DRAFT

Overview

ROSS (short for Research Object Storage System) is a multi-protocol, web-based storage system, currently based on the Dell EMC Elastic Cloud Storage (ECS) system.  The primary 4.2 PB storage unit for ROSS resides on the UAMS campus.  In the coming months the UAMS unit will be connected with a 2 PB sister unit sited at the University of Arkansas in Fayetteville, to allow for a limited amount of off-site, replicated storage.  ROSS currently has 23 storage nodes, with 15 at UAMS and 8 at UA Fayetteville.   Future expansions could increase the number of storage nodes and capacity.

As the name implies, ROSS is an object store.  An object store really does not have a directory tree structure like conventional POSIX file systems do.  (Grace's storage systems are POSIX file systems.)  Instead, objects have names and live in buckets, essentially a one-level directory hierarchy.  People often name objects using what looks like a POSIX-style file path.  For example "directory1/directory2/directory3/directory4/filename" might be an object name.  The RESTful APIs used to access ROSS can do a prefix search, listing just the objects that start with a particular prefix. For example "directory1/directory2/" would pull back all the objects that start with that prefix.  So in a way one can mimic a POSIX file system employing an object naming convention that uses the equivalent of POSIX path names.  Unfortunately, prefix search is quite inefficient, making the simulated directory lookup slower than a native POSIX file system.  Nevertheless, some tools exist that can mimic a POSIX-style file system using an object store behind the scenes.  With a proper naming convention, it is fairly easy to backup and restore POSIX directory trees to and from an object store fairly efficiently.  This is what makes ROSS an excellent backing storage system for Grace.  (A backing storage system is the larger, but slower storage system that back ends a smaller, but faster storage cache in a hierarchical storage model.  Grace's storage system should be considered as cache (fast temp storage).)

Remember, the HPC Admins do not back up Grace's shared storage system, since it is only intended to be a staging or scratch area (aka a cache) used to run HPC jobs.  In other words, the primary copy of data should be elsewhere, not on Grace.  ROSS is one option for keeping the primary copy of data.  In fact, the main reason UAMS purchased ROSS is as the primary location for storing research data.

Data in ROSS starts as triple replicated on 3 different storage nodes, but then transforms to erasure coding giving resilience without major costs.  ROSS currently uses 12/16 error coding, meaning that for every 12 blocks of data stored it actually writes 16 blocks.  Up to four blocks could be damaged before ROSS is no longer able to recover data if a fifth block were damaged.  In contrast, most RAID systems only have a 1 or 2 drive redundancy.  ROSS scatters the 16 blocks across the storage nodes in a data center, improving the performance of data retrievals and further improving the resilience of the system (less chance that a node failure blocks access).  Data can be accessed from any of the storage nodes.  Currently the two systems (UAMS and UARK) are isolated from each other, not allowing replication.  But the plans to join the two are in progress.  Eventually all of the content in ROSS, regardless of which campus it physically lives on, would be accessible from either campus.  Of course, if data living on one campus is accessed from the other campus, the access will be somewhat slower due to networking delays and bandwidth limitations going between campuscampuses.

As mentioned, soon Grace ROSS will have an option for replicating data in Fayetteville.  However, even in this case I would not consider the copy in Fayetteville as a true backup copy.  Replication is good for maintaining data that needs high availability and equivalent performance regardless of which campus the data is accessed from.  Replication also doubles the storage cost since it reduces consumes available storage capacity at twice the rate that non-replicated storage does.  We still recommended that researchers keep backup or archive copies of data somewhere else, even if replication is turned on.

For data that meets OURRstore requirements (relatively static, STEM related, not clinical nor regulated), archiving to OURRstore is a very cost effective option to get resilient, long-term, offsite copies for little extra money (just media and shipping costs).  Other options for backup include the UAMS campus Research NAS or cloud storage such as One Drive, Box, Azure, Google, Amazon, or IBM.  USB or bare drives are also an option for backup, but not recommended, as they are quite error prone if not stored and managed properly.  Some departments might have departmental storage systems that could hold backup copies of data, though the caveat of proper management of those devices still applies.

Costs for using ROSS

The College of Medicine (COM) Associate Dean of Research, in consultation with an advisory group at UAMS, recommended charging an up-front payment of $70 per TB for the use of ROSS, good for the 5 year life of the storage.  This is less than half the current replacement plus maintenance cost for the underlying hardware ($159 per TB).  If one wants replicated storage (i.e. one copy in Little Rock, one copy in Fayetteville), the cost would be double, or $140 per TB for the pair of copies.

...

