Raid Bottleneck and drive type

[JohanZoli]

Gents,

 

I would like to setup 2 esxi host based each one on:

> Dell R620 - Dual 2697v2 - 256Gb Ram

> Dual 10Gb network

> H710 controller (I believe the mini version as it is a 1U server)

> the chassis can hold up to 8 x 2.5 drives in the front

 

I have

> 16 drives 800Gb HGST SAS 12G that have tons of IOs and very good sequential read / write (8 drives per Dell R620)

> 16 drives 1Tb Micron 1100 SATA 6G (8 drives per R620)

 

Considering that I suspect the H710 is going to be a bottleneck here

> is it relevant to waste the HGST 12G in these hosts ? 

 

Considering I also have 4 x NVME Drives Samsung PM981 (3200Mb read / > 2000 write)

> would it be relevant to install them on adapter and use this space for database and cache ? 

> do you know any cards or mecanism that would do raid 1 (2xNVME drives per R620) ? (without having to sell a kidney)

 

 

Last but not least, I also purchased 2 massive JBOD (4U / 60 disk / SAS 3 - 12G - Zoning capabilities) and  I already have around 700Tb of spinning drive.

Should  I save the 800Gb HGST 12g I mentioned earlier and make some ZFS ultra fast storage ?  As I will make sure the enclosure, the SSD and the controller are 12G compatible ?

 

Thanks in advance for the feedback

 

Cheers

 

Johan

 

 

 

 

 

[leadeater]

Personally with the hardware you have I would have the ESXi hosts as compute only and have a storage server presenting storage to them over NFS or iSCSI or both (some things like block storage more).

[JohanZoli]

21 minutes ago, leadeater said:

Personally with the hardware you have I would have the ESXi hosts as compute only and have a storage server presenting storage to them over NFS or iSCSI or both (some things like block storage more).

Hi,

 

Thanks for the fast feedback ! Much appreciated.

 

The plan would be something like:

 

Step 1: Storage and Compute on the two Dell R620 knowing that it's not ideal. I have a price for hosting constraint on the one hand and my storage strategy is not fnished on the other hand. Having everything on two 1U Servers would make sense while some part of the project are "work in progress"

 

Step 2 : While the first phase is beeing implemented, I will have to work on making the best out of:

> 2 HGST JBOD - 4U - 60 Disk - Top Load (2 x QSFP connections)

> 28 x 8Tb SAS helium HDD

> 14 x 10 Tb SATA helium HDD

> 34 x 6Tb SATA HDD 7200

> 60 x 3Tb SAS HDD

> 24 x 800Gb SSD

> 6 x 1Tb NVME PM 981

 

Not all these drives will have to be installed in the JBOD

 

I am wondering if I need to go for

> CEPH

> Gluster

> 2 big ZFS / FreeNas with old school rsynch

 

The idea is, I need both resiliency and performance out of my storage. This storage will be mostly "file oriented"  SMB or NFS share. 

 

Regarding, the space I require for the VMs, I don't need 3 or 4 To, regarding the cache, 2Tb would be plenty.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

[leadeater]

35 minutes ago, JohanZoli said:

The idea is, I need both resiliency and performance out of my storage. This storage will be mostly "file oriented"  SMB or NFS share. 

Ceph is probably going to be above what you need then, as much as I like it. Probably just do Gluster over ZFS.

 

I would use the Microns in the R620s then migrate over to external storage when that is ready. Reason being that SATA is single port so can't be dual pathed, I would only use SAS in the external enclosures so you can dual path properly to two HBAs. Probably not actually an issue since the 60 bay enclosures have backplane expanders but I like to stick to SAS whenever I can as it makes it easier and you'll never have to worry about knocking out half the disks if you need to unplug a cable.