Caching VDI VM Reads and Writes to In-Host Intel Optane SSD Improves VM Latencies

Use Case:
VDI Performance

Location:
San Leandro, California

Challenges:

  • Customer wanted low VM latencies for their end-user VDI VMs. However, CEPH storage latencies were not very consistent and would spike from time to time.

Benefits:

  • VirtuCache ensures that VDI VM latencies are under 5ms at all times.

Caching VDI VM Reads and Writes to In-Host Intel Optane SSD Improves VM Latencies

The Virtunet Difference
The customer selected VirtuCache because:
  • VirtuCache is the only solution in the market that can improve storage performance of CEPH, even an All-Flash CEPH, as was the case here.
  • The only other alternative was to replace CEPH with a high-end All-Flash array, which would have been much more expensive.
  • April 29, 2020:

    By caching all VM read and write operations to Intel Optane NVME SSDs installed in VMware hosts, latencies of VDI VMs was much reduced. This resulted in improved VM performance and quicker VM boot times.

    Customer’s VDI Environment.

    City of San Leandro uses VMware’s Horizon software to serve virtual desktops to 300 end users. Horizon was running in a 9-host ESXi 6.7 cluster. Their shared storage was built using open source CEPH storage software. CEPH was running in VMs, and it pooled together locally attached SSDs in ESXi hosts and presented this storage over iSCSI to the same ESXi hosts.

    High Storage Latencies in CEPH.

    CEPH exhibits high latencies, especially for writes. This is due to high software overhead in CEPH, especially in the CEPH process that makes two copies of all data across different nodes (for data protection). Also, when CEPH is run in VMs, as was the case here, you have the additional overhead of VMware’s VMFS filesystem. Even though the customer was using SSDs with CEPH, write latencies were high from time to time.

    VirtuCache configuration.

    VirtuCache was configured to cache to 375GB P4800X Intel Optane SSD per host.

    In VirtuCache, you can apply three caching policies either at the Datastore or VM level:

    1. Write Through Caching policy caches reads but not writes.
    2. ‘Write-Back with One Replica’ caching policy caches reads and writes, and mirrors (replicates) writes to one additional host in the ESXi cluster. VirtuCache does this to protect against data loss if the host fails.
    3. ‘Write-Back with No Replica’ works like ‘Write-Back with One Replica’ caching policy, except that it does not replicate writes to another host, which means that if a host fails, there will be data loss for the VMs that have this policy.

    The customer has deployed non-persistent VDI VMs using Horizon View Linked Clones. Since these VDI VMs are non-persistent, they are OK with data loss in case of host failure. Hence ‘Write-Back with No Replica’ caching policy is applied to these VMs. The advantage of this policy versus the ‘Write-Back with One Replica’ policy is that there is no network overhead when it comes to caching writes, and so VM write latencies are really low. VM write latency is equal to the latency of the cache media, versus in the case of ‘Write-Back with One Replica’, where the VM write latency is the sum of network latency and cache media latency.

    For the customer’s VDI management VMs like Connection Server, Composer Server, and vCenter VM, ‘Write-Back with One Replica’ caching policy is used, since there can be no data loss in these VMs.

    Performance Improvement with VirtuCache.
    VDI VM latency. 20 milliseconds, before VirtuCache. 1 millisecond, with VirtuCache caching to Intel Optane P4800X SSD.
    Download Trial Contact Us