This page describes the approximate performance that Cloud Bigtable can provide under optimal conditions, factors that can affect performance, and tips for testing and troubleshooting Bigtable performance issues.
Performance for typical workloads
Bigtable delivers highly predictable performance that is linearly scalable. When you avoid the causes of slower performance described below, each Bigtable node can provide the following approximate throughput, depending on which type of storage the cluster uses:
|SSD||up to 10,000 rows per second||or||up to 10,000 rows per second||or||up to 220 MB/s|
|HDD||up to 500 rows per second||or||up to 10,000 rows per second||or||up to 180 MB/s|
These estimates assume that each row contains 1 KB of data.
In general, a cluster's performance scales linearly as you add nodes to the cluster. For example, if you create an SSD cluster with 10 nodes, the cluster can support up to 100,000 rows per second for a typical read-only or write-only workload.
Plan your Bigtable capacity
Trade-off between high throughput and low latency
When planning your Bigtable clusters, it is important to think about the trade-off between throughput and latency. Bigtable is used in a broad spectrum of applications, and different use cases can have different optimization goals. For example, for a batch data processing job, you might care more about throughput but less about latency. On the other side, an online service that serves user requests might prioritize lower latency over throughput. As a result, it is important to plan the capacity accordingly.
The numbers in the Performance for typical workloads section are achievable when you prioritize throughput, but the tail latency for Bigtable under such a load might be too high for latency-sensitive applications. In general, Bigtable offers optimal latency when the CPU load for a cluster is under 70%. For latency-sensitive applications, however, we recommend that you plan at least 2x capacity for your application's max Bigtable queries per second (QPS). This capacity ensures that your Bigtable cluster runs at less than 50% CPU load, so it can offer low latency to front-end services. This capacity also provides a buffer for traffic spikes or key-access hotspots, which can cause imbalanced traffic among nodes in the cluster.
Trade-off between storage usage and performance
Another consideration in capacity planning is storage. The storage capacity of a cluster is determined by the storage type and the number of nodes in the cluster. When the amount of data stored in a cluster increases, Bigtable optimizes the storage by distributing the amount of data across all the nodes in the cluster.
You can determine the storage usage per node by dividing the cluster's storage utilization (bytes) by the number of nodes in the cluster. For example, consider a cluster that has three HDD nodes and 9 TB of data. Each node stores about 3 TB, which is 18.75% of the HDD storage per node limit of 16 TB.
When storage utilization increases, workloads can experience an increase in query processing latency even if the cluster has enough nodes to meet overall CPU needs. This is because the higher the storage per node, the more background work such as indexing is required. The increase in background work to handle more storage can result in higher latency and lower throughput.
For latency-sensitive applications we recommend that you keep storage utilization per node below 60%. If your dataset grows, add more nodes to maintain low latency.
For applications that are not latency-sensitive, you can store more than 70% of the limit, as explained in Storage per node.
Run your typical workloads against Bigtable
Always run your own typical workloads against a Bigtable cluster when doing capacity planning, so you can figure out the best resource allocation for your applications.
Google's PerfKit Benchmarker uses YCSB to benchmark cloud
services. You can follow the
PerfKitBenchmarker tutorial for Bigtable
to create tests for your own workloads. When doing so, you should tune the
parameters in the benchmarking config
yaml files to make sure that the
generated benchmark reflects the following characteristics in your production:
- Total size of your table. This can be proportional, but use at least 100 GB.
- Row data shape (row key size, number of columns, row data sizes, etc.)
- Data access pattern (row key distribution)
- Mixture of reads vs. writes
Refer to Testing performance with Bigtable for more best practices.
Causes of slower performance
There are several factors that can cause Bigtable to perform more slowly than the estimates shown above:
- You read a large number of non-contiguous row keys or row ranges in a single read request. Bigtable scans the table and reads the requested rows sequentially. This lack of parallelism affects the overall latency, and any reads that hit a hot node can increase the tail latency. See Reads and performance for details.
- The table's schema is not designed correctly. To get good performance from Bigtable, it's essential to design a schema that makes it possible to distribute reads and writes evenly across each table. See Designing Your Schema for more information.
- The rows in your Bigtable table contain large amounts of data. The performance estimates shown above assume that each row contains 1 KB of data. You can read and write larger amounts of data per row, but increasing the amount of data per row will also reduce the number of rows per second.
- The rows in your Bigtable table contain a very large number of cells. It takes time for Bigtable to process each cell in a row. Also, each cell adds some overhead to the amount of data that's stored in your table and sent over the network. For example, if you're storing 1 KB (1,024 bytes) of data, it's much more space-efficient to store that data in a single cell, rather than spreading the data across 1,024 cells that each contain 1 byte. If you split your data across more cells than necessary, you might not get the best possible performance. If rows contain a large number of cells because columns contain multiple timestamped versions of data, consider keeping only the most recent value. Another option for a table that already exists is to send a deletion for all previous versions with each rewrite.
The cluster doesn't have enough nodes. A cluster's nodes provide compute for the cluster to handle incoming reads and writes, keep track of storage, and perform maintenance tasks such as compaction. You need to make sure that your cluster has enough nodes to satisfy the recommended limits for both compute and storage. Use the monitoring tools to check whether the cluster is overloaded.
