Towards Marrying Files to Objects

by   Kunal Lillaney, et al.

To deal with the constant growth of unstructured data, vendors have deployed scalable, resilient, and cost effective object-based storage systems built on RESTful web services. However, many applications rely on richer file-system APIs and semantics, and cannot benefit from object stores. This leads to storage sprawl, as object stores are deployed alongside file systems and data is accessed and managed across both systems in an ad-hoc fashion. We believe there is a critical need for a transparent merger of objects and files, consolidating data into a single platform. Such a merger would extend the capabilities of both object and file stores while preserving existing semantics and interfaces. In this position paper, we examine the viability of unifying object stores and file systems, and the various design tradeoffs that exist. Then, using our own implementation of an object-based, POSIX-complete file system, we experimentally demonstrate several critical design considerations.




Object Files and Schemata: Factorizing Declarative and Procedural Knowledge in Dynamical Systems

Modeling a structured, dynamic environment like a video game requires ke...

Fragmented Objects: Boosting Concurrency of Shared Large Objects

This work examines strategies to handle large shared data objects in dis...

How Big Are Peoples' Computer Files? File Size Distributions Among User-managed Collections

Improving file management interfaces and optimising system performance r...

Autoplot: A browser for scientific data on the web

Autoplot is software developed for the Virtual Observatories in Heliophy...

Uncovering the epistemological and ontological assumptions of software designers

The ontological and epistemological positions adopted by information sys...

Implementation of an Android Framework for USB storage access without root rights

This bachelor thesis describes the implementation of an Android framewor...
This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

1 Introduction

Object storage has gained significant traction in recent years owing to an explosion in the amount of unstructured data and the popularity of applications using RESTful interface to access such data. Some estimates predict that by 2019 more than 30% of the storage capacity in data centers will be provided by object storage 

[5], and that by 2020 the world will produce 44 zetabytes of unstructured data each year [25]. Object stores can offer cost of ownership and scalability improvements over traditional file systems. Users and enterprises increasingly look to object storage as an economical solution for storing unstructured data [6] [7]. Object stores are characterized by Web-style access, flat namespaces, immutable data, relaxed consistency, and rich user-defined metadata. Most object stores support simple data access operations: GET, PUT, and DELETE. Existing applications are often unable to benefit from object storage because they rely on traditional file system interfaces and features. Many applications, for example genomic workflows [15] [13], are dependent on a namespace hierarchy and file pointers which are not supported by object storage. Objects do not support updating in place, and need to be updated and rewritten as a unit; existing big data workflows may need to be rearchitected to use object storage in order to avoid the performance penalty introduced by incremental updates [26]. Given the flat name space in object stores, some space organization operations such as creation and deletion of directories could help analysis applications manage data. Adding file access protocols to object stores would broaden their use cases and increase their adoption rate. With the emergence of low-cost cloud and on-premise object storage [2] [14] [11], object storage file systems can be a cost effective alternative to block-based file systems. File system interfaces over object storage would allow for a seamless migration of existing applications [12]. This would reduce storage costs without requiring porting applications to a new interface. It would also aid the current efforts of cloud providers towards storage consolidation to eliminate storage sprawl—the spread of data across different media and interfaces. Storage sprawl causes numerous issues, including over-provisioning, cost inflation, reduced backup efficiency, and poor quality of service (QoS) [23]. The key to eliminating storage sprawl is dual access—the ability to read and write data through both file system interfaces and object storage APIs. However, there are many design considerations that expose tradeoffs in the quality and performance of dual access. We explore these tradeoffs using our implementation. We recommend an initial design for dual access that chooses a simple file-to-object mapping and a more complex indirect naming scheme. This design allows data access from the native object API without modifications, but makes identifying data across systems more complex in order to avoid data copies during file system metadata operations. Other choices are discussed with quantified tradeoffs. We also observe that write-back caching is critical to making dual access efficient. We conclude that file systems built on object stores can eliminate storage sprawl while realizing a large fraction of the performance of the underlying object store.

