097: GreyBeards talk open source S3 object store with AB Periasamy, CEO MinIO

Ray was at SFD19 a few weeks ago and the last session of the week (usually dead) was with MinIO and they just blew us away (see videos of MinIO’s session here). Ray thought Anand Babu (AB) Periasamy (@ABPeriasamy), CEO MinIO, who was the main presenter at the session, would be a great invite for our GreyBeards podcast. Keith and I had a ball talking with AB.

Why object store

There’s something afoot in object storage space over the last year or so. It seems everybody is looking to deploy object store whether that be on prem, in CoLo facilities and in the cloud. It could be just the mass of data coming online but that trend has remained the same for years no. No it’s something else.

It all starts with AWS and S3. Over the last couple of years AWS has been rolling out new functionality that only works with S3 and this has been driving even more adoption of S3 as well as other object storage solutions.

S3 compatible object stores are available in just about every cloud service, available from major (and minor) storage vendors and in open source from MinIO.

Why S3 is so popular

Because object store is accessed via RestFUL interfaces, traditionally most implementations used their own API to access it. But when AWS created S3 (simple storage service) with their own API/SDK to access it, it somehow became the de-facto standard interface for all other object stores. S3 compatibility became a significant feature that all object stores had to support.

Sometime after that MinIO came into existence. MinIO provides a 100% open source, fully AWS S3 compatible object store that you can run anywhere on prem, in CoLo facilities and indeed in the cloud. In fact, there exist customers that run MinIO in AWS AB says this is probably just customers using a packaged software solution which happens to include MinIO but it’s nonetheless more expensive than AWS S3 as it uses EC2 instances and EBS storage to create an object store

Customers can access MinIO object stores with the AWS S3 SDK or the MinIO SDK. and you can access AWS S3 storage with AWS S3 SDK or use MinIO SDK. Occosionally, AWS S3 updates have broken MinIO’s SDK but these have been later fixed by AWS. It seems AWS and MinIO are on good terms.

AB mentioned that as customers get up to a few PBs of AWS S3 storage they often find the costs to be too high. It’s at this point that they start looking at other object storage solutions. But because MinIO is 100% S3 compatible and it’s open source many of these customers deploy it in their own data center facilities or in colo environments.

For those customers that want it, MinIO also offers an S3 gateway. With the gateway on prem customers can use S3 or standard file services to access S3 object storage located in the cloud. The gateway also works in the public cloud and can support both AWS s3 as well as Microsoft Blob storage as a backend.

MinIO matches AWS S3 features

AWS S3 has a number of great features and MinIO has matched or exceeded them all, step by step. AWS S3 has cross region replication options where customers can replicate S3 data from one region to another. MinIO supports both asynchronous replication of S3 data and synchronous replication (using RADIO).

But MinIO adds support for erasure coding within a fault domain. Default is Nx2 erasure coding which duplicates all your data so as long as 1/2 of your servers and storage are available you continue to have access to all your data. But this can be configured down like 12+4 where data is split accross 16 servers any four of which can fail and you can still access data.

AWS customers can use a Snowball (standalone storage device) to transfer data to or from S3 storage. AWS Snowball implements a subset of S3 API and requires a NAS staging area of equivalent size to migrate data out of S3. MinIO has support for Snowball’s limited S3 API and as such, Snowball’s can be used to migrate data into or out of MinIO. MinIO has a blog post which describes their support for AWS Snowball.

AWS also offers S3 Lambda services or server less computing services where compute services can be invoked when data is loaded in a bucket and then turned off when no longer needed. AWS Lambda depends on AWS messaging and other services to work properly. But MinIO supports Lambda like functionality using other open source services. AB mentions MQTT and Kafka services. MinIO has another blog post discussing their Lambda like services based on Kafka.

AWS recently implemented Snowflake a SQL database server for unstructured data that uses S3 storage to hold data. Ray and Keith almost choked on that statement as unstructured data and databases never used to be uttered in the same breath. But what AWS has shown was that you can use object store for database data as long as you are willing to load the table into memory and process it there and then unload any modified table data back into the object store. Indexing of the object data seems to be done as the data is being loaded and is also being done in a (random IO) cache or in memory and once done can also be unloaded into the object store.

