- CEO and co-founder at Minio Inc.
- Board member and angel investor at H2O.ai
- Technical advisory board at Storj Labs Inc.
- Angel investor at Gitter.im, Procurify, Treasure Data Inc., Helpshift
- Venture advisor at Nexus Venture Partners
- Advisor at Postman
- Member board of advisors at Fastor Systems
- Board member at Free Software Foundation of India
- Charter Member at TiE
- Office of the CTO at Red Hat
- CTO and co-founder at Gluster Inc. (acquired in 2011 by Red Hat)
- Angel investor at Push Computing Inc.
- CTO and VP engineering at California Digital Corporation
- Software engineer at Newgen Software Technologies
StorageNewsletter: Could you refresh about the root and genesis of the company, with background of founders? And how many people do you employ today?
AB Periasamy: In 2015, Garima [Kapoor], Harsha [Harshavardhana], and I launched Minio to address the explosive growth in data storage market.
Enterprise users are looking for an Amazon S3 compatible alternative to replace their NAS and Hadoop HDFS infrastructure. Public cloud is expensive for petascale storage needs and often vulnerable to data breaches. Three years later, it looks like we chose the right path. The enterprise market is hot again.
Prior to Minio, I co-founded Gluster in 2005 to build a petascale distributed file system. Red Hat acquired Gluster in 2011 and has done a fantastic job advancing it.
Minio team has grown to 30 members now. We also get a lot of help from our community of ~300 code contributors and ~3,000 slack members.
Minio has raised so far $23 million, it seems that you have a frugal approach regarding spending?
Venture capital money should always be spent frugally.
With Gluster and Ceph, Minio is a reference in open source storage, how do you see the object storage category and industry direction?
Object storage has emerged as the primary storage technology for cloud-native applications. Unlike file and block, object storage allows applications to securely access data from across the Internet with a simple REST API. It has powerful features like encryption, erasure-code, SQL-SELECT and lambda computing that are relevant for today’s needs. Object storage wisely traded compatibility to legacy infrastructure for simplicity and scale.
Minio’s use case is growing as horizontally as the cloud itself. I would like to wear blinders and stay focused on a few vertical use cases though. Most opportunity is in displacing enterprise NAS and Hadoop HDFS for streaming data workloads, database snapshots, video streaming and medical imaging.
How do you measure the adoption of Minio? What about US vs. Europe vs. APAC? What’s the total capacity you have under management?
We look for growth trends around code contribution, community participation, and adoption stats. US, Europe and Japan rank amongst the top three bases for Minio’s deployments. Adoption is continuing to grow strong with 100 million docker pulls, 300 code contributors and 3,000 community members on slack channel approximately. We have no idea how much data is actively stored on Minio. We cannot collect this data without user consent.
We hear a lot of things about software-defined storage (SDS). What is your definition?
In the early days of Gluster, it was hard for people to conceive the idea of downloadable storage software that operated at petascale. Operating system architects ridiculed user-space file systems as toys. A decade later, storage systems are today written in Go and interfaced with RESTful APIs. The industry has come a long way.
SDS systems remove the need for inflexible and specialized hardware by moving the functionality to software layer. Literal definition aside, SDS as a category seems to have lost its identity. Every storage appliance vendor has an SDS offering today. The mindset of a hardware team is very different from a software team. Unless they rewrote the product from scratch as a software-only offering with an entirely new engineering team, I would be skeptical about its SDS capability. Then you would also need software sales people, software marketing people and so on.
What are the advantages of your object storage solution? What are the unique technologies you built and embed?
Minio’s advantages are high-performance, simplicity, and cloud-native.
- High performance: Minio is positioned to take full advantage of solid stage storage innovations. 1PB in 1U is going to be a reality in 2019. Minio can saturate Mellanox 100GbE interconnects with Intel NVMe SSDs. Minio is fast because of its light-weight architecture and assembly language optimizations in erasure code, encryption and hashing with AVX512, AVX2 and SSE SIMD instructions. High performance is crucial for big data and AI/ML workloads.
- Simplicity without compromise: Simplicity is the foundation for scalability. Minio differentiates itself from other unified file+block+object storage systems by focusing only on object storage. Running a Minio service is no more complicated than a distributed set of web servers. It is strictly consistent and light-weight. You can pull the plug under heavy load and restart without corruption. Minio server does not require a meta-data store because each of the objects are individually erasure-coded. Other object storage systems use Cassandra or home-grown distributed key-value stores to manage their meta-data in an eventual consistency model. All meta-data servers are prone to corruption and complexity at scale.
Within a 23MB static binary, Minio packs all the features necessary to operate a large scale private cloud storage service.
- Cloud-native: Minio is designed for cloud-native architecture from the ground up. Operators manage storage infrastructure with the same set of Kubernetes commands and APIs as compute infrastructure. Applications make a claim to the orchestrator and a new set of Minio server instances are provisioned instantly for each tenant, isolated from one another, and with the same durability as you’d get if you ran Minio as a binary on bare metal or VMs.
Why did you choose the GO language?
Go is practical and the language of choice for modern infrastructure projects. It has the right proportions of C’s performance and Python’s rapid prototyping. Haskell is a great choice, but we could not have grown this prevalent in the cloud-native community.
In 2017 many companies announced a multi-cloud strategy and it means different things for different people, what is your definition and how do you deliver this approach?
Customers want the ability to adopt a mix of public cloud, private cloud, and legacy infrastructure without requiring application rewrites. Multi-cloud is about portability of applications and standardization of APIs. Multi-cloud is not to be confused with unified global namespace across multiple clouds.
Minio has storage plugins for NAS, SAN and public cloud storage services to appear as Amazon S3 compatible storage. Plugins include NFS, vSAN, Azure Blob, Google Cloud Storage, Amazon S3, Pivotal Cloud Foundry, Backblaze B2, Alibaba COSS, Joyent Manta, Sia and a dozen private forks.
The product is open source but you generate revenue I think, what do you sell in fact? And do you do it?
Yes, we do. Enterprise customers prefer commercial open source products compared to its proprietary alternatives because they receive more value in return. Customers buy a Minio One annual subscription based on their capacity utilization.
What is your business model? Any OEM or strategic alliances? What about channel partners?
The business model is similar to that of Red Hat. We have been working on tighter integration with selected partners for a while. Expect a series of strategic partnership announcements soon. Minio for Pivotal Cloud Foundry and Minio for Azure are live already.
What about your international presence?
We are taking international customers, though we do not have a local presence outside of the US. Europe and Japan will be our next target.
How did you finish 2017?
For the first three years, we kept our focus on product maturity and market adoption. 2017 marked the end of phase one for Minio. The highlight of 2017 was the rate of growth of adoption. Docker pulls grew by 100x.
What do you plan to add in the product in 2018?
We are working on features for the enterprise market like flash optimization, compression, single-sign-on and access management.
Another area of focus is to optimize Minio for streaming data workloads. AI/ML and analytic databases will be able to query 5TB compressed objects in CSV, JSON and Parquet formats with high-speed SQL SELECT statements.
Minio Secures $20 Million Series A
And launches multi-cloud object storage solution.
2017.09.21 | Press Release
Start-Up Profile: Minio, in Open Source Object Storage
For developers to build their own cloud storage, just raising $3.3 million seed funding
by Jean Jacques Maleval | 2015.06.25 | News