- Compute - If the CPU of your Bigtable cluster is overloaded, adding more nodes can improve performance by spreading the workload across more nodes.
- Storage - If your storage usage per node has become higher than recommended, you need to add more nodes to maintain optimal latency and throughput, even if the cluster has enough CPU to process requests. This is because increasing storage per node increases the amount of background maintenance work per node. For details, see Trade-offs between storage usage and performance.
The Bigtable cluster was scaled up or scaled down recently. After you increase the number of nodes in a cluster to scale up, it can take up to 20 minutes under load before you see a significant improvement in the cluster's performance. When you decrease the number of nodes in a cluster to scale down, try not to reduce the cluster size by more than 10% in a 10-minute period to minimize latency spikes.
The Bigtable cluster uses HDD disks. In most cases, your cluster should use SSD disks, which have significantly better performance than HDD disks. See Choosing between SSD and HDD storage for details.
There are issues with the network connection. Network issues can reduce throughput and cause reads and writes to take longer than usual. In particular, you might see issues if your clients are not running in the same zone as your Bigtable cluster, or if your clients run outside of Google Cloud.
You are using replication but your application is using an out-of-date client library. If you observe increased latency after enabling replication, make sure the Cloud Bigtable client library that your application is using is up to date. Older versions of the client libraries might not be optimized to support replication. See Bigtable client libraries to find your client library's GitHub repository, where you can check the version and upgrade if you need to.
You enabled replication but didn't add more nodes to your clusters. In an instance that uses replication, each cluster must handle the work of replication in addition to the load it receives from applications. Underprovisioned clusters can cause increased latency. You can verify this by checking the instance's CPU usage charts in the Google Cloud console.
Because different workloads can cause performance to vary, you should perform tests with your own workloads to obtain the most accurate benchmarks.
Bigtable performs best with large tables that are frequently accessed. For this reason, if you start sending requests after a period of no usage, you might observe high latency while Bigtable reestablishes connections.
If you know that you will sometimes be sending requests to a Bigtable table after a period of inactivity, you can try the following strategies to keep your connection warm and prevent this high latency. These can also help performance during periods of low QPS.
- Send a low rate of artificial traffic to the table at all times.
- Configure the connection pool to ensure that steady QPS keeps the pool active.
If you are using the Cloud Bigtable client for Java, you can use the following strategies as well:
During cold starts or periods of low QPS, the number of errors that Bigtable returns is more relevant than the percentage of operations that return an error.
Replication and performance
Enabling replication will affect the performance of a Bigtable instance. The effect is positive for some metrics and negative for others. You should understand potential impacts on performance before deciding to enable replication.
Replication can improve read throughput, especially when you use multi-cluster routing. Additionally, replication can reduce read latency by placing your Bigtable data geographically closer to your application's users.
Although replication can improve availability and read performance, it does not increase write throughput. A write to one cluster must be replicated to all other clusters in the instance. As a result, each cluster is expending CPU resources to pull changes from the other clusters. Write throughput might actually go down because replication requires each cluster to do additional work.
For example, suppose you have a single-cluster instance, and the cluster has 3 nodes:
If you add nodes to the cluster, the effect on write throughput is different than if you enable replication by adding a second 3-node cluster to the instance.
Adding nodes to the original cluster: You can add 3 nodes to the cluster, for a total of 6 nodes. The write throughput for the instance doubles, but the instance's data is available in only one zone:
With replication: Alternatively, you can add a second cluster with 3 nodes, for a total of 6 nodes. The instance now writes each piece of data twice: when the write is first received and again when it is replicated to the other cluster. The write throughput does not increase, and might go down, but you benefit from having your data available in two different zones:
In these examples, the single-cluster instance can handle twice the write throughput that the replicated instance can handle, even though each instance's clusters have a total of 6 nodes.
When you use multi-cluster routing, replication for Bigtable is eventually consistent. As a general rule, it takes longer to replicate data across a greater distance. Replicated clusters in different regions will typically have higher replication latency than replicated clusters in the same region.
As explained in Write throughput, when an instance uses replication, each cluster in the instance must handle the work of replication in addition to the load it receives from applications. For this reason, a cluster in a multi-cluster instance often needs more nodes than a cluster in a single-cluster instance with similar traffic.
App profiles and traffic routing
Depending on your use case, you will use one or more app profiles to route your Bigtable traffic. Each app profile uses either multi-cluster or single-cluster routing. The choice of routing can affect performance.
Multi-cluster routing can minimize latency. An app profile with multi-cluster routing automatically routes requests to the closest cluster in an instance from the perspective of the application, and the writes are then replicated to the other clusters in the instance. This automatic choice of the shortest distance results in the lowest possible latency.
An app profile that uses single-cluster routing can be optimal for certain use cases, like separating workloads or to have read-after-write semantics on a single cluster, but it will not reduce latency in the way multi-cluster routing does.
To understand how to configure your app profiles for these and other use cases, see Examples of Replication Settings.
Dropping row ranges
If possible, avoid dropping a row range in an instance that uses replication because the operation is slow and the CPU usage increases during the operation.