Now Snowflake uses S3 but it’s not available on prem. MinIO has a number of data base partners that make use of their object store as a backend to host a Snowflake like service onprem. AB mentioned Spark and Splunk but there are others as well.

We ended up the discussion with what does it mean to have 20K stars on GitHub. AB said if you did a java script getting 20K stars would be easy but you just don’t see this sort of open source popularity for storage systems. He said the number is interesting but the growth rate is even more interesting.

The podcast runs ~47 minutes. AB was a great to talk tech with. Keith and I could have talked all afternoon with AB. It was very hard to stop the recording as we could have talked with him for another hour or more. AB said he doesn’t like to do podcasts or videos but he had no problem with us firing away questions. Listen to the podcast to learn more.

This image has an empty alt attribute; its file name is Subscribe_on_iTunes_Badge_US-UK_110x40_0824.png
This image has an empty alt attribute; its file name is play_prism_hlock_2x-300x64.png

Anand Babu Periasamy, CEO MinIO

AB Periasamy is the CEO and co-founder of MinIO. One of the leading thinkers and technologists in the open source software movement, AB was a co-founder and CTO of GlusterFS which was acquired by RedHat in 2011. Following the acquisition, he served in the office of the CTO at RedHat prior to founding MinIO in late 2015. AB is an active angel investor and serves on the board of H2O.ai and the Free Software Foundation of India.

He earned his BE in Computer Science and Engineering from Annamalai University.

094: GreyBeards talk shedding light on data with Scott Baker, Dir. Content & Data Intelligence at Hitachi Vantara

Sponsored By:

At Hitachi NEXT 2019 Conference, last month, there was a lot of talk about new data services from Hitachi. Keith and I thought it would be a good time to sit down and talk with Scott Baker (@Kraken-Scuba), Director of Content and Data Intelligence, at Hitachi Vantara about what’s going on with data operations these days and how customers are shedding more light on their data.

Information supply chain

Something Scott said in his opening remarks caught my attention when he mentioned customer information supply chains. The information supply chain is similar to manufacturing supply chains, but it’s all about data. Just like manufacturing supply chains where parts and services come from anywhere and are used to create products/services for customers,

information supply chains are about the data used in their organization operations. Information supply chain data is A) being sourced from many places (or applications); B) being added to by supply chain processing (or other applications); and C) ultimately used by the organization to supply a product/service to customers.

But after the product/service is supplied the similarity between manufacturing and information supply chains breaks down. With the information supply chain, data is effectively indestructible, is infinitely re-useable and can live forever. Who throws data away anymore?

The problem most organizations have with information supply chains is once the product/service is supplied, data is often put away never to be seen again or as Scott puts it, goes dark.

This is where Hitachi Content intelligence (HCI) comes in. HCI is designed to take (unstructured or structured) data and analyze it (using natural language and other processing tools) to surround it with information and other metadata, so that it can become more visible and useful to the organization for the life of its existence.

Customers can also use HCI to extract and blend data streams together, automating the creation of an information rich, data repository. The data repository can readily be searched to re-discover or uncover attributes about the data not visible before.

Scott also mentioned the Hitachi Pentaho Platform which can be used to make real time decision from structured data. Pentaho information can also be fed into HCI to provide more intelligence for your structured data.

But HCI can also be used to analyze other database data as well. For instance, database blob and text elements can be fed to and analyzed by HCI. HCI analysis can include natural language processing and other functionality to tag the data by adding key:value information, all of which can be supplied back to the database or Pentaho to add further value to structured data.

Customers can also use HCI to read and transform database tables into XML files. XML files can be stored in object stores as objects or in file systems. XML data could easily be textually indexed and be searched by various tools to better understand the structured data information

We also talked about Hadoop data that can be offloaded to Hitachi Content Platform (HCP) object storage with a stub left behind. Once data is in HCP, HCI can be triggered to index and add more metadata, which can then later be used to decide when to move data back to Hadoop for further analysis.