Restrictions on data that can be stored on ROSS

There are few restrictions on the types of data that can be stored on ROSS.  Almost anything is allowed, as long as the data storage complies with both governmental, IRB, and UAMS rules and regulations.

Keep in mind that any campuses that are participants in ARCC, and by extension, the Arkansas Research Platform (ARP), have access to ROSS.  Unlike the UAMS Research NAS, which is locked down behind UAMS firewalls hence only accessible inside the UAMS Campus, ROSS is located in the ARCC Science DMZ, a private network only accessible by a limited number of campuses, both within Arkansas and potentially beyond.  As such, it is inappropriate to store unecrypted in ROSS fully identified patient (PHI) or personal (PII) data, as it could be a violation of UAMS HIPAA or FERPA policies.  ROSS does have the ability to restrict access to buckets and to do server-side, data-at-rest encryption, but these capabilities have not been evaluated as to whether or not they are sufficient for HIPAA or FERPA compliance.  For now, ROSS should not be used for data that is regulated by HIPAA, FERPA, or any other governmental regulation.  De-identified human subject data is allowed.

...

The server-side encryption can either be turned on at the namespace level, where all buckets in a namespace are required to be encrypted, or on the bucket level for namespaces that do not have 'encryption required' turned on.  If you want namespace level encryption, please inform the HPC admins when requesting a namespace.  The encryption choice must be made at when the namespace or bucket creation time, created and cannot be changed afterwards.  (One can copy data from an unencrypted bucket to an encrypted one, then destroy the unencrypted bucket, or visa versa, should a change be needed after bucket creation.)

Preparing to Use ROSS 

If you are are good with the data restrictions and prepared to cover costs that might be incurred for using ROSS, there are steps that must be  be completed before you can actually move data between ROSS and Grace.  First, you must request or be assigned to a namespace.  Second, decide which APIs you might use. Third, your namespace administrator using the ECS Portal may wish to pre-create the buckets that you might use.  Creating buckets in the ECS Portal can be a more convenient than creating them using APIs or tools.  Finally,  get get access credentials for the APIs that you might use.

Requesting a namespace assignment on ROSS

All users of ARP facilities, including Grace, may ask for credentials in the "arp" namespace.  When the administrators create your credentials, they will also create a bucket in that namespace for your use.  The bucket will have the same name as your home directory on Grace or Pinnacle, which should be the same as your username.  After creating your account, the administrators will place your S3 secret key into the file named ".ross/s3key" in your home directory.  Your S3 ID is your ROSS object user name in the arp namespace, which is "<username>@arp", where <username> is your login name for Grace or Pinnacle.

If your lab, department, project, or group would like to purchase their own namespace for their exclusive use, please contact us at HPCAdmin@uams.edu.  All namespaces have namespace administrators namespaces have namespace administrators who manage buckets and object users within that namespace.  A particular username@domain can only be a namespace administrator for one namespace.  If a particular person needs to be the namespace administrator in more than one namespace, for example a personal namespace as well as a group namespace, than one namespace, for example namespaces for two different groups, they must use different login names and domainsplus domains for each namespace they wish to administer.  This is why we suggest using your HPC credentials for personal namespaces, and other credentials (e.g. UAMS AD) for group namespacesalways qualifying namespace users (object or administrative) with the "@<namespace>" suffix, where <namespace> is the name of the namespace that the user is assigned to.

Don't forget that a namespace administrator is not the same as an object user.  See Access to Terminology Used By the Research Object Store System (ROSS) for the difference between a namespace administrator and an object user.  An object user has a different API-specific set of credentials.

The steps to gain access to a namespace differ, depending on whether the namespace already exists or not. 

New namespaces

To initiate access to ROSS for a new personal or group (e.g. project, lab, department) namespace, please send a request via e-mail to to HPCAdmin@uams.edu.  In your request,

• Please indicate whether this is for a personal namespace (e.g. primary storage for processing data on Grace), or for a group (shared storage). 
  • For a personal namespace, please indicate
    • your name
    • your e-mail
    • your departmental affiliation
    • what your login name (not your password) and domain you will use to access ROSS's administrative interface, e.g. johndoe@hpc.uams.edu (for personal namespaces we prefer that you use your HPC usernameyou use your HPC username)
    • why you do not wish to be part of the "arp" namespace (where most personal accounts go)
  • For a group namespace, please give
    • a name and brief description for the group
    • the primary e-mail contact for the group
    • the departmental affiliation of the group
    • who will be the namespace administrators - we need
      • their names
      • their e-mail address
      • their login name (not their password) and domain e.g. janedoe@ad.uams.edu (for group namespaces we generally prefer campus Active Directory usernames (e.g. the name the namespace administrator might use to login to Outlook Mail))
      • you may ask for more than one namespace administrator
      • if all the members of a particular campus AD group should be namespace administrators, you could also just give us the name and domain of the group instead of their individual names
• Please estimate approximately how much storage you or your group intend to use in ROSS for the requested namepace, divided into local and replicated amounts.  Remember that replicated data costs twice as much as non-replicated data.  The HPC Admins will use this information in setting the initial quotas and for capacity planning.  You will be allowed to request increases in quota if needed and space is available. 
• We would also appreciate a brief description of what you will be using the storage for.  The "what it is used for" assists us in drumming up support (and possibly dollars) for expanding the system. 