2 Related work

Object storage file systems are not the same as file systems designed to use object-based storage devices (OSDs), for example Lustre [22]. The object storage file systems that we explore were developed to operate over existing, generic object storage systems, and to support data and application portability. Object storage file systems can be broadly categorized using two properties. First, whether they are generic and support multiple back-end object stores, important for providing flexibility and avoiding vendor lock-in. Second, whether they are compliant with POSIX standards, which we refer to as POSIX complete. S3Fuse [18], Goofys [10], RioFS [17], GCSfuse [9], Blobfuse [4], SVFS [24] and MarFS [12] are some examples of file systems which are generic, but not POSIX complete. Most of these systems do not support POSIX features such as symlinks, hardlinks, or file attributes (chmod), and have poor performance for random writes. CephFS [29] is a popular example of a POSIX complete file system; however, it can only function with its own RADOS object store. Some recent systems are both generic and complete, but lack support for dual access. These do not address the issue of storage sprawl. S3QL [19] is POSIX complete and generic but does not support distributed access. BlueSky [28] uses caching gateways to aggregate writes in log segments that are later pushed to an object store. SCFS [3] is a generic system with near-POSIX semantics that supports distributed access.

3 Design considerations

We believe that dual access to the data through both object and file-system interfaces is critical to eliminate storage sprawl. For this paper, we assume a limited, yet representative, object interface with the following operations: (1) PUT(name, data) adds a named object, (2) GET(name) retrieves an object, and (3) DEL(name) deletes an object by name. The file system interface has POSIX-defined operations, such as file create, open, read, write, close, delete, rename, as well as directory operations. Object and file-system interfaces have a number of fundamental namespace and data access differences. For instance, in object stores users cannot create directories and subdirectories of objects, and cannot operate on directories as a whole, e.g., rename them. (Though buckets of objects are supported by many object stores they cannot be nested and their number is often limited, e.g., to around 100 per account in AWS S3 [1]). Data access differences include the inability of object storage to perform an in-place partial update of the data, a common operation in many file system workloads. In light of these disparities, providing dual access is challenging. In fact, the goal of dual access is often in direct conflict with that of achieving high performance. We present design considerations for overlaying a file system on generic object storage and explore the corresponding performance impacts. In the following text we refer to an implementation of an abstract file system working on top of an object store as ObjectFS.

3.1 File-to-object mapping

A fundamental question in the design of ObjectFS is how to map files to objects.  11 mapping represents a whole file with a single object in the object store. This mapping allows simple and intuitive dual access to the data from the user perspective. 11 mapping can drastically reduce the performance of file writes because a small modification to a file requires a GET and a PUT on the complete object.  1N mapping splits an individual file into multiple objects, each storing a segment of the file. The segments can be of a fixed size, as in a traditional block-based file system, or of variable size, as in an extent-based file system. Splitting a file into multiple objects enables faster in-file updates by only writing smaller-sized objects which map to the updated parts of the file. However, accessing the data from the object interface in 1N mapping is no longer intuitive and requires additional metadata in object-based user applications.  N1 mapping packs multiple files into a single object. This can improve performance when a subset of small-sized files tend to be accessed together. Accessing data through the object interface is even more complicated with N1 mapping than with 1N mapping.  Hybrid mapping varies the mapping within the same file system. For example, ObjectFS could create new objects for each incoming write (as extents) and then reassemble them into complete objects in the background. This hybrid mapping trades consistency of object and file system views of the data for performance.