Finally, Keith mentioned that he just got back from KubeCon and there was an increasing cry for data being used with containerized applications. Scott mentioned HCP for Cloud Scale, the newest member of the HCP object store family, focused on scale out capabilities to provide highly consistent, object storage performance for customers that need it. Customers running containerized workloads use scale-out capabilities to respond to user demand and now they have on premises object storage that can scale with them, as needs change.

The podcast ran ~24 minutes. Scott was very knowledgeable about data workflows, pipelines and the need for better discovery tools. We had a great time discussing information supply chains and how Hitachi can help customers optimize their data pipelines. Listen to the podcast to learn more.

This image has an empty alt attribute; its file name is Subscribe_on_iTunes_Badge_US-UK_110x40_0824.png
This image has an empty alt attribute; its file name is play_prism_hlock_2x-300x64.png

Scott Baker, Director of Content and Data Intelligence at Hitachi Vantara

Scott Baker is, and has been, an active member of the information technology, data analytics, data management, and data protection disciplines for longer than he is willing to admit.

In his present role at Hitachi, Scott is the Senior Director of the Content and Data Intelligence organization focused on Hitachi’s Digital Transformation, Data Management, Data Governance, Data Mobility, Data Protection and Data Analytics solutions which includes Hitachi Content Platform (HCP), HCP Anywhere, HCP Gateway, Hitachi Content Intelligence, and Hitachi Data Protection Solutions.

Scott is a VMware Certified Professional, recognized as a subject matter expert, industry speaker, and author. Scott has been a panelist on topics related to storage, cloud, information governance, data security, infrastructure standardization, and social media topics. His educational background includes an MBA, Master’s & Bachelor’s in Computer Science.

When he’s not working, Scott is an avid scuba diver, underwater photographer, and PADI Scuba Instructor. He has a passion for public speaking, whiteboarding, teaching, and traveling the world.

93: GreyBeards talk HPC storage with Larry Jones, Dir. Storage Prod. Mngmt. and Mark Wiertalla, Dir. Storage Prod. Mkt., at Cray, an HPE Enterprise Company

Supercomputing Conference 2019 (SC19) is coming to Denver next week and in anticipation of that show, we thought it would be a good to talk with some HPC storage group. We contacted HPE and given their recent acquisition of Cray, they offered up Larry and Mark to talk about their new ClusterStor E1000 storage system.

There are a number of components that go into Cray supercomputers and besides the ClusterStor, Larry and Mark mentioned their new SlingShot cluster interconnect which is Ethernet based with significant enhancements to congestion handling. But the call focused on ClusterStor.

What is ClusterStor

ClusterStor, is a Lustre file system hardwareappliance. Lustre has always been popular with the HPC crowd as it offered high bandwidth file services. But Lustre often took a team of (PhD) scientists to configure, deploy and run properly because of all the parameters that had to be setup for optimum performance.

Cray’s ClusterStor was designed to make configuring, deploying and running Lustre a lot simpler with a GUI and system defaults that provided an optimal running environment. But if customers still want access to all Lustre features and functionality, all the Lustre parameters can still be tweaked to personalize it.

What sort of appliance

The ClusterStore team has created a Lustre storage appliance using two systems, a 2U-24 NVMe SSD system and a 4U-106 disk drive system. Both systems use PCIe Gen 4 buses which offer 2X the bandwidth of Gen 3 and NVMe Gen 4 SSDs. Each ClusterStore E1000 appliance comes with 2 servers for HA and the storage behind it.

Larry said the 2U NVMe Gen 4 appliance offers 80GB/sec of read and 60GB/sec of write data bandwidth. And a full rack of these, could support ~2.5TB/sec of data bandwidth. One TB/sec seems like an awful lot to the GreyBeards, 2.5TB/sec, out of this world.

We asked if it supported InfiniBAND interconnects? Yes, they said it supports the latest generation of InfiniBAND but it also offers Cray’s own (SlingShot) Ethernet interconnect, unusual for HPC environments. And as in any Lustre parallel file system, servers accessing storage use Lustre client software.

