Performace one big vs several small

[chell]

Hi

I'm wondering if anyone has compared the performance of running several smaller starwind drives verses one big drive.
Star wind is on a raid 6 32TB (6x8tb), plus 1TB SSD for caching.
I'm going to be running 3 clusters. 2 for file storage and one for the virtual machines.
I'm deciding between running 3 x 4TB LSFS starwind drives. each will have their own ssd cache, ram cache and duplication enabled.
Or
running 1 x 12tb starwind drive LSFS with the entire ssd cache, ram cache and duplication enabled. Within this drive I'll create 2 x 4tb thick drives. I assume the duplication of the 12tb drive will allow me to create the thick drives without taking up unneeded space.
this should be a more efficient use of ssd and ram cache and of duplication as much of the data duplicated between the 3 clusters. Of course the cost is that any data written/read from the thick drives needs to go through starwind twice.
This is all assuming starwind will even work in this arrangement.
I guess i could use windows iscsi target server instead. if there is a way to have starwind and Microsoft iscsi server to co exist on a system.

[Petas3]

Hi,

1] Deduplication is not ideal for virtualization - you should avoid it when deploying virtual machines
2] SSD cache is nearly useless for data storage - I would dedicate it all for virtualization device
3] RAM disk is extremely important if working with RAID6, dedicate most for VM storage - but I think if you have only 1 Starwind Node it wont enable write defer features automatically and forcing it is asking for trouble...

I would do 2 devices:

1] VM - LSFS, No Deduplication, 100% SSD cache, 80-100% RAM cache, No write defer on single node - be prepared write performance will be bad
2] Storage - LSFS, Deduplication (mind it takes a lot of RAM to run it on Starwind, you can still use Windows Deduplication) - you can add it to Windows and make a CSV HA share in Cluster Manager

Running multiple devices can increase performance depending on workload and application optimizations.
You can have both Starwind iSCSI and Windows iSCSI, but you need to edit Starwind iSCSI port to different value in cfg file while Starwind service is stopped. Know, that Starwind is kind of faster .

[PoSaP]

I would not agree with

1] Deduplication is not ideal for virtualization - you should avoid it when deploying virtual machines

Actually, it can help to save space.
But in general, you gave good explanation

[Boris (staff)]

Indeed, deduplication is all about saving space when deploying dozens of similar VMs. It can be nearly useless if you have a small number of VMs or when most of them have different operating systems installed. Deduplication is great when the same OS is installed on most VMs. chell, does your environment look similar to this? If so, LSFS could be good for you. Otherwise you might benefit from the flat image files. Just consider the deduplication ratio and calculate whether it is better to have your data deduplicated and use only 1/3 of your underlying storage capacity for StarWind storage devices or use flat images with full capacity of your disk array.

L2 (SSD) cache is a nice thing for read intensive environments. chell, is your environment read-intensive? StarWind VSAN uses L2 cache in write-through mode and this improves only read operations with no write operations fostered. If your environment is not read-intensive, I would follow the piece of advice by Petas3 and use the SSD volume for mission critical VMs that need faster storage under the StarWind VSAN provisioned disks.

Generally, before deciding in favor of LSFS you would need to properly analyze your environment and define how good this will be for you.

[chell]

Is it possible to turn deduplication off for an existing volume?

[Oleg(staff)]

This option is not implemented at the moment.