Simplivity VM block size

[ACA_RAY]

Hi there,

I'm hoping it's not too late for a Simplivity expert to perhaps enlighten me regarding the below.

I note in the Omnistack administration guide that it is suggested that NTFS volumes on the Windows VM's be formatted in 8K block/allocation size rather than the default of 4K for "best performance".

We have a lot of VM's that are formatted in 4K and would be a massive task to change them over. New volumes I have created have always been in 8K.

Basically my question is, what do you mean by "best performance"? Is it the speed of the VM itself, the Simplivity operations such as backups or copying a VM?

Does having the block size of 4K affect the de-duping and compression ratios compared to 8?

Would welcome your thoughts.

Cheers,

ACA

[StorageMike]

Hi

I've taken the following from the VMware guide - https://h20195.www2.hpe.com/v2/Getdocument.aspx?docname=a00019351enw.  it's a bit salesy but should help answer

HPE SimpliVity hyperconverged infrastructure performs hardware-assisted inline deduplication, compression, and optimization on all data at inception across all phases of the data lifecycle (primary, backup, WAN, archive, and in the cloud), usingfine data granularity of just 8 KB. By driving efficiencies at the point of origin andby offloading processor-intensivedata handling functions on to purpose-built data acceleration hardware, the solution optimizes the use of system resources and minimizes I/O and network traffic, accelerating data clone, backup, restore, and mobility operations.

HPE SimpliVity Data Virtualization PlatformData Management layer performs inline data deduplication, compression, and optimization on all data, at inception, across all tiers (DRAM, SSD, and HDD) and phases of the data lifecycle (primary, backup, WAN, archive, and in the cloud). All data is deduplicated, compressed, and optimized in 8 KB granularity by theHPE OmniStackAccelerator Cardinline before it is written to disk. With HPE SimpliVity, deduplication and compression does not require any intervention by the administrator since it is always on and accelerated by the HPE OmniStack Accelerator Card. Utilizing dedicated hardware ensures that deduplication can be performed 100% of the time nomatter the workload and without performance impact to the application.

[scott_svt · Accepted solution]

Hi @ACA_RAY,

To answer your question directly, It is more about minimising any potential overhead in terms of performance and capacity. It is important to note however that this would only apply in constrained environments where IOPS and latency are peaking. To be honest, I never suggest to anyone to convert things like C: drives due to the overhead of doing so, for such little real world gain. For and subsequent drives, it really depends on the data. Fileservers can obviously get some benefit, as can the likes of SQL DATA drives (but arguable these are misconfigured if they are not set to 64k - a multiple of 8k anyway). Also be aware of the 'noisy neigbour' effect. An improvement in one monster VM may have system-wide benefits, even for those VMs on 4k.

TL;DR: Yes there can be a small impact in terms of dedupeability and potentially raw performance at the very top of the range on constrained systems, but this has to be tempered by the amount of data and subsequent impact we are really talking about. Heavy hitting VMs (either large or those with high performance requirements) may benefit from 8k (or multiples thereof), but really, I wouldn't be going back to convert C: drives and the like. Stick to those drives that might offer the best bang for the buck and note that any improvement would be system-wide, not just for the machines that were converted to 8k.

Cheers,

Scott

[ACA_RAY]

Thanks @scott_svt that's answered perfectly. Really appreciate the insight, thank you.