Existing namespace

If you wish to access an existing group namespace as an object user, please contact the namespace administrator of that namespace and ask to be added as an object user for that namespace.  For the "arp" namespace, please contact "HPCAdmin@uams.edu".  If you need assistance determining what namespaces are available and who are the namespace administrators feel free to contact the HPC Admins via e-mail (hpcadmin@uams.edu).

If you supervised an employee or student who is no longer with you, for example due to a change of assignment or leaving the university, then you are responsible for the final disposition of the storage used by that person.  In this case you may request access to their personal namespace by sending that request to the HPC Admins via e-mail (hpcadmin@uams.edu).  If the person is still with UAMS, the HPC Admins will coordinate with the two of you, in case the person wishes keep their namespace, including portions of the existing data.

APIs supported by ROSS

ROSS supports multiple object protocol APIs.  The most common is S3, followed by Swift (from the Open Stack family).  Although ROSS can support the NFS protocol, our experiments show that the native NFS support is slower for many use cases than other options that simply emulate a POSIX file system using object protocols.  But NFS is available for those who would like to use it.  ROSS can also support HDFS (used in Hadoop clusters) and a couple of Dell/EMC proprietary protocols (ATMOS and CAS); however, we have not tested any of these other protocols, hence the HPC Admins could only offer limited support for these other options. 

...

The file-based protocols NFS and HDFS must be enabled on a bucket at creation time, if you intend to use them, since the underlying system adds additional metadata when file access is enabled.  When enabled, NFS and HDFS, too, can be used interchangeably with the S3 and Swift object protocols on the same bucket.  However, a bucket with file access enabled loses some of the object features, such as life cycle management.  So consider carefully before enabling file access.  (You can still access a bucket using file system semantics with some external tools even when the bucket is not enabled for file access.)

Creating a bucket

There are three ways to create buckets (aka containers in the Swift world):

1. Many object tools have options to create and manipulate buckets, which is quite convenient, generally portable (i.e. would work on any object storage system), but also limited.  In general, tools cannot create buckets with ECS options, such as encryption, replication and file access.  You have to use one of the other bucket creation options.  Object tools only need the object user's access credentials.  Please see the tools' documentation for details.
2. The RESTful APIs, including S3, Swift, and an ECS-specific management API, can be used to create and manipulate buckets.  With the appropriate headers and parameters, ECS options can be enabled.  More information can be had in the ECS Data Access Guide and the ECS API Reference (hint - use the search function). 
   1. The protocol-specific S3 or Swift REST APIs need an object user's credentials, and the namespace administrator must have given appropriate permissions to the object user.
   2. The ECS-specific management APIs (which we do not support users in using) require namespace administrator credentials.
3. Use the ECS Portal.  This is the simplest option that gives full control over bucket characteristics.  The ECS Portal can only be accessed by the namespace administrator.  (That would be you, for your personal namespace.)  The namespace administrator logs into the ECS Portal (https://ross.hpc.uams.edu) with the credentials tied to that namespace.  Details on how to use the ECS Portal to manage buckets can be found in the Buckets chapter of the ECS Administration Guide.

Getting access credentials

In order to use either S3 or Swift, you need credentials. 

...

For Swift credentials, Note the "Swift Group" and "Swift Password" fields.  If this is a new object users, you may need to fill in the group and password for Swift.  Note that we do not currently run a keystone server, hence only Swift v1 or v2 credentials work.

Using ROSS

After getting access credentials for your chosen APIs, you are ready to use ROSS.  

...

In theory, a bucket can hold millions of files.  We have noticed, though, that with path prefix searching the more objects that have a particular prefix, the longer the search takes.  While this is not dissimilar to POSIX filesystems, where the more files there are in a directory. the longer it takes to do a directory lookup, on an object store, being a flat namespace (no directory hierarchy), such searches can be much slower than the typical POSIX file system.  The total number of objects in a bucket also has a mild, though not dramatic, impact on the look up speed particular objects in a bucket.  Again, keep this in mind when planning your use of ROSS.  One way to get around this speed penalty is to use an external database (e.g. sqlite) to track in which buckets and objects your data resides in.

