Poor read performance - 6.5.0 On Ultra 4 Plus

[Vez]

I upgraded to 6.5.0 a while ago, mainly to get the new plex updates, and have read of a slight performance hit, and I expectedthat.

However write performance is exceptional, able to write at 100Mbs (I used to get 80Mb before upgrade)

Read has turned pittiful at ~30Mbs. (I used to get 100MB before upgrade)

All via SMB.

I believe this is bottle necking my VMware cluster, which is anoying to say the least (although that using NFS).

Have I missed a setting somewhere?  Why is read so slow!!!

[Vez · Accepted solution]

Might have found a solution:

https://community.spiceworks.com/topic/1581799-readynas-516-extremely-slow-nfs-performance

Will test a little later.

[Vez]

Right I have shut down my VMware cluster and read performance is as expected 100mb via SMB.  Previously I was still able to get 100mb through put on my desktop with the VMware cluster running!

So this would indicate to me an issue with NFS?  Any help would be welcomed.

[Vez]

Right, single VM, 2 NFS threads, and SMB file transfers are ~30mb.  Turn the VM off and I get 100Mb again from SMB.

[StephenB]

---

@Vez wrote:

Right, single VM, 2 NFS threads, and SMB file transfers are ~30mb.  Turn the VM off and I get 100Mb again from SMB.

---

Do you have any way to gauge how much I/O the VM is doing?

[Vez]

Its a 2012 domain controller sat there only serving DNS - with zero accounts on so no authentication.  I will check the vsphere logs later after spining it up again.

[Vez]

I will have a full graph within an hour, however with 2 VM's running a single ESXi host (the other has zero VM's running at this time for this test) are utilising a Datastore @:

Average Read: 239.65kbps

Average Write 101.022kbps

First VM is an Ubunto server (VPN server with no conections) - the Virtual disc is cached entirely on the host on SSD - so very little if any NFS usuage.

Second VM is 2012R2 - presently doiong windows updates.  All it does is server DNS for my home network (25% of the disc sits on SSD Cache).

SMB performance whist almost nothing is going on: an average of ~35Mbs.

This is with 2 NFS threads.  Something appears distictly wrong with prioritisation here.  This was never an issue before changing to 6.

Thoughts?  (Graph still compiling).

[Vez]

Revisiting this as I have finally resolved the dropbox issue:

I/O graph:

So as you can see, its sat doing literally nothing, yet renders down the SMB file transfer to circa 35 mbs.

This is particularly ponient again, as I am just creating a couple of new VM's (one at a time I may add) and now am back at 30mbs read spead from my windows workstation!  (Graph taken from the time of original issue).

Could this be a memory related constraint?

[Vez]

It appears if you have a folder shared via both CIFS and NFS, you get this impacted CIFS performance.

I have removed CIFS from said VMware stores and everything is flying again.  Read and Write on CIFS is now constantly maxing my Gbit connection to my workstation.

I will attempt to get vCenter up and running to see if this is actually the issue or not.

[mdgm-ntgr]

What are your settings for bit-rot protection and snapshots for the shares used by ESXi? It's recommended to have both of those disabled. Has that always been the case?

What about NFS async?

[Vez]

Yes both dissabled.

I spoke way too soon, it took me most of the evening to build a windows8 VM.

I attempted a test of speed from an RDP to VMDK via an already provisioned server,  Heady heights of 400kbps, and then it stopped after 10 seconds.

If would appear something is preventing NFS from functioning correctly, the fastest I have seen an ESXi host transfer data via NFS peaked at 4mbs.

Again CIFS is crippled whilst NFS is "busy" doing more or less nothing in the grans scheme of things.

(Can you give me info on VMware Certification for the 626x - though I am incredibly hesitant to invest if this cant be resolved).

[Vez · Accepted solution]

Might have found a solution:

https://community.spiceworks.com/topic/1581799-readynas-516-extremely-slow-nfs-performance

Will test a little later.

[mdgm-ntgr]

You can have a read of How to enable async mode in ReadyNAS OS 6 for improved NFS performance

[Vez]

Finally it appears my VMware cluster is functional again without crippling itself and the rest of the NAS.

Still interested to know performance stats for the 626x however, as this will likely be a pending purchase.

[mdgm-ntgr]

What would you like to know?

I can't test things with my RN626X at the moment as it's tied up at the moment with another project, but I could ask a colleague.

I did test transcoding video using my RN626X and that was a big step up from my 516 which is a big step up from the Pro 6 which is much better than the Ultra 4 Plus. I'm sure you'll love the performance that the RN626X provides.

In the live Q&A recently we were asked:

"what is the difference between Readynas rn626x & RN716X - so the older model actually has more ram, speed but the newer model is cheaper? both 10GB"

The answer was:

RN716X is Intel Xeon Ivy Bridge E3-1265Lv2 2.5GHz (3.5GHz Turbo) Quad Core with 16GB ECC

RN626X is Intel® D-1521 Quad Core 2.4GHz Xeon Server Processor with 8GB DDR4 ECC RAM
 

The E3-1265Lv2 uses a 22nm manufacturing process, while the D-1521 processor is much newer and uses a 14nm process. The D1521 will operate faster at a lower clock speed.