THE IMPLEMENTATION OF PREEMPTIVE GARBAGE COLLECTION TO REDUCE GARBAGE COLLECTION-INDUCED TAIL-LATENCY IN SSD RAID
Tail-latency is a well-known problem in SSD RAID. It significantly increases I/O latency which in turn makes performance-critical system servers unable to meet common reliability requirements, e.g. I/O latency should be <=100ms at the 98th percentile. One of the main causes of tail-latency in SSD...
Saved in:
Main Author: | |
---|---|
Format: | Final Project |
Language: | Indonesia |
Online Access: | https://digilib.itb.ac.id/gdl/view/43786 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Institut Teknologi Bandung |
Language: | Indonesia |
id |
id-itb.:43786 |
---|---|
spelling |
id-itb.:437862019-09-30T10:49:45ZTHE IMPLEMENTATION OF PREEMPTIVE GARBAGE COLLECTION TO REDUCE GARBAGE COLLECTION-INDUCED TAIL-LATENCY IN SSD RAID Lutta Putra, Martin Indonesia Final Project tail-latency, garbage collection, SSD, preemptive garbage collection, RAID INSTITUT TEKNOLOGI BANDUNG https://digilib.itb.ac.id/gdl/view/43786 Tail-latency is a well-known problem in SSD RAID. It significantly increases I/O latency which in turn makes performance-critical system servers unable to meet common reliability requirements, e.g. I/O latency should be <=100ms at the 98th percentile. One of the main causes of tail-latency in SSD is the Garbage Collection (GC) process. GC renders SSD unable to serve I/O during the process. Recently proposed solutions include ttRAIS which make use of parity reconstruction within RAID to nearly eliminate tail-latency caused by an SSD doing GC. We propose Preemptive Garbage Collection (PGC) to further leverage the effects brought by ttRAIS. By using FEMU, an accurate state-of-the-art flash emulator, we show that PGC by itself is able to reduce tail-latency up to 75%, while PGC and ttRAIS reduce the tail-latency up to 99.9%. text |
institution |
Institut Teknologi Bandung |
building |
Institut Teknologi Bandung Library |
continent |
Asia |
country |
Indonesia Indonesia |
content_provider |
Institut Teknologi Bandung |
collection |
Digital ITB |
language |
Indonesia |
description |
Tail-latency is a well-known problem in SSD RAID. It significantly increases I/O latency which in turn makes performance-critical system servers unable to meet common reliability requirements, e.g. I/O latency should be <=100ms at the 98th percentile. One of the main causes of tail-latency in SSD is the Garbage Collection (GC) process. GC renders SSD unable to serve I/O during the process. Recently proposed solutions include ttRAIS which make use of parity reconstruction within RAID to nearly eliminate tail-latency caused by an SSD doing GC. We propose Preemptive Garbage Collection (PGC) to further leverage the effects brought by ttRAIS. By using FEMU, an accurate state-of-the-art flash emulator, we show that PGC by itself is able to reduce tail-latency up to 75%, while PGC and ttRAIS reduce the tail-latency up to 99.9%. |
format |
Final Project |
author |
Lutta Putra, Martin |
spellingShingle |
Lutta Putra, Martin THE IMPLEMENTATION OF PREEMPTIVE GARBAGE COLLECTION TO REDUCE GARBAGE COLLECTION-INDUCED TAIL-LATENCY IN SSD RAID |
author_facet |
Lutta Putra, Martin |
author_sort |
Lutta Putra, Martin |
title |
THE IMPLEMENTATION OF PREEMPTIVE GARBAGE COLLECTION TO REDUCE GARBAGE COLLECTION-INDUCED TAIL-LATENCY IN SSD RAID |
title_short |
THE IMPLEMENTATION OF PREEMPTIVE GARBAGE COLLECTION TO REDUCE GARBAGE COLLECTION-INDUCED TAIL-LATENCY IN SSD RAID |
title_full |
THE IMPLEMENTATION OF PREEMPTIVE GARBAGE COLLECTION TO REDUCE GARBAGE COLLECTION-INDUCED TAIL-LATENCY IN SSD RAID |
title_fullStr |
THE IMPLEMENTATION OF PREEMPTIVE GARBAGE COLLECTION TO REDUCE GARBAGE COLLECTION-INDUCED TAIL-LATENCY IN SSD RAID |
title_full_unstemmed |
THE IMPLEMENTATION OF PREEMPTIVE GARBAGE COLLECTION TO REDUCE GARBAGE COLLECTION-INDUCED TAIL-LATENCY IN SSD RAID |
title_sort |
implementation of preemptive garbage collection to reduce garbage collection-induced tail-latency in ssd raid |
url |
https://digilib.itb.ac.id/gdl/view/43786 |
_version_ |
1821998976003473408 |