How Bigtable optimizes your data over time
To store the underlying data for each of your tables, Bigtable shards the data into multiple tablets, which can be moved between nodes in your Bigtable cluster. This storage method enables Bigtable to use two different strategies for optimizing your data over time:
- Bigtable tries to store roughly the same amount of data on each Bigtable node.
- Bigtable tries to distribute reads and writes equally across all Bigtable nodes.
Sometimes these strategies conflict with one another. For example, if one tablet's rows are read extremely frequently, Bigtable might store that tablet on its own node, even though this causes some nodes to store more data than others.
As part of this process, Bigtable might also split a tablet into two or more smaller tablets, either to reduce a tablet's size or to isolate hot rows within an existing tablet.
The following sections explain each of these strategies in more detail.
Distributing the amount of data across nodes
As you write data to a Bigtable table, Bigtable shards the table's data into tablets. Each tablet contains a contiguous range of rows within the table.
If you have written only a small amount of data to the table, Bigtable will store all of the tablets on a single node within your cluster:
As more tablets accumulate, Bigtable will move some of them to other nodes in the cluster so that the amount of data is balanced more evenly across the cluster:
Distributing reads and writes evenly across nodes
If you've designed your schema correctly, then reads and writes should be distributed fairly evenly across your entire table. However, there are some cases where you can't avoid accessing certain rows more frequently than others. Bigtable helps you deal with these cases by taking reads and writes into account when it balances tablets across nodes.
For example, suppose that 25% of reads are going to a small number of tablets within a cluster, and reads are spread evenly across all other tablets:
Bigtable will redistribute the existing tablets so that reads are spread as evenly as possible across the entire cluster:
Test performance with Bigtable
If you're running a performance test for an application that depends on Bigtable, follow these guidelines as you plan and execute your test:
- Test with enough data.
- If the tables in your production instance contain a total of 100 GB of data or less per node, test with a table of the same amount of data.
- If the tables contain more than 100 GB of data per node, test with a table that contains at least 100 GB of data per node. For example, if your production instance has one four-node cluster, and the tables in the instance contain a total of 1 TB of data, run your test using a table of at least 400 GB.
- Test with a single table.
- Stay below the recommended storage utilization per node. For details, see Storage utilization per node.
- Before you test, run a heavy pre-test for several minutes. This step gives Bigtable a chance to balance data across your nodes based on the access patterns it observes.
- Run your test for at least 10 minutes. This step lets Bigtable further optimize your data, and it helps ensure that you will test reads from disk as well as cached reads from memory.
Troubleshoot performance issues
If you think that Bigtable might be creating a performance bottleneck in your application, be sure to check all of the following:
- Look at the Key Visualizer scans for your table. The Key Visualizer tool for Bigtable generates new scan data every 15 minutes that shows the usage patterns for each table in a cluster. Key Visualizer makes it possible to check whether your usage patterns are causing undesirable results, such as hotspots on specific rows or excessive CPU utilization. Learn how to get started with Key Visualizer.
- Try commenting out the code that performs Bigtable reads and writes. If the performance issue disappears, then you're probably using Bigtable in a way that results in suboptimal performance. If the performance issue persists, the issue is probably not related to Bigtable.
Ensure that you're creating as few clients as possible. Creating a client for Bigtable is a relatively expensive operation. Therefore, you should create the smallest possible number of clients:
- If you use replication, or if you use app profiles to identify different types of traffic to your instance, create one client per app profile and share the clients throughout your application.
- If you don't use replication or app profiles, create a single client and share it throughout your application.
If you're using the HBase client for Java, you create a
Connectionobject rather than a client, so you should create as few connections as possible.
Make sure you're reading and writing many different rows in your table. Bigtable performs best when reads and writes are evenly distributed throughout your table, which helps Bigtable distribute the workload across all of the nodes in your cluster. If reads and writes cannot be spread across all of your Bigtable nodes, performance will suffer.
If you find that you're reading and writing only a small number of rows, you might need to redesign your schema so that reads and writes are more evenly distributed.
Verify that you see approximately the same performance for reads and writes. If you find that reads are much faster than writes, you might be trying to read row keys that do not exist, or a large range of row keys that contains only a small number of rows.
To make a valid comparison between reads and writes, you should aim for at least 90% of your reads to return valid results. Also, if you're reading a large range of row keys, measure performance based on the actual number of rows in that range, rather than the maximum number of rows that could exist.
Use the right type of write requests for your data. Choosing the optimal way to write your data helps maintain high performance.
Check the latency for a single row. If you observe unexpected latency when sending
ReadRowsrequests, you can check the latency of the first row of the request to narrow down the cause. By default, the overall latency for a
ReadRowsrequest includes the latency for every row in the request as well as the processing time between rows. If the overall latency is high but the first row latency is low, this suggests that the latency is caused by the number of requests or processing time, rather than by a problem with Bigtable.
- Learn how to design a Bigtable schema.
- Find out how to monitor Bigtable performance.
- Learn how to troubleshoot issues with Key Visualizer.
- View sample code for programmatically adding nodes to a Bigtable cluster.