3.2 Object naming policy

Although the naming of the objects is tightly coupled to file-to-object mapping, we discuss naming separately to isolate and demonstrate relevant difficulties. For simplicity, we assume 11 mapping.  file-name policy names an object identically to the corresponding file. Such a policy allows intuitive dual access to the data, but with a substantial caveat. Two files with identical names but in different directories cause a conflict in the flat object namespace. So, this policy is applicable only in limited scenarios, e.g., when a read-only file system is deployed on a pre-populated object storage to perform analytics.  file-path policy creates an object named after the file’s complete path. This policy is both convenient for dual access and avoids conflicts in the object storage namespace. However, a rename of a file requires a GET and a PUT, making metadata operations slow. A directory rename requires a GET-PUT sequence for every file in the directory and performance scales down as the total size of all files in a directory grow. Another limitation of file-path policy is its inability to support hardlinks (different files referring to the same data).  inode-number policy names the file using the file system inode number as the object name. The assumption here is that ObjectFS, similar to a traditional UNIX file system, maintains a mapping of file paths to inode numbers. File paths are translated to inode numbers using a lookup procedure. The inode-number policy hinders dual access, because the inode number needs to be looked up. Similarly, files created through the object store will need to reference file system metadata for a name. inode-number performs renames quickly as no objects need to be moved: only the mapping is updated.  user-defined policies allow the user to drive the naming scheme. In one potential implementation, the inode in ObjectFS records the name of the corresponding object. When a new file is created, ObjectFS executes a user-defined naming policy to derive the object name. The naming policy can take as an input such file system information as the file name and path, owner, inode number, and more. A corresponding naming policy is required to generate full file paths based on the properties of any objects created directly in the object store. user-defined policies are more flexible than inode-number but need to be carefully designed to be convenient, and avoid naming conflicts.

3.3 Metadata

ObjectFS could potentially use several different locations to persistently store its metadata.  in-object placement stores file metadata (e.g., owner, permissions, timestamps) in the object store itself. One option is to store metadata in the same object as data, but this requires cloud-native applications to deal with metadata during GETs and PUTs, which compromises dual access. Another option maintains separate metadata objects: one per file or one per a group of files. In this case, dual access is not directly hindered but “confusing” metadata objects are visible in the results of a LIST request. Furthermore, object stores typically exhibit high latency, which would metadata operations (e.g., accessing or updating atime or uid). This typically leads to poor overall performance.  in-object-meta relies on the fact that the majority of object storage implementations can store user-defined metadata in association with an object. Access to user-defined metadata is independent of access to the object data, and has comparatively lower latency. The concept is similar to extended attributes in file systems. This approach offers for dual metadata access in addition to dual data access. Object-based applications can request file-system metadata through the object interface. However, it relies of a richer object API that is not generic.  independent stores the file system metadata in a storage solution separate from the object store. A key-value store with high scalability and low access latency is one feasible configuration. In this case, metadata operations like inode lookup or stat() would not require slow accesses to the object storage. A downside is the higher system complexity and the need to maintain additional storage system for metadata. independent precludes accessing metadata through the object interface. Although, ObjectFS could asynchronously write metadata from the metadata store to corresponding objects in one of the other formats.

3.4 Caching

Caching plays an integral role in file system performance. For ObjectFS, both read and write caches are important because the underlying object storage has high latencies and operates efficiently only when transferring large objects. We limit our discussion to two fundamental design options:  local cache has its independent instances on every node where the file system is mounted. Each cache instance buffers data read or written by the local node. RAM or a local SSD can be used for cache space. For local cache, ObjectFS needs to maintain cache consistency between nodes using, e.g., lock tokens [21]. Since object-based accesses do not go through the file system cache, cloud-native applications could see outdated versions of the data until caches are synced.  unified cache is a distributed and shared tier between file system clients. Data cached by one client can quickly be fetched from the cache by other clients. Redis [20] and Memcached [8] are systems suitable to implement a unified cache. Caching nodes can be collocated with file system mount nodes or deployed in a separate cluster. A unified cache may re-export an object interface so that object-based applications access the same data consistently and realize the benefits of the cache.

4 Implementation

Figure 1: An overview of ObjectFS and the data flow across different components.

