addition to VM IOPS and Throughput graphs mentioned above. So you can see how well-balanced your disks groups if you have more than one per host.
You can also see the analysis of used capacity now.
Health Services now has even more items to test and monitor. It is enabled by default and runs all tests every hour.
Apart from checking the health of all key services and different aspects of VSAN configuration (Networking, Storage, etc) it also checks all physical components of VSAN, e.g. controller driver, against VSAN Hardware Compatibility List.
Finally, VSAN 6.2 makes it way easier to upload support bundles to VMware Support to speed up the troubleshooting process.
Even though we haven't seen any new features of vSphere there is huge amount of information to digest - just have a look at other announcements of VMware Workspace One and Horizon 7.
