Best disk-performance for high-workload database VM

[SnoppyFloppy]

Hi

I'm about to deploy a VM dedicated to a database and I want to know how to get the best disk-performance.

Unfortunately I don't have a spare ssd at the moment that I can pass-through so I'm thinking that an iSCSI-share on one of my ssd zpools (raid10-equivalent) seems like the best option.

But I would like a second opinion before I go about with it.

Thanks in advance.

 

[sretalla]

Generally Databases want lots of IOPS for their logs (and to a lesser extent for the DB itself).

SSDs are far superior to HDDs for IOPS.

If you are running the VM on TrueNAS, why wouldn't you add the disk as a ZVOL directly to the VM rather than iSCSI?

 

[hescominsoon]

I have a clinet where i have zvols exposed over ISCSI to a hyper-v host directly. The iscsi extents are used for a server 2016 vm, a windows nav 2016 nav vm, and a mssql VM. Performance works very well and is rock stable.

 

[ChrisRJ]

I'm about to deploy a VM dedicated to a database and I want to know how to get the best disk-performance.

What does that mean in absolute figures? People have vastly(!) different ideas here, so that statement alone does not help much ;-)

 

[SnoppyFloppy]

If you are running the VM on TrueNAS, why wouldn't you add the disk as a ZVOL directly to the VM rather than iSCSI?

My thinking is that if I create a dataset rather than a zvol I can adjust the record size to 16K to match the default page size of the DB which afaik can't be done on a zvol. So as I see it, there's basicly three options and I don't know which performes best for a DB:
1) optimized dataset + iSCSI
2) optimized dataset + NFS
3) un-optimized zvol + mount

What does that mean in absolute figures?

I cant tell you. I just want to get the pest possible performance out the hardware at hand.

 

[fahadshery]

My thinking is that if I create a dataset rather than a zvol I can adjust the record size to 16K to match the default page size of the DB which afaik can't be done on a zvol. So as I see it, there's basicly three options and I don't know which performes best for a DB:
1) optimized dataset + iSCSI
2) optimized dataset + NFS
3) un-optimized zvol + mount



I cant tell you. I just want to get the pest possible performance out the hardware at hand.

you can try phoronix-test-suite benchmarks

 

[sretalla]

My thinking is that if I create a dataset rather than a zvol I can adjust the record size to 16K to match the default page size of the DB which afaik can't be done on a zvol.

When you create a ZVOL, you have the option to select the block size... I wouldn't call that "can't be done".

Of course a block device is then formatted with its own filesystem which may match (or not) that block size, so you need to think about that based on the filesystem chosen.