To illustrate the design choices of Section 3, we developed an ObjectFS prototype using FUSE [27]. Our implementation is simple and modular to facilitate experimentation with various ObjectFS configurations. Figure 1 depicts ObjectFS’s high-level architecture. ObjectFS’s user-space daemon is responsible for the main logic of the file system: to perform file lookups, reads, writes, etc. ObjectFS uses an independent metadata service that is abstracted as a key-value store. In this paper we use Redis, an in-memory key-value store, because of its low latency, distributed design, support of transactions, and ability to persistently store in-memory data. The object library communicates with object storage using a common subset of object operations. We currently support AWS S3 [1] and OpenStack Swift [16]. Many object stores support multi-part upload and download of large objects. If available, ObjectFS utilizes this feature to improve performance. ObjectFS can be configured to run with or without a cache. Currently we use Redis to cache data. By default, data is fetched into the cache on a file open and is flushed back to the object store on a file close. ObjectFS supports all major POSIX operations, and we are able to successfully boot a Linux OS directly from an object store using ObjectFS. Our implementation is open source and is available for collaborative development and reuse at our GitHub repository (

5 Evaluation

Our evaluation quantitatively demonstrates some key design trade-offs presented in Section 3. We chose four basic workloads—streaming reads, streaming writes, random writes, and renames—and measured performance on various ObjectFS designs. We used Amazon Web Services (AWS) as a testbed [2]. An ObjectFS client was mounted on a t2.2xlarge compute instance with 8 vCPUs, 32GB of RAM, and AWS moderate network bandwidth. ObjectFS’s metadata server was deployed on the same instance. We used AWS S3 object storage with the default standard class of storage as a backend. The S3 buckets had default settings with object logging enabled and object versioning and transfer acceleration disabled. Streaming reads: Read experiments demonstrate that ObjectFS tracks the performance of the underlying object store for sequential workloads. We perform sequential reads, in 4 MB record sizes, on files stored as objects in 11 mapping. We measure the I/O throughput of S3 and ObjectFS when varying the file size from 64MB to 1GB and using S3’s multipart download with 2, 4, and 8 threads. Multipart downloads divide the object access into parts, parallelized over multiple threads. Figure 2 shows that multipart downloads mitigate the performance overhead of ObjectFS and that ObjectFS realizes a large fraction of S3’s potential bandwidth. The small remaining overhead comes from metadata operations and caching overhead.

Figure 2: Streaming read performance on native S3 and ObjectFS. mp represents multipart download with the corresponding number of threads used.
Figure 3: Streaming write performance on native S3 and ObjectFS. mp represents multipart upload with the corresponding number of threads used.

Streaming writes: Write experiments demonstrate that caching is critical to realizing performance in ObjectFS. The experiment performs sequential writes, in 4 MB record sizes, on files stored as objects in 11 mapping. We measure the I/O throughput of S3 and ObjectFS when varying the file size from 64MB to 1GB and using S3’s multipart upload with 2, 4, and 8 threads. ObjectFS implements a write-back cache: a write is stored locally in a Redis memory store and written back to the object store when the object is closed. Figure 3 shows that caching enables reasonable write throughput when compared with native S3 bandwidth, and that multi-part uploads reduce overhead and increase throughput. Multi-part uploads overlap data transfer with metadata operations in multiple threads. Without a write-back cache, each write results in a read-modify-write in the object store, reducing throughput to less than 2MB/sec. With caching, we aggregate writes in the cache and issue many transfers in parallel. Increasing multi-part uploads beyond 8 threads shows no more performance improvement. We theorize that the physical footprint of an object is limited to a few storage servers and that more threads result in smaller messages to the same set of servers.

Figure 4: Random write performance for different file object mapping designs without no caching (left) and with write-back caching (right).

Random writes: Evaluating the performance of random writes demonstrates the performance tradeoffs among the different file to object mappings. We perform writes of 4 MB to random file offsets aligned to 4 MB using different mapping schemes: 11, 1N (1 MB chunks) and 1N (4 MB chunks). Figure 4 (left) shows throughput for write-through workloads without caching. In this scenario, 11 is much worse than 1N, 1.7 MB/s versus 15 MB/s. With 11 mapping, each 4 MB write performs a partial write or read-modify-write against the underlying object, whereas 1N mappings write entire object(s). More importantly, all data rates are remarkably low without caching. With write caching, Figure 4 (right), ObjectFS defers individual writes and avoids read-modify-writes to realize an order-of-magnitude performance improvement. This experiment uses 8 I/O threads, doing multi-part upload for 11 and parallel transfers to mulitple objects for 1N. The 1N mappings are slightly slower than 11 due to overhead for RESTful calls to more objects. Caching raises the random-write throughput of ObjectFS close to the sequential performance of S3.

Figure 5: Rename latency for different object naming designs. Experiments look at renaming a single large file (left) and many 1 MB files (right).

Metadata performance: We also examine the performance associated with different file naming conventions that affect dual access. We perform two experiments: The first, Figure 5 (left), renames files of different sizes. With full path naming, a rename results in an S3 server-side copy of the object; the performance of rename operations thus scales with the file size, taking 2 seconds for a 64MB file and 30 seconds for 1GB file. When naming by inode number, rename is fast (less than 0.005 seconds) and does not depend on file size; in this case renames in ObjectFS are metadata-only operations. The second experiment, Figure 5 (right), renames directories with varying number of files, each 1MB in size. With full path naming, latency increases with the number of files. With inode naming, latency is consistent and low.

6 Discussion and Conclusions

Dual access to data through object and file system APIs is feasible with a judicious choice of design options. Based on our evaluation, we propose a specific design that preserves object APIs and incurs only minor overheads when accessing data through the ObjectFS file system. Specifically, a 11 file to object mapping allows the object store APIs to access data without assembling data from multiple objects. We recommend an indirect naming scheme based on naming objects by file system inode number. This choice enables the system to perform metadata operations without copying, but adds complexity to object access, which must resolve the file system name to an inode number. This is only one design; our evaluation quantifies tradeoffs and thus can aid in future designs of object storage file systems. We also conclude that write-back caching is a critical technology for deploying object-based file systems. Caching aggregates multiple writes in memory, converting many synchronous writes into fewer larger asynchronous writes. Without caching, object file systems have low throughput and high latency. This would limit the applications that could adopt object based file systems to those that perform synchronous writes infrequently. Acknowledgements: This work was funded by a study grant from IBM Research.


