Expected performance (Hyper-V cluster with 40Gb)

Software-based VM-centric and flash-friendly VM storage + free version

Moderators: anton (staff), art (staff), Max (staff), Anatoly (staff)

Post Reply
SteveMC
Posts: 1
Joined: Mon Sep 24, 2018 3:16 pm

Mon Sep 24, 2018 3:56 pm

Hi,

We have a 2-node Windows Server 2016 Hyper-V Hyper-Converged Cluster with VSAN 8.0.4724, and are experiencing unexpectedly slow write performance (maximum of around 600MB/s write vs 3600GB/s read).
I'm after any hints as to where to look for configuration adjustments that may improve things.

The detailed set up is as follows:

2 x physical Server 2016 DataCenter Hosts, each with:
  • 2 x Intel Xeon E5-2620v4 CPUs
  • 512GB DDR4 RAM
  • 256GB RAID 1 SSDs (for OS)
  • 80TB of available SAS storage on a RAID 6 array, running on an Avago MegaRAID SAS 9361-8i controller (with 512MB of cache)
  • 2 x Mellanox ConnectX-4 40Gb copper NICs (1 assigned to StarWind Sync, the other assigned to iSCSI)
  • 2 x Intel 10Gb copper NICs (for client access, teamed and managed using Microsoft SCVMM)
  • 2 x Intel 1Gb copper NICs (management only)
I've got a 20TB StarWind image file set up in HA mode, which is in sync between the two nodes, and have set up iSCSI as per the guide at https://www.starwindsoftware.com/resour ... erver-2016.

Enabling iSER on the 40Gb NICs did increase the write-speed a fair amount, but I'm wondering if there's anything else I should check?

Many thanks,

Stephen
Boris (staff)
Staff
Posts: 805
Joined: Fri Jul 28, 2017 8:18 am

Mon Sep 24, 2018 4:27 pm

It's RAID6 that is causing slow performance for you. With RAID6 write performance is sometimes a real pain.
Guys from Storagecraft have described it pretty well in one of their articles:
RAID 6, after RAID 10, is probably the most common and useful RAID level in use today. RAID 6, however, is based off of RAID 5 and has another level of parity. This makes it dramatically safer than RAID 5, which is very important, but also imposes a dramatic write penalty. Each write operation requires the disks to read the data, read the first parity, read the second parity, write the data, write the first parity and then finally write the second parity.
Dramatic write penalty is what you experience in your system.This is why we recommend using only RAID10 for HDD-based arrays as specified in our related KB article https://knowledgebase.starwindsoftware. ... ssd-disks/
Post Reply