SNIA-SDC23-Modrzyk-Bridging-the-Gap-BetweenH-SMRs-and-SDS_1.pdf

1、1|2023 SNIA&Leil Storage O copyrightVirtual ConferenceSeptember 28-29,2021Bridging the Gap Between Host Managed SMR Drives and Software-Defined StoragePresented byPiotr Modrzyk2|2023 SNIA&Leil Storage O copyrightSpeakerPiotr ModrzykPrincipal Architect at Leil StorageX-googler and Creator of LizardFS

2、3|2023 SNIA&Leil Storage O copyrightOutline Brief intro to SaunaFS Simplified SaunaFS architecture Chunks SMR restrictions Problems for conventional Chunks Solution Divide the Chunks into Metadata and Data Handle non-sequential writes:fragment the chunks&garbage collection SMR libraries overview and

3、 why ZoneFS Testing framework extension for zoned devices Inspecting the content of zones with a graphical UI tool 4|2023 SNIA&Leil Storage O copyrightBrief intro to SaunaFS SaunaFS is a Distributed File System written mostly in C+which implements concepts introduced by Google File System.SaunaFS is

4、 divided into:Metadata Servers(master,shadows and metaloggers)Data Servers(chunkservers)Clients(native Linux/Windows,NFS)In the Chunkserver side:Files are divided into Chunks(up to 64 MiB)(chunks are logically divided into Blocks of 64 KiB,which is the minimum block size)For each block,4 bytes of CR

5、C are also stored in the Chunk metadata.See https:/en.wikipedia.org/wiki/Google_File_System5|2023 SNIA&Leil Storage O copyrightSimplified writing processThe client wants to create a file and to write data to this file:The client asks the Master server where(Chunkservers)to put the first chunk.The cl

6、ient connects directly to the Chunkservers and starts sending the data in Blocks of 64 KiB+4 B of CRC.The Chunkservers check the Blocks CRC against the received data and write to the storage devices,metadata is updated if needed.6|2023 SNIA&Leil Storage O copyrightEC4+2 file with 2 Chunks Example Fi