  • [1] Amazon Simple Storage Service (S3). Note: External Links: Link Cited by: §3, §4.
  • [2] Amazon Web Services Pricing. Note: External Links: Link Cited by: §1, §5.
  • [3] A. Bessani, R. Mendes, T. Oliveira, N. Neves, M. Correia, M. Pasin, and P. Verissimo (2014) SCFS: A Shared Cloud-backed File System. In 2014 USENIX Annual Technical Conference (USENIX ATC 14), Philadelphia, PA, pp. 169–180. External Links: ISBN 978-1-931971-10-2, Link Cited by: §2.
  • [4] Blobfuse. Note: External Links: Link Cited by: §2.
  • [5] A. Chandrasekaran, R. Bala, and G. Landers (2017) Critical Capabilities for Object Storage. Gartner. Cited by: §1.
  • [6] A. Cockroft, C. Hicks, and G. Orzell (2011) Lessons Netflix learned from the AWS outage. Netflix Techblog. Cited by: §1.
  • [7] E. Deelman, G. Singh, M. Livny, B. Berriman, and J. Good (2008) The Cost of Doing Science on the Cloud: The Montage Example. In High Performance Computing, Networking, Storage and Analysis, 2008. SC 2008. International Conference for, pp. 1–12. Cited by: §1.
  • [8] B. Fitzpatrick (2004-08) Distributed Caching with Memcached. Linux J. 2004 (124), pp. 5–. External Links: ISSN 1075-3583, Link Cited by: §3.4.
  • [9] GCSfuse. Note: External Links: Link Cited by: §2.
  • [10] Goofys. Note: External Links: Link Cited by: §2.
  • [11] Google Cloud Platform Pricing. Note: External Links: Link Cited by: §1.
  • [12] J. Inman, W. Vining, G. Ransom, and G. Grider (2017) MarFS, a Near-POSIX Interface to Cloud Objects. ; Login 42 (LA-UR–16-28720; LA-UR–16-28952). Cited by: §1, §2.
  • [13] B. Langmead, K. D. Hansen, and J. T. Leek (2010) Cloud-scale RNA-sequencing differential expression analysis with Myrna. Genome biology 11 (8), pp. R83. Cited by: §1.
  • [14] Microsoft Azure Pricing. Note: External Links: Link Cited by: §1.
  • [15] A. O’Driscoll, J. Daugelaite, and R. D. Sleator (2013) ‘Big data’, Hadoop and cloud computing in genomics. Journal of Biomedical Informatics 46 (5), pp. 774–781. External Links: ISSN 1532-0464 Cited by: §1.
  • [16] OpenStack Swift. Note: External Links: Link Cited by: §4.
  • [17] RioFS. Note: External Links: Link Cited by: §2.
  • [18] S3Fuse. Note: External Links: Link Cited by: §2.
  • [19] S3QL. Note: External Links: Link Cited by: §2.
  • [20] S. Sanfilippo and P. Noordhuis Redis. Note: External Links: Link Cited by: §3.4.
  • [21] F. B. Schmuck and R. L. Haskin (2002) GPFS: A Shared-Disk File System for Large Computing Clusters.. In FAST, Vol. 2. Cited by: §3.4.
  • [22] P. Schwan et al. (2003) Lustre: Building a file system for 1000-node clusters. In Proceedings of the 2003 Linux symposium, Vol. 2003, pp. 380–386. Cited by: §2.
  • [23] H. Smith (2016) Data Center Storage: Cost-Effective Strategies, Implementation, and Management. CRC Press. External Links: ISBN 9781439834886 Cited by: §1.
  • [24] SVFS. Note: External Links: Link Cited by: §2.
  • [25] V. Turner, J. F. Gantz, D. Reinsel, and S. Minton (2014) The Digital Universe of Opportunities: Rich Data and the Increasing Value of the Internet of Things. IDC Analyze the Future, pp. 5. Cited by: §1.
  • [26] K. Vahi, M. Rynge, G. Juve, R. Mayani, and E. Deelman (2013-10) Rethinking Data Management for Big Data Scientific Workflows. In 2013 IEEE International Conference on Big Data, Vol. , pp. 27–35. External Links: Document, ISSN Cited by: §1.
  • [27] B. K. R. Vangoor, V. Tarasov, and E. Zadok (2017) To FUSE or Not to FUSE: Performance of User-Space File Systems.. In FAST, pp. 59–72. Cited by: §4.
  • [28] M. Vrable, S. Savage, and G. M. Voelker (2012) BlueSky: A Cloud-backed File System for the Enterprise. In Proceedings of the 10th USENIX Conference on File and Storage Technologies, FAST’12, Berkeley, CA, USA, pp. 19–19. External Links: Link Cited by: §2.
  • [29] S. A. Weil, S. A. Brandt, E. L. Miller, D. D. E. Long, and C. Maltzahn (2006) Ceph: A Scalable, High-performance Distributed File System. In Proceedings of the 7th Symposium on Operating Systems Design and Implementation, OSDI ’06, Berkeley, CA, USA, pp. 307–320. External Links: ISBN 1-931971-47-1, Link Cited by: §2.