SG0-001 · Question #306
A four-way cluster is attached to two fabrics. Each member has two HBAs connected to each fabric. Each fabric has two subsystem connections. Each cluster member is a primary for one application. All L
The correct answer is C. SCSI Queue Depth too low. A large read pending at the host during multi-threaded sequential I/O backup is most likely caused by a SCSI Queue Depth that is too low, preventing the host from issuing sufficient concurrent I/O requests.
Question
A four-way cluster is attached to two fabrics. Each member has two HBAs connected to each fabric. Each fabric has two subsystem connections. Each cluster member is a primary for one application. All LUNs appear on all storage ports with workload equally distributed across all. Each host has two dedicated HBAs for tape that can each see four independent LTO II (35MB/s) tape drives. All files are large and involve sequential I/O. Performance reporting shows a large read pending at the host during multi-threaded backup. What is the most probably cause for the large read pending at the host?
Options
- Anative command queuing
- BSCSI Queue Depth too high
- CSCSI Queue Depth too low
- Dthrottling threshold is too high
How the community answered
(51 responses)- A16% (8)
- B8% (4)
- C45% (23)
- D31% (16)
Why each option
A large read pending at the host during multi-threaded sequential I/O backup is most likely caused by a SCSI Queue Depth that is too low, preventing the host from issuing sufficient concurrent I/O requests.
Native Command Queuing (NCQ) is a SATA feature and not typically relevant for enterprise SCSI/FC tape drives and HBAs in this context.
A SCSI Queue Depth that is too high typically leads to increased latency at the storage target, not directly to the host having pending reads because it cannot issue commands.
A low SCSI Queue Depth restricts the number of I/O commands that a host can send to a tape drive simultaneously. For multi-threaded backups involving large sequential files, this limitation starves the I/O pipeline, causing the host to have pending read requests as it waits for available command slots.
A throttling threshold that is too high means less aggressive throttling, which would allow more I/O to proceed and is unlikely to cause pending reads due to resource limitation.
Concept tested: SCSI Queue Depth, I/O performance tuning, multi-threaded I/O
Source: https://learn.microsoft.com/en-us/windows-server/administration/performance-tuning/subsystem/storage/storage-performance
Topics
Community Discussion
No community discussion yet for this question.