ClusterStor Data Services

But on the backend, where normally one would see only LDISKFS for backend storage, ClusterStor also offers ZFS. Larry and Mark said that LDISKFS is faster but ZFS offers more functionality like snapshots and data compression.

Many of the Top 100 & Top 500 supercomputing environments are starting to deploy ML DL (machine learning-deep learning) workloads along with their normal HPC activities. But whereas HPC work has historically depended on bandwidth to read, write and move large files around, ML DL deals with small files and needs high IOPS. ClusterStor was designed to satisfy both high bandwidth and high IOPS workloads.

In previous HPC Lustre flash solutions, customers had to deal with the complexity of where to place data, such as on flash or on disk. But with net ClusterStor E1000, the system can do all this for you. That is it will move data from disk to flash when it sees an advantage to doing so and move it back again when that advantage is gone. But, just as with Lustre configuration parameters above, customers can still pre-stage data to flash.

The other challenge for HPC environments is extreme size. Cray and others are starting to see requirements for Exascale (exabyte, 10**18) byte) storage systems. In fact, Cray has a couple of ClusterStor E1000 configurations of 400PB or more already, As these systems age they may indeed grow to exceed an exabyte.

With an exabyte of data, systems need to support billions of files/inodes and better metadata services and indexing. ClusterStor offers optimized inode indexing and search to enable HPC users to quickly find the data they are looking for. Further, ClusterStor offers, data at rest encryption and supports virtual file systems, for multi-tenancy.

With a ZFS backend, ClusterStor can supply data compression and snapshots. Cray has tested ZFS compression on HPC scientific ( mostly already application compressed) data and still see ~30% reduction is storage footprint. At an exabyte of storage 30% can be a significant cost reduction

The podcast ran long, ~46 minutes. Larry and Mark had a good knowledge of the HPC storage space and were easy to talk with. Matt’s an old ZFS hand, so wanted to take even more about ZFS. I had a good time discussing ClusterStor and Lustre features/functionalit and how the HPC workloads are changing. Listen to the podcast to learn more. [The podcast was recorded on November 6th, not the 5th as mentioned in the lead in, Ed.]

This image has an empty alt attribute; its file name is Subscribe_on_iTunes_Badge_US-UK_110x40_0824.png
This image has an empty alt attribute; its file name is play_prism_hlock_2x-300x64.png

Larry Jones, Director Storage Product Management

Larry Jones is a director of storage product management for Cray, a Hewlett Packard Enterprise company.

Jones previously held senior product management roles at Seagate, DDN and Panasas.

Mark Wiertalla, Director Storage Product Marketing

Mark Wiertalla is a product marketing director for Cray, a Hewlett Packard Enterprise company.

Prior to Cray, Wiertalla held product manager roles at EMC and SGI.

90: GreyBeards talk K8s containers storage with Michael Ferranti, VP Product Marketing, Portworx

At VMworld2019 USA there was a lot of talk about integrating Kubernetes (K8s) into vSphere’s execution stack and operational model. We had heard that Portworx was a leader in K8s storage services or persistent volume support and thought it might be instructive to hear from Michael Ferranti (@ferrantiM), VP of Product Marketing at Portworx about just what they do for K8s container apps and their need for state information.

Early on Michael worked for RackSpace in their SaaS team and over time saw how developers and system engineers just loved container apps. But they had great difficulty using them for mission critical applications and containers of the time had a complete lack of support for storage. Michael joined Portworx to help address these and other limitations in using containers for mission critical workloads.

Portworx is essentially a SAN, specifically designed for containers. It’s a software defined storage system that creates a cluster of storage nodes across K8s clusters and provides standard storage services on a container level granularity.

As a software defined storage system, Portworx is right in the middle of the data path, storage they must provide high availability, RAID protection and other standard storage system capabilities. But we talked only a little about basic storage functionality on the podcast.

Portworx was designed from the start to work for containers, so it can easily handle provisioning and de-provisioning, 100s to 1000s of volumes without breaking a sweat. Not many storage systems, software defined or not, can handle this level of operations and not impact storage services.

