This blog is the second in the machine learning series following the previous one, which discussed Alluxio’s solution to improve training performance and simplify data management. With the help of Alluxio, loading data from cloud storage, training and caching data can be done in a transparent and distributed way as a part of the training process, thus improving training performance and simplifying data management. In this blog 2 of the series, we focus on comparing traditional solutions with Alluxio’s.
Alluxio Community Day VIII
Leading experts present their latest ML & AI use cases, share Alluxio & Iceberg integration experience, and explain how to accelerate analytics workloads with Alluxio & Spark.
Upcoming Webinar: Architecting a Heterogeneous Data Platform Across Clusters, Regions, and Clouds
Learn how Alluxio can dramatically improve end-to-end distributed machine learning speeds in the cloud.
We’re hiring! Join our team and build the future of data orchestration. See open positions >
Alluxio enables compute
Data locality
Bring your data close to compute.
Make your data local to compute workloads for Spark caching, Presto caching, Hive caching and more.
Data Accessibility
Make your data accessible.
No matter if it sits on-prem or in the cloud, HDFS or S3, make your files and objects accessible in many different ways.
Data On-Demand
Make your data as elastic as compute.
Effortlessly orchestrate your data for compute in any cloud, even if data is spread across multiple clouds.
“zero-copy” burst user spotlight: walmart
Why Walmart chose Alluxio’s “Zero-Copy” burst solution:
- No requirement to persist data into the cloud
- Improved query performance and no network hops on recurrent queries
- Lower costs without the need for creating data copies
See more on how Alluxio powers Walmart’s “zero-copy” burst solution in their presentation >
Featured Use Cases and Deployments
Managing data copies/app changes when bursting compute to cloud?
Zero-copy hybrid bursting with no app changes to intelligently make remote data accessible in the public cloud.
Expanding compute capacity across geo-distributed data centers?
Zero-copy bursting across data centers for Presto, Spark, and Hive with no app changes on data stored in HDFS.
Interact with Alluxio in any stack
Pick a compute. Pick a storage. Alluxio just works.
-- Pointing Table location to Alluxio
CREATE SCHEMA hive.web
WITH (location = 'alluxio://master:port/my-table/‘)
// Using Alluxio as input and output for RDD
scala> sc.textFile("alluxio://master:19998/Input")
scala> rdd.saveAsTextFile("alluxio://master:19998/Output")
// Using Alluxio as input and output for Dataframe
scala> df = sqlContext.read.parquet("alluxio://master:19998/Input.parquet")
scala> df.write.parquet("alluxio://master:19998/Output.parquet”)
-- Pointing Table location to Alluxio
hive> CREATE TABLE u_user (
userid INT,
age INT)
ROW FORMAT DELIMITED
FIELDS TERMINATED BY '|'
LOCATION 'alluxio://master:port/table_data';
Create and Query table stored in Alluxio
hbase(main):001:0> create 'test', 'cf'
hbase(main):002:0> list ‘test'
# Accessing Alluxio after mounting Alluxio service to local file system
$ ls /mnt/alluxio_mount
$ cat /mnt/alluxio_mount/mydata.txt
ALLUXIO
$ ./bin/alluxio fs mount \
--option aws.accessKeyId=<AWS_ACCESS_KEY_ID> \
--option aws.secretKey=<AWS_SECRET_KEY_ID> \
alluxio://master:port/s3 s3a://<S3_BUCKET>/<S3_DIRECTORY>
$ ./bin/alluxio fs mount \
alluxio://master:port/hdfs hdfs://namenode:port/dir/
$ ./bin/alluxio fs mount \
--option
fs.azure.account.key.<AZURE_ACCOUNT>.blob.core.windows.net=<AZURE_ACCESS_KEY> \
alluxio://master:port/azure
wasb://<AZURE_CONTAINER>@<AZURE_ACCOUNT>.blob.core.windows.net/<AZURE_DIRECTORY>/
$ ./bin/alluxio fs mount \
--option fs.gcs.accessKeyId=<GCS_ACCESS_KEY_ID> \
--option fs.gcs.secretAccessKey=<GCS_SECRET_ACCESS_KEY> \
alluxio://master:port/gcs gs://<GCS_BUCKET>/<GCS_DIRECTORY>
$ ./bin/alluxio fs mount \
--option aws.accessKeyId=<AWS_ACCESS_KEY_ID> \
--option aws.secretKey=<AWS_SECRET_KEY_ID> \
--option alluxio.underfs.s3.endpoint=http://<rgw-hostname>:<rgw-port> \
--option alluxio.underfs.s3.disable.dns.buckets=true \
alluxio://master:port/ceph s3a://<S3_BUCKET>/<S3_DIRECTORY>
$ ./bin/alluxio fs mount alluxio://master:port/nfs /mnt/nfs
powered by alluxio
What’s Happening
In this blog, we provide an overview of Alluxio’s AI/ML model training solution. For more details about the reference architecture and benchmarking results, please refer to the full length whitepaper.
Running inference at scale is challenging. In this blog, we will share our observations and the practice to use Alluxio to speed up the I/O performance for large-scale ML/DL offline inference at Microsoft Bing.
Alluxio has raised $50 million in a Series C round of funding, capital the company will use to fuel the growth of its global operations and continue building out the capabilities of its data orchestration software for managing large-scale distributed data workloads.
Intel and Alluxio collaborate to measure a 20-25% price/performance improvement over the prior generation for machine learning models with PyTorch on AWS. This collaboration … Continued
Alluxio started as a virtual distributed file system, a research project out of the AMPLab at U.C. Berkeley. Alluxio foresaw the need for agility … Continued