Performance of Storage I/O Control (SIOC) with SSD Datastores – vSphere 6.5
A brand new blog post from VMware – VMware VROOM! Blog
We would like to make a repost of this interesting topic.
Please find here the blog post from Julie Brodeur
With Storage I/O Control (SIOC), vSphere 6.5 administrators can adjust the storage performance of VMs so that VMs with critical workloads will get the I/Os per second (IOPS) they need. Admins assign shares (the proportion of IOPS allocated to the VM), limits (the upper bound of VM IOPS), and reservations (the lower bound of VM IOPS) to the VMs whose IOPS need to be controlled. After shares, limits, and reservations have been set, SIOC is automatically triggered to meet the desired policies for the VMs.
A recently published paper shows the performance of SIOC meets expectations and successfully controls the number of IOPS for VM workloads.
We tested SIOC under a variety of conditions using the storage micro-benchmark Flexible I/O Tester (FIO) on two vSphere 6.5 hosts each with 8 VMs (for a total of 16 VMs) which shared one VMFS datastore that was made up of SSDs from an EMC VNX5700 hybrid storage array. Our tests showed the following:
- Users can combine the three performance features (share, limit, and reservation) together to give a detailed performance specification (policy) for each VM. We tested the sample policies, provided in a vSphere cloud VM, for high priority, medium priority, and low priority VMs in our experiments to prove the efficiency of policy-based management.
- In real datacenters, over a long period, storage usage is not stable. It is very normal to see sudden spikes in I/O requests amid a low-demand period. We mimicked the datacenter environment by introducing different types of VMs like logging, production, and others. We assigned each of the VMs a different priority level, and we tested the system performance over a long period to make sure that the overall system performance was
- Uneven configuration is also very common among hosts. It is entirely possible that some hosts carry a heavier workload compared to others. We wanted to make sure that heavy computation demands on a host did not impact the I/O performance of the VMs located on that host. To mimic this condition, we created an uneven setting, using a shared datastore, and proved that the policy was still met, regardless of the number of VMs on each host.
To learn more about the performance experiments and test results, download Performance Implications of Storage I/O Control–Enabled SSD Datastores.
For instructions on how to use SIOC, see the vSphere 6.5 Resource Management guide > Chapter 8: Managing Storage I/O Resources.
The original post can be found here.
Some other links about Storage I/O Control: