When sizing Humio, your choice depends on your usage patterns, so we recommend first doing an example setup to see how Humio works with your data. The following provides some examples. For clustered setups, and/or setups with data replication, we currently recommend contacting Humio Technical Support for specific suggestions.
With Humio, the limiting factor is usually query speed, not ingest capacity. Query speed depends upon the number of CPUs, available RAM, and disk speed. The following numbers are all general recommendations; your exact circumstances may be different.
A running Humio instance will use something like 10-20% of available RAM — the rest of available memory
is used for OS-level file system cache. If you’re, for example, on an AWS
m4.4xlarge (60 GB RAM, 16 vCPUs), then you would typically see
~10 GB used for running Humio, and ~50 GB available for file system caches. The 50 GB cache represents compressed Humio
data files which are optimized for querying; this way you will typically observe that recently read or written data
is faster to query.
Data is typically compressed 5-10x, depending on what it is. Your mileage may wary, but short log lines (HTTP access logs, syslog) compress better than longish JSON logs (such as those coming from Metricbeat).
For data available as compressed data in the OS-level cache, Humio generally provides query speed at 1 GB/s/vCPU,
or 1 GB/s/hyperthread. So, on a
m4.4xlarge instance with 16 vCPUs, you observe ~16GB/s queries. If your compression
ratio is 6x, then that means that the last 50GB x 6 = 300GB of ingested data can be read at this speed.
You should be able to keep 48 hours of data accessible for fast querying for a good experience, so
we recommend that a
m4.4xlarge is used for up to 150 GB/day ingest. This will let you do a full scan of
a day’s worth of data in less than 10 seconds. Many queries will be faster because you usually narrow the search
by specifying tags or other time limitations.
On this setup, a 300 GB/day ingest will use ~5% of your CPU load, so there is plenty of headroom for data spikes and running dashboards.
Searches going beyond what fits in the OS-level file system caches can be significantly slower, as they depend on disk I/O performance. If you have sufficiently fast NVME-drives or similar that can deliver the compressed data as fast as the CPU cores can decompress them, then there is virtually no penalty for doing searches that extend beyond the size of the page cache. We built Humio to run on local SSDs, so it is not (presently) optimized to run on high-latency EBS storage or slow spinning disks. But it will work, especially if most searches are “live” or within the page cache. Humio reads files using “read ahead” instructions to the Linux kernel, which allows for file systems on spinning disks to read continuous ranges of blocks rather than random blocks.
Your machine has 64 GB of RAM, 8 hyper threads (4 cores), and 1 TB of storage. Your machine can hold 460 GB of ingest data compressed in RAM, and process 8 GB/s. In this case that means that 10 seconds worth of query time will run through 80 GB of data. So this machine fits an 80 GB/day ingest, with +5 days data available for fast querying.
You can store 7.2 TB of data before your disk is 80% full, corresponding to 90 days at 80 GB/day ingest
For AWS, we recommend starting with these instance types. This represents setups that can hold 48 hours compressed ingest data in RAM; powerful enough to do a full table scan of ~24 hours of data in less than 10 seconds.
|Instance Type||Daily Ingest||RAM||vCPUs||Notes|
||600 GB||256||64 (2 CPUs)||run with per-instance cluster|
||400 GB||160||40 (2 CPUs)||run with per-instance cluster|
||200 GB||60||16||single node per instance|
||100 GB||40||8||single node per instance|
||30 GB||15||4||single node per instance|
For multi-socket machines (with multiple physical CPUs), we recommend running Humio as an per-instance cluster, with each Humio node tied to a single physical CPU.
With EBS storage, you will see query performance drop ~100x when querying beyond what is loaded into memory caches.
Alternatively, for a little more convenience and performant setup, you can run
instances which have more RAM, and let you store the data on local SSDs.
|Instance Type||Daily Ingest||RAM||vCPUs||Notes|
||600 GB||488||64 (2 CPUs)||run with per-instance cluster|
||300 GB||244||32 (2 CPUs)||run with per-instance cluster|
||150 GB||122||16||single node per instance|
||70 GB||61||8||single node per instance|
For instance an
i3.4xlarge would be suitable for 150 GB/day ingest, holding 5 days
of data in cache, and because of the SSDs this would be avoiding the “cliff” when
the cache runs full. The 3.8 TB SSD would hold ~150 days of ingest data.
With ephemeral SSD storage, you’d want to set up EBS instances for live backup (and Kafka storage), so that you can load the Humio data onto a fresh machine quickly. Humio live backup live-replicates all data to a separate network drive such that data loss is prevented even for ephemeral disks. For more information, see Backup.
Running many live queries/dashboards is less of an issue with Humio than most other similar products, because these are kept in-memory as a sort of in-memory materialized view. The time spent on updating them is part of the ingest flow and thus having many live queries increases the CPU usage for the ingest process. When initializing such queries, it does need to run a historic query to fill in past data, and that can take some time in particular if it extends beyond what fits within the compressed-data in memory horizon.
FIO is a great tool for testing the IO performance of your system. It is available as a APT package on Ubuntu too:
sudo apt install fio
fio either with all options on the commandline or through a “jobfile”.
sudo fio --filename=/data/fio-test.tmp --filesize=1Gi --bs=256K -rw=read --time_based --runtime=5s --name=read_bandwidth_test --numjobs=8 --thread --direct=1
Here are a sample contents for a jobfile that somewhat mimics how
Humio reads from the file system. Make sure to replace
/data/fio-tmp-dir with the path to somewhere on the disk where your
humio-data would be located, once installed.
[global] thread rw=read bs=256Ki directory=/data/fio-tmp-dir direct=1 [read8] stonewall size=1Gi numjobs=8
fio --bandwidth-log ./humio-read-test.fio # Clean tmp files from fio: rm /data/fio-tmp-dir/read8.?.?
and then look for the lines at the bottom of the very detailed report similar to…
Run status group 0 (all jobs): READ: bw=3095MiB/s (3245MB/s), 387MiB/s-431MiB/s (406MB/s-452MB/s), io=8192MiB (8590MB), run=2375-2647msec
This example is an NVME providing ~3 GB/s read performance. This allows for searching up to (9*3) GB/s of uncompressed data if the system has sufficient number of CPU cores according to the rules of thumb above. This NVME is well matched to a CPU with 32 hyper-threads (16 hardware cores.)