Write-through cache in Azure Managed Instance for Apache Cassandra
Azure Managed Instance for Apache Cassandra now has a write-through cache (preview) feature through the use of L-series virtual machines (VMs). This implementation can help you minimize tail latency and enhance read performance, particularly for read-intensive workloads. These specific VM types are equipped with locally attached disks to increase input/output operations per second (IOPS) for read operations and to reduce tail latency.
Important
Write-through cache is in public preview. This feature is provided without a service-level agreement, and we don't recommend it for production workloads. For more information, see Supplemental Terms of Use for Microsoft Azure Previews.
Benefits of using write-through cache
Reduced tail latency: This implementation focuses on minimizing tail latency to enhance the user experience, particularly in time-sensitive applications, by reducing delays in read operations.
Optimization of read-intensive workloads: This implementation is for scenarios with heavy read workloads. The design specifically targets and improves read performance. It acknowledges the demands of applications where frequent read operations are the norm.
Empowerment of localized disks: The designated VM types feature locally attached disks, a key element that empowers architecture with increased IOPS for read operations. This design choice aims to optimize the efficiency and responsiveness of data retrieval.
Increased IOPS: Locally attached nonvolatile memory express (NVMe) disks boost IOPS for reads and reduce tail latency. This implementation seeks to provide a comprehensive solution that addresses performance bottlenecks for read-intensive workloads.
How to access L-series VMs
For existing Azure Managed Instance for Apache Cassandra clusters, you can access L-series VMs by adding a new datacenter:
On the Data Center pane, select the Add button.
In the Sku Size box, select your preferred L-series VM type.
Note
You can access new features of automatic replication when you're adding a new datacenter. Learn more.
After you create the datacenter, confirm that it appears in your list of datacenters.
Read-intensive scenario with write-through cache
The following analysis explores a sample test that compares the IOPS and latency between premium managed disks and NVMe for reads. Through a detailed examination of performance metrics in this comparison, you can get valuable insights into the effectiveness of write-through cache for locally attached NVMe disks.
IOPS for premium managed disks and local NVMe disks
The following example shows the IOPS of a RAID 0 array with four premium managed disks (present-day configuration).
The following example shows the IOPS of a RAID 1 array with local NVMe disks and premium managed disks.
The example that uses locally attached NVMe disks has a better read performance of 248K IOPS.
Latency for premium managed disks and local NVMe disks
The following example shows the throughput of a RAID 0 array with four premium managed disks (present-day configuration).
The following example shows the throughput of a RAID 1 array with local NVMe disks and premium managed disks.
The example that uses locally attached NVMe disks has a higher throughput capacity performance of 14.9 Gbps.
Note
Local disks are ephemeral and might be lost. But data stays intact in the network-attached premium managed disks, because we mirror data between them by using RAID technology. You can monitor the health of the RAID array through Azure Monitor metrics of RaidArrayDegradedMetric and RaidArrayRebuildMetric.