Portworx supports both synchronous and asynchronous (snapshot based) replication solutions. As all synchronous replication, system write performance is dependent on how far apart the storage nodes are, but it can provide RPO=0 (recovery point objective) for mission critical container applications.

Portworx takes this another step beyond just data replication. They also replicate container configuration (YAML) files. We’re no experts but YAML files contain an encapsulation of everything needed to understand how to run containers and container apps in a K8s cluster. When one combines replicated container YAML files, replicated persistent volume data AND an appropriate external registry, one can start running your mission critical container apps at a disaster site in minutes.

Their asynchronous replication for container data and configuration files, uses Portworx snapshots , which are sent to an alternate site. But they also support asynch replication to any S3 compatible storage via CloudSnap.

Portworx also supports KubeMotion, which replicates/copies name spaces, container app volume data and container configuration YAML files from one K8s cluster to another. This way customers can move their K8s namespaces and container apps to any other Portworx K8s cluster site. This works across on prem K8s clusters, cloud K8s clusters, between public cloud provider K8s clusters s or between on prem and cloud K8s clusters.

Michael also mentioned that data at rest encryption, for Portworx, is merely a tick box on a storage class specification in the container’s YAML file. They make use use of KMIP services to provide customer generated keys for encryption.

This is all offered as part of their Data Security/Disaster Recovery (DSDR) service. that supports any K8s cluster service whether they be AWS, Azure, GCP, OpenShift, bare metal, or VMware vSphere running K8s VMs.

Like any software defined storage system, customers needing more performance can add nodes to the Portworx (and K8s) cluster or more/faster storage to speed up IO

It appears they have most if not all the standard storage system capabilities covered but their main differentiator, besides container app DR, is that they support volumes on a container by container basis. Unlike other storage systems that tend to use a VM or higher level of granularity to contain container state information, with Portworx, each persistent volume in use by a container is mapped to a provisioned volume.

Michael said their focus from the start was to provide high performing, resilient and secure storage for container apps. They ended up with a K8s native storage and backup/DR solution to support mission critical container apps running at scale. Licensing for Portworx is on a per host (K8s node basis).

The podcast ran long, ~48 minutes. Michael was easy to talk with, knew K8s and their technology/market very well. Matt and I had a good time discussing K8s and Portworx’s unique features made for K8s container apps. Listen to the podcast to learn more.

This image has an empty alt attribute; its file name is Subscribe_on_iTunes_Badge_US-UK_110x40_0824.png
This image has an empty alt attribute; its file name is play_prism_hlock_2x-300x64.png

Michael Ferranti, VP of Product Marketing, Portworx

Michael (@ferrantiM) is VP of Product Marketing at Portworx, where he is responsible for communicating the value of containerization and digital transformation to global architects and CIOs.

Prior to joining Portworx, Michael was VP of Marketing at ClusterHQ, an early leader in the container storage market and spent five years at Rackspace in a variety of product and marketing roles

85: GreyBeards talk NVMe NAS with Howard Marks, Technologist Extraordinary and Plenipotentiary, VAST Data Inc.

As most of you know, Howard Marks was a founding co-Host of the GreyBeards-On- Storage podcast and has since joined with VAST Data, an NVMe file and object storage vendor headquartered in NY with R&D out of Israel. We first met with VAST at StorageFieldDay18 (SFD18, video presentation). Howard announced his employment at that event. VAST was a bit circumspect at their SFD18 session but Howard seems to be more talkative, so on the podcast we learn a lot more about their solution.

VAST Data is essentially an NFS-S3 object store, scale out solution with both stateless, VAST Data storage servers and JBoF drive enclosures with Optane and NVMe QLC SSDs. Storage servers or JBoFs can be scaled independently. They don’t support tiering or DRAM caching of data but instead seem to use the Optane SSDs as a write buffer for the QLC SSDs.

At the SFD18 event their spokesperson said that they were going to kill off disk storage media. (Ed’s note: Disk shipments fell 18% y/y in 1Q 2019, with enterprise disk shipments at 11.5M units, desktop at 24.5M units and laptops at 37M units).