Tools recommendations for accessing ROSS

The bottom line - any tool that uses any of the supported protocols theoretically could work with ROSS.  There are a number of free and commercial tools that work nicely with object stores like ROSS.  Like most things in life, there are pros and cons to different tools.  Here is some guidance as to what we look for in tools.

...

The following sections describe using ecs-sync and rclone for moving data between ROSS and Grace's cluster storage.

ecs-sync

The ecs-sync program is specifically designed for the parallel bulk moving of data from one storage technology to another.  It comes from the Dell/EMC support labs, and is what EMC support engineers use for migrating data.  It certainly is possible to install ecs-sync outside of Grace, for example on a lab workstation to rapidly move data to or from ROSS.  However in this article we only discuss the use case of moving data between ROSS and Grace using the ecs-sync installed by the HPC Admins.

...

<?xml version="1.0" encoding="UTF-8" standalone="no"?>
<syncConfig xmlns="http://www.emc.com/ecs/sync/model">  
    <options>
        <!-- ... the global settings ... -->
    </options>

    <source>
        <filesystemConfig>
            <!-- ... file copy settings ... -->
        </filesystemConfig>
    </source>

    <target>
        <ecsS3Config>
            <!-- ... ecs s3 copy settings ... -->
        </ecsS3Config>         
    </target>
</syncConfig>

...

The easiest way to create an xml config file is to copy an exiting template, and just change the pertinent values.  To help guide, here are examples of the 3 sections with descriptions of the parameters

Global Options

This example shows available global parameter options.  These are typical values, and may be changed.  The important options are deleteSource, fourceSync, recursive, retryAttempts, threadcount, verify and verifyOnly,

<syncConfig xmlns="http://www.emc.com/ecs/sync/model">
    <options><syncConfig xmlns="http://www.emc.com/ecs/sync/model">   <options>
        <bufferSize>524288</bufferSize>
        <!-- <dbConnectString>dbConnectString</dbConnectString>
        <dbEncPassword>ChangeMe1234</dbEncPassword> -->
        <dbFile>~<dbFile>/home/YourHomeDirectory/ecs-dbFile</dbFile>
        <dbTable>ecs-dbTable<_sync_log</dbTable>
        <deleteSource>false</deleteSource>
        <estimationEnabled>true</estimationEnabled>
        <forceSync>false</forceSync>
        <ignoreInvalidAcls>false</ignoreInvalidAcls>
        <logLevel>quiet</logLevel>
        <monitorPerformance>true</monitorPerformance>
        <recursive>true</recursive>
        <rememberFailed>false</rememberFailed>
        <retryAttempts>2</retryAttempts>
        <!-- <sourceListFile>sourceListFile</sourceListFile> -->
        <syncAcl>false</syncAcl>
        <syncData>true</syncData>
        <syncMetadata>true</syncMetadata>
        <syncRetentionExpiration>false</syncRetentionExpiration>
        <threadCount>24</threadCount>
        <timingWindow>1000</timingWindow>
        <timingsEnabled>false</timingsEnabled>
        <verify>false</verify>
        <verifyOnly>false</verifyOnly>
    </options>

The following table describes all the global options.

File System Options

This example shows the file system as a source.  It could also be a target.  The filesystem storage type plugin reads and writes to or from a directory.  The example shows typical values, which generally need to change to fit your situation.

...

Ross (ECS S3) Options

ROSS uses the ECS S3 storage system type.  This example shows ROSS as a destination.  It could also be a target.  The example shows typical values, some of which must be changed for the sync to work, such as accessKey, bucketName, keyPrefix, and secretKey.

...

rclone

Although not as fast nor efficient as ecs-sync, the rclone program (https://rclone.org/) is one of the most versatile tools for accessing object stores such as ROSS.  It includes features for browsing various types of object and cloud storage systems, as well as local files using Posix file system conventions, using commands such as ls (list objects/files in a path), lsd (list directories/containers/buckets in a path), copy, move, delete, etc.  For some object/cloud storage systems, rclone can mount buckets as if they were a network file systems, though with reduced functionality and speed, so that users can use familiar commands to access bucket contents instead of the rclone commands or object API.  But the most popular feature of rclone is the ability to sync a directory tree to a bucket on an object store using a familiar rsync-like syntax.  The rclone site includes documentation on how to install, configure, and use rclone.  The rclone program is licensed under the permissive, open-source MIT license, hence is free to use and distribute.

...