VMware Explore 2022 US, the new name of VMware’s historic VMworld conference, was filled with interesting announcements and sessions. However, what stood out among the most anticipated sessions was the unveiling of the vSphere 8.0, two and a half years after vSphere 7.0. And as you know, along with vSphere also comes VSAN, the distributed object storage solution that lets you leverage local disks to provide shared storage across the nodes in the cluster. Now VSAN is a well-oiled machine by now and most vSphere admins know how it works or has had to deal with it at some point in their career.
With VSAN 8.0, VMware has made important changes to VSAN by offering an alternative storage architecture that is optimized for modern hardware, namely NVMe-based TLC flash devices. This decision was driven by the fact that the existing design didn’t account so well for modern high-performing NVMe flash devices as opposed to SAS/SATA flash drives. This alternative method includes a new flexible storage architecture named Express Storage Architecture or ESA. The existing architecture (should we call it legacy already?) is now referred to as Original Storage Architecture or OSA. In this blog, we will have a look at VSAN 8.0 ESA, what it is made of, and its benefits.
New log-structured file system
While the architecture in VSAN 8 ESA is mostly based on the original VSAN (OSA), new layers have been added in the IO path to optimize performances and how data/metadata are handled.
A new patented and log-structured file system called LFS acts as one of the new layers in the VSAN stack. It reduces the time to ingest data and store metadata. Along with it comes an optimized log-structured object manager and data structure that aims at reducing the overhead needed for metadata processing. The end goal is to write data to storage devices without contention and achieve near device-level speeds.
Compression operations are moved up the VSAN stack
By moving compression operations up the stack where incoming writes are received, VSAN 8 ESA brings several significant enhancements:
- Reduced CPU overhead: The data needs only be compressed once instead of once for each copy of the data
- Reduced network payloads: Because the data is moved in a compressed state, less data needs to be conveyed on the network
- 4x improvements in compression ratios for every 4LB block written
- Compression managed per VM with Storage Policy Based Management (SPBM) as opposed to being a datastore-wide option in VSAN OSA
RAID-1 performances for RAID5/6
Just like with physical disk pools configured in RAID groups, the original VSAN architecture (VSAN OSA) implied that you sort of had to make a compromise on performance when configuring VMs with Erasure coding (RAID-5 or RAID-6) which would bring the most storage capacity as opposed to RAID-1 that would offer the best performance while wasting 50% of storage space.
With VSAN 8.0 ESA, you now get the same performance in RAID-5/6 mode as in RAID-1, hence maximizing the storage to performance ratio. This feat is partly thanks to the new LFS (log-structured file system). LFS drastically changes the way data is accessed and stored as the VSAN LFS.
In order to speed up operations, LFS will ingest incoming writes, coalesce/package them with metadata and commit them to a durable log for the object in what is called the performance leg, once done an ACK is sent back to the VM. In the performance leg, data is mirrored twice for a VM with SPBM RAID-5 and thrice for a VM with SPBM RAID-6.
As the performance leg section of the storage fills up, LFS then commits the data part of objects to the capacity leg, in a more traditional full striped fashion across the hosts while the metadata is moved to a metadata log.
With VSAN ESA, RAID5-6 erasure coding will be the recommended type for all workloads while RAID-1 mirroring will only be useful in stretched clusters and 2-nodes clusters. This also means you can now get RAID-5 erasure coding in 3-node clusters (although 4-nodes are still recommended).
Storage pools replace disk groups
Another major change in VSAN 8 ESA is that disk groups are gone. Mind you I said ESA, if you decide to go with VSAN 8 OSA you will still run the “legacy” method of configuring disks. Because of the new LFS and the fact that VSAN 8 ESA nodes will be full NVMe devices, there is no longer a need to have a cache and a capacity tier. Instead, all disks contribute to both cache and capacity in the VSAN datastore.
This means that less space is wasted due to having devices dedicated to caching as all of them contribute to capacity. On top of it, it significantly simplifies the setup and administration of the solution.
Note that it is supported to run storage devices of different sizes in the same host but like with OSA, VMware encourages organizations to aim toward consistency. Either way, early versions of VSAN 8 ESA will only support VSAN Ready Nodes which will obviously follow best practices. Be aware that the requirements to use VSAN 8 ESA restrict you to using only certified NVMe devices with a minimum of 4 of each along with 25Gbps networking.
Faster snapshots
As you know by now, VSAN 8 ESA opened a lot of doors for VMware to improve the platform. One of these is the introduction of a new native snapshotting mechanism. There aren’t many technical details about it yet but this new way of taking snapshots should make consolidation 100x faster and is fully supported with VADP so backup products like Vembu BDRSuite can benefit from it.
What happens to VSAN OSA (Original Storage Architecture)?
While VSAN 8 ESA will probably be where the bulk of the effort is focused at VMware, VSAN OSA is still supported and a valid architecture to run your VSAN clusters. When you create your VSAN cluster, you now get to choose between ESA and OSA.
No new features or specific announcements around VSAN OSA, apart from an increase in maximum cache tier size from 600GB to 1.6TB and vSAN Proactive insights, a new feature that warns users of potential hardware/software compatibility or support gaps (available in both ESA and OSA).
Wrapping up
VMware delivered a lot of interesting talks and features at VMware Explore 2022. VSAN 8 ESA is a personal favorite as it rethinks and improves a foundational component of the on-premise platform. While it isn’t possible to go from VSAN OSA to VSAN ESA since you need specific drives, we expect a large number of organizations will leverage this as the de-facto standard going forward, especially given the decreasing prices of NVMe storage devices.
Follow our Twitter and Facebook feeds for new releases, updates, insightful posts and more.
