GCP Latency Reduction with Cloud Volumes ONTAP

As more and more applications migrate to the cloud, latency becomes an important benchmark for application performance.

When it comes to highly transactional applications with stringent performance requirements, storage latency can have a negative impact that will be felt all the way down to the end users. In Google Cloud, NetApp has a new way to help you avoid that.

In this blog we’ll cover the new Cloud Volumes ONTAP in Google Cloud support for NVMe caching via Flash Cache® intelligent caching. Coupled with Cloud Volumes ONTAP using the high write speed option, Google Cloud users now have a powerful tool to reduce overall storage latency.

Read on or follow these links to jump below:

• What Is Latency?
• What Causes GCP Latency?
• Reduce GCP Latency Using Cloud Volumes ONTAP with NVMe Caching
• Conclusion: Don’t Let Google Cloud Latency Slow You Down Anymore
• FAQs

What Is Latency?

A storage system's response time to a read or write request is referred to as its latency. If the time it takes to complete the request, i.e., the latency, is higher than the expected response time, it’s going to have a negative impact on the people and applications that depend on that storage.

There are several factors in cloud that could lead to increased latency:

• Network delays
  With data stored in the cloud and users everywhere around the world trying to access that data, good network connectivity is crucial to retrieving and sending back that data to user devices. Even within the cloud, there can be multiple hops for that data to take— between the web, application, and database servers—before it reaches a user. If there’s any delay in the network, it will increase latency.
• File systems
  Data stored in file systems can come with additional overheads, as file systems need to manage metadata when data is accessed. This could impact applications that need higher IOPS, and add to the latency.
• Infrastructure constraints
  Data slows as it passes through different devices. Moving across switches, routers, and firewalls in cloud data centers all has an impact on the overall latency. Latency is also affected by the server hardware and virtualization layer hosting the server, which have limitations in terms of data throughput and IOPS.
• Routing challenges
  Data passes through multiple networks to reach the user, and the route for this transit may not always be the most optimal. Inefficient paths can add to latency woes.

As cloud offers managed storage options and takes care of aspects such as networking, routing, and infrastructure, customers can adopt specific configurations to help reduce the overall storage latency.

What Causes GCP Latency?

Depending on the choices made for the computer, storage, and network settings, Google Cloud-specific configurations may result in an increase or decrease in GCP latency. Let’s take a look at some of these considerations that customers should be aware of.

• Google Cloud connectivity
  Customers can access applications hosted in GCP directly over the internet, or over hybrid connectivity options, such as VPNs and Interconnect.
  
  Of these solutions, only Interconnect guarantees dedicated bandwidth and availability, which helps reduce GCP latency. Users accessing data over a VPN or the internet are dependent on the stability of the network, and could in turn face higher GCP network latency.

• Virtual Private Cloud (VPC) architecture
  Google Cloud VPC is global in nature and offers a single virtual network that can span across multiple regions. Customers can also opt for a shared VPC where the same network is being used between different projects.
  
  Additional services such as load balancers, web application firewalls, and network virtual appliances, can also be part of the overall VPC architecture. All of these can add to the number of hops required for the data to reach the user from the storage layer, adding to GCP latency.
• Storage layer
  Google Cloud provides HDD- and SSD-based disks that support different IOPS and throughputs, depending on the disk type. Additionally, each VM type has a limitation on the maximum IOPS it can support. Selecting the right storage type and VM configuration plays an important role in reducing the overall latency. Making the wrong choice usually has a noticeable effect on latency.

Reduce GCP Latency Using Cloud Volumes ONTAP with NVMe Caching

For Google Cloud users, there’s a new way to reduce the effects of latency in your operations: Cloud Volumes ONTAP now supports NVMe caching on Google Cloud.

Cloud Volumes ONTAP is an enterprise grade solution from NetApp that delivers advanced storage management capabilities in all the leading cloud platforms: AWS, Azure, and Google Cloud. While AWS and Azure users have benefitted from NVMe caching for some time, Cloud Volumes ONTAP on Google Cloud can now take advantage of this feature as well.

Flash Cache offers a simple and straightforward solution for Cloud Volumes ONTAP customers to reduce Google Cloud Platform latency, without compromising storage efficiency. Flash Cache uses a local NVMe storage module to improve storage latencies and deliver better performance. NVMe intelligent caching enables real-time caching of recently accessed data and metadata to improve read performance. Read-intensive workloads such as databases, file services, LOB applications, and email can greatly benefit from this latency reduction.

In addition to NVMe caching, customers can accelerate application performance by enabling the high write speed setting for Cloud Volumes ONTAP.

You can choose from normal or high write speed while configuring Cloud Volume ONTAP:

• With normal write speed, data is written directly to the disk, thereby reducing the possibility of data loss during an outage
• With high write speed, the data is buffered in memory before writing to disk, thereby improving write performance

Starting from Cloud Volumes ONTAP version 9.13.0, Cloud Volumes ONTAP read/write performance is accelerated by Flash Cache and the high write speed option. The following GCP instance types deployed as HA pairs can take advantage of Flash Cache and high write speed:

• n2-standard-16
• n2-standard-32
• n2-standard-48 (new)
• n2-standard-64 (new)

Flash Cache and high write speed will both be enabled on the supported instances in this release. An upcoming release will include the capability to activate each of these features independently.

Conclusion: Don’t Let Google Cloud Latency Slow You Down Anymore

Storage latency is a major pain point for cloud administrators. Since there can be many factors that come together to increase latency, it can be difficult to pinpoint and eliminate specific bottlenecks. With Cloud Volumes ONTAP and NVMe intelligent caching Flash Cache technology, you can eliminate latency concerns to a great extent.

For more, check out this customer case study about using NVMe intelligent caching to speed up the high-demand design process in the EDA vertical.

FAQs

• What is latency in Google Cloud?

  Latency in Google Cloud is the time taken for applications to provide a meaningful response to user requests. Choosing the right storage type and optimizing the overall application architecture will help with reducing latency and deliver improved performance.

• How do I check Google Cloud latency?

  Google Cloud provides a Google Cloud latency dashboard that tracks that latency metrics for different traffic types. Using this dashboard, you can view the latency metrics between regions where the VMs are hosted and user devices accessing them from different geographies. In addition to Google Cloud latency between regions, it also provides visibility of Google Cloud latency between zones. There are also third-party tools that perform Google Cloud latency tests.

• Is AWS faster than GCP?

  Both AWS and GCP have multiple data centers over the globe connected with low latency connectivity between them and can be used for hosting applications with low latency and high performance requirements. The performance and latency in each platform will be dependent on the architecture and component specific configurations. Hence, it is difficult to definitely say whether one platform is faster than the other.