The hardware

The VAST Data storage servers are in a 2U/4 server configuration, that runs interface protocols (NFS & S3), data reduction (see below), data reformating/buffering etc. They are stateless servers with all the metadata and other control state maintained on JBoF Optane drives.

Each drive enclosure JBoF has 12 Optane SSDs and 44 U.2 QLC (no DRAM/no super cap) SSDs. This means there are no write buffers on the QLC SSDs that can lose data when power failures occur. The interface to the JBoF is NVMeoF, either RDMA-RoCE Ethernet or InfiniBand (customer selected). Their JBoFs have high availability, with dual fabric modules that support 2-100Gbps Ethernet/InfiniBand ports per module, 4 per JBoF.

Minimum starting capacity is 500TB and they claim support up to Exabytes. Although how much has actually been tested is an open question. They also support billions of objects/files.

Guaranteed better data reduction

They have a rather unique, multi-level, data reduction scheme. At the start, data is chunked in variable length chunks. They use heuristics to determine the chunk size that fits best. (Ed note, unclear which is first in this sequence below so presented in (our view of) logical order)

  • 1st level computes a similarity hash (56 bit not SHA1), which is used to determine a similarity level with any other currently stored data chunk in the system.
  • 2nd level uses a ZSTD compression algorithm. If a similarity is found, the new data chunk is compressed with the ZSTD compression algorithm and a reference dictionary used by the earlier, similar data chunk. If no existing chunk is similar to this one, the algorithm identifies a semi-unique reference dictionary that optimizes the compression of this data chunk. This semi-unique dictionary is stored as metadata.
  • 3rd level, If it turns out to be a complete duplicate data chunk, then the dedupe count for the original data chunk is incremented, a pointer is saved to the original unique data and the data discarded. If not a complete duplicate of other data, the system computes a delta from the closest “similar’ block and stores just the delta bytes, includes a pointer to the original similar block and increments a delta block counter.

So data is chunked, compressed with a optimized dictionary, be delta-diffed or deduped. All data reduction is done post data write (after the client is ACKed), and presumably, re-hydrated after being read from SSD media. VAST Data guarantees better data reduction for your stored data than any other storage solution.

New data protection

They also supply a unique Locally Decodable Erasure Coding with 4 parity (-like) blocks and anywhere from 36 (single enclosure leaving 4 spare u.2 SSDs) to 150 data blocks per stripe all of which support up to 4 device failures per stripe. 

The locally decodable erasure coding scheme allows for rebuilds without having to read all remaining data blocks in a stripe. In this scheme, once you read the 4 parity (-like) blocks, one has all the information calculated from up to ¾ of the remaining drives in the stripe, so the system only has to read the remaining ¼ drives in the stripe to reconstruct one, two, three, or four failing drives.  Given their data stripe width, this cuts down on the amount of data needing to be read considerably. Still with 150 data drives in a stripe, the system still has to read 38 drives worth of QLC SSD data to rebuild a data drive.

In addition to all the above, VAST Data also reblocks the data into much larger segments, (it writes 1MB segments to the QLC drives) and uses a heat map along with other heuristics to separate actively written data from less actively written data, thus reducing garbage collection, write amplification.

The podcast is a long and runs over ~43 minutes. Howard has always been great to talk with and if anything, now being a vendor, has intensified this tendency. Listen to the podcast to learn more.

Howard Marks, Technologist Extraordinary and Plenipotentiary, VAST Data, Inc.

Howard Marks brings over forty years of experience as a technology architect for hire and Industry observer to his role as VAST Data’s Technologist Extraordinary and Plienopotentary. In this role, Howard demystifies VAST’s technologies for customers and customer requirements for VAST’s engineers.

Before joining VAST, Howard ran DeepStorage an industry test lab and analyst firm. An award-winning speaker, he has appeared at events on three continents including Comdex, Interop and VMworld.

Howard is the author of several books (all gratefully out of print) and hundreds of articles since Bill Machrone taught him journalism at PC Magazine in the 1980s.

Listeners may also remember that Howard was a founding co-Host of the Greybeards-on-Storage Podcast.