Computational Storage Devices (CSDs), which are storage devices including general-purpose, special-purpose, and/or re-configurable processing units, are now becoming commercially available from different vendors. CSDs are capable of running software that usually runs on the host CPU – but on the storage device, where the data reside. Thus, a server with one or more CSDs may improve the overall performance and energy consumption of software dealing with a large amount of data. With the aim of fostering CSD’s research and adoption, this position paper argues that commercially available CSDs are still missing a wealth of functionalities that should be carefully considered for their widespread deployment in production datacenters. De facto, existing CSDs ignore (heterogeneous) resource management issues, do not fully consider security nor multi-user, nor data consistency, nor usability. Herein, we discuss some of the open research questions, and to what degree several well-known programming models may help solving them – considering also the design of the hardware and software interfaces.
|Number of pages||6|
|Publication status||Published - 11 Jan 2021|
|Event||Conference on Innovative Data Systems Research 2020 - Virtual Confernce|
Duration: 11 Jan 2021 → 15 Jan 2021
|Conference||Conference on Innovative Data Systems Research 2020|
|Abbreviated title||CIDR 2020|
|Period||11/01/21 → 15/01/21|