Mount Tigris as a Filesystem

For AI workloads, most runtimes assume data is accessible via a filesystem. By mounting a Tigris bucket as a filesystem, data is no longer limited by local storage nor memory capacity.

We provide TigrisFS a fork of the original yandex-cloud/geesefs project with optimizations for Tigris. TigrisFS is a FUSE-based filesystem that mounts S3-compatible storage as a local filesystem. You can use TigrisFS to mount any Tigris bucket as if it was a normal Linux filesystem.

note

TigrisFS does not obfuscate the objects in the bucket, and does not require an additional metadata service (unlike JuiceFS). This means you can access the objects directly via the S3-compatible API using the standard S3 SDKs, tools, and libraries, as well as via the TigrisFS filesystem.

Prerequisites:

• A Tigris account
• A bucket you want to mount into your system (mybucket)
• A target location you want to mount your bucket to (~/data)

note

TigrisFS uses FUSE to mount buckets into the filesystem. FUSE has limitations involving multiple users accessing the same mountpoint. If you need multiple users (human or service) to access the same bucket, consider creating multiple systemd services mounting the bucket in each user's home directory.

Installing TigrisFS

Download the latest release for your machine from the list of TigrisFS releases

Debian Linux

You may need to install FUSE utils (fuse3 or fuse RPM/Debian package) first.

wget https://github.com/tigrisdata/tigrisfs/releases/download/v1.2.1/tigrisfs_1.2.1_linux_amd64.deb
dpkg -i tigrisfs_1.2.1_linux_amd64.deb

MacOS ARM64

You may need to install osxfuse/macfuse first.

curl -sL https://github.com/tigrisdata/tigrisfs/releases/download/v1.2.1/tigrisfs_1.2.1_darwin_arm64.tar.gz | sudo tar -xz -C /usr/local/bin

Configure credentials

Create a new access key with edit permissions on your target bucket in the Tigris Dashboard. Configure the aws-cli to utilize the credentials as described in the guide here.

Mount bucket

Using systemd

When installed from the DEB or RPM package, TigrisFS installs a systemd service to simplify mounting and configuration.

Mount as current user:

systemctl --user start tigrisfs@<bucket>

Mount as root:

systemctl start tigrisfs@<bucket>

The bucket is mounted at $HOME/mnt/tigrisfs/<bucket> for the current user, or /mnt/tigrisfs/<bucket> for root.

To automount the bucket at boot or user login, enable the systemd service:

systemctl enable tigrisfs@<bucket> # as root
systemctl --user enable tigrisfs@<bucket> # as current user

Using the command line

To mount the bucket using the command line, run the following command:

/usr/local/bin/tigrisfs <bucket> <mountpoint>

You did it! You now have a performant filesystem backed by bottomless object storage.

Maximizing performance

Cache sizing

TigrisFS uses cache for both read buffers when it needs to read data from your Tigris bucket, and write buffers when it needs to write data to Tigris. The cache can be configured using the --memory-limit flag when mounting the bucket. The default value is 1GB. You can increase it to 4GB or more, depending on available memory on your machine. For example, below is how to mount a bucket with a 4GB cache:

/usr/local/bin/tigrisfs --memory-limit 4096 <bucket> <mountpoint>

Improving write performance

TigrisFS uses a write-back cache to improve write performance. This means that when you write data to the mounted filesystem, it is first written to the cache and then flushed to the Tigris bucket in the background.

As mentioned above, the cache is used for both read and write buffers, so increasing the cache size will also improve write performance.

The other way to improve write performance is to increase the number of flushing threads. By default, TigrisFS uses 16 flushing threads. You can increase this number using the --max-flushers flag when mounting the bucket. The recommended value is 32 or more, depending on the number of CPU cores available on your machine.

/usr/local/bin/tigrisfs --memory-limit 4096 --max-flushers 64 <bucket> <mountpoint>

Improving performance for small files

Apart from the data cache mentioned above, TigrisFS also uses a metadata cache to improve performance for metadata operations. The metadata cache is enabled by default. For small files, the metadata cache can significantly improve performance, as it reduces the number of requests to the Tigris bucket for metadata operations.

To improve performance for small files, you can increase the TTL of the cached entries in the metadata cache using the --stat-cache-ttl flag when mounting the bucket. The default value is 60 seconds. For example, below is how to mount a bucket with a 15 minutes metadata cache TTL:

/usr/local/bin/tigrisfs --memory-limit 4096 --max-flushers 64 --stat-cache-ttl 15m <bucket> <mountpoint>