Data shared by Facebook users is increasingly growing. Binary Large Objects (BLOBs) is a class of data that needs modification due to an increase in delete rates. BLOBs in Facebook incorporate documents, photos, traces, videos, source code, and head bumps. The storage capacity of the type of data named above is large. However, the imminent perception that Facebook has grown makes the BLOB storage to change. Today storage capacity is growing, and the diversity in sizes facilitates an increase in delete rates. For instance, in 2014, Facebook recorded 400 billion photos in storage capacity, and today the number is growing at an alarming rate. Notably, Facebook used Haystack as a storage capacity module that has become inefficient today due to warm BLOB. Warm BLOB is fault-tolerant, and its content request rate is lower than that of the Haystack. Thus, this paper provides a comprehensive research review of Facebook’s Warm Blob Storage System, a case for generous storage, BLOB storage design, and overall storage system.
The correlation between warm and cold storage is essential since the creation of warm storage systems. Facebook adopted a warm storage system because of age and different temperature zones. Storage facilities today also use warm storage to reduce the magnitude of loads in their storage system. Arguably, the case for warm storage incorporates temperature zones, different temperature zones, the growth of warm content, and volumes (Muralidhar et al., 2014). Concerning the existence of temperature zones, the Facebook storage facility contents are often hot while receiving requests and cools off upon receiving fewer requests. Over time, request rates decrease, and the decreasing rates show that temperature zones exist. For example, content established less than a day receives more requests in the storage system compared to a one year content.
In the case of warm storage, the research paper depicts temperature, thus creating differently in temperature zones narrative displayed in the context above. The analogy only differentiates hot from warm content in the storage system. Differentiating the latter from the former depend on temperature zones and request rates. Given that warm temperatures, zones have low request rates; the delete rates are high in this case. Concerning the growth of warm content, today’s growth percentage is more extensive in the overall BLOB content. Today’s monthly intervals of warm content are at 89%, which is an extensive improvement from the hot content (Muralidhar et al., 2014). On the other hand, volumes establish file system metadata that allows storage systems to have logical volumes. Categorically there are two types of volumes, namely unlocked and locked types volumes. Unlocked volumes support reads, delete, and creation, while locked volumes only allow deletes and reads. In this regard, locked volumes do not allow upon transitioning.
According to Muralidhar et al. (2014), BLOB storage design is an appliance that keeps components focused, simple, and well-matched for any job. BLOB designs, in this case, is an explanation of the types of volumes named above. In the design, there is data and index file. The data file is associated with metadata that holds the BLOB, while index files a look-up structure of snapshots in the storage machine’s in-memory. Besides, there is a journal file in the design that helps track deleted files in the storage system. An overall storage system consists of the controller, router tier, transformer tier, caching stack, and hot storage with Haystack (Muralidhar et al., 2014). The controller maintains unlocked volumes and ensures smooth control and functioning of the system. The controller also creates new physical volumes that back the unlocked volumes during garbage collection and compaction. The controller directly influences the smooth transitioning of the system from a broader perspective.
Router tier machines often identical and are interfaces of a BLOB storage system. The primary function of the router tier is sending operations and while adding new subsystems. In essence, the router tier extracts a volume, maps the volume physically in a machine, and sends requests later. In case of any failures, the router tier also facilitates deletes in the storage system. Notably, the transformer tier is essential in cropping and resizing photos in the right size that can easily be retrieved from the BLOB (Muralidhar et al., 2014). In Facebook, the transformer tier also allows warm storage and separates computation enabling the storage tier to run independently in the storage system. The caching stack lowers requests due to the warm storage that enables BLOB reads. An overall storage system also consists of Haystack with fault tolerance to host, disk, and racks. Haystack mostly has a hot storage system designed to utilize the IOPS. In most cases, it handles most deletes and a higher reading rate.
In conclusion, this paper provides an in-depth review of Facebook’s Warm Blob Storage System in conjunction with other aspects such as the case for warm storage, BLOB storage design, and overall storage system. The paper further addresses the case for warm storage by establishing the differentiation between cold, warm, and hit storage system. In warm storage, the article also depicts temperature zones’ existence and the differentiation of temperature that influences the Warm Blob Storage system. The context also talks about BLOB storage design that explains the different types of volumes. In this prospect, the two types of volumes depicted above involve the unlocked and locked types of volumes. The paper further discusses the overall storage system in the Warm Blob storage system. Notably, in this section, the article highlights the controller, route tier, transformer, caching stack, and hot haystack storage system as the primary aspect enabling the overall storage system. For example, the controller helps create new physical volumes that back the unlocked volumes during garbage collection and compaction, leading to a smooth transition in the overall system.