BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:CLyDE Mid-Flight: What we have learnt so far about the SSD-Based I
 O Stack
DTSTART:20140214T120000
DTEND:20140214T140000
DTSTAMP:20260510T025353Z
UID:b0673dc182861e1c0867c6bbb4ba102254afdc32efbe5078460178bd
CATEGORIES:Conferences - Seminars
DESCRIPTION:prof. Philippe Bonnet\, Associate professor at IT University o
 f Copenhagen\nThe quest for energy proportional systems and the growing pe
 rformance gap between processors and magnetic disks has led to the adoptio
 n of SSDs as secondary storage of choice for a large range of systems.  I
 ndeed\, SSDs offer great performance (tens of flash chips wired in paralle
 l can deliver hundreds of thousands accesses per second) with low energy c
 onsumption. This evolution introduces a mismatch between the simple disk m
 odel that underlies the design of today’s database systems and the compl
 ex SSDs of today’s computers. This mismatch leads to unpredictable perfo
 rmance\, with orders of magnitude slow down in IO latency that can hit an 
 application anytime. To attack this problem\, the obvious approach is to c
 onstruct models that capture SSDs' performance behaviour. However\, our pr
 evious work has shown the limits of this approach because (a) performance 
 characteristics and energy profiles vary significantly across SSDs\, and (
 b) performance varies in time on a single device based on the history of a
 ccesses. The CLyDe project is based on the insight that the strict layerin
 g that has been so successful for designing database systems on top of mag
 netic disks is no longer applicable to SSDs. In other words\, our central 
 hypothesis is that the complexity of flash devices cannot be abstracted aw
 ay as it results in unpredictable and suboptimal performance. We postulate
  that database system designers need a clear and stable distinction betwee
 n efficient and inefficient patterns of access to secondary storage\, so t
 hat they can adapt space allocation strategies\, data representation or qu
 ery processing algorithms. We propose that (i) SSDs should expose this dis
 tinction instead of aggressively mitigating the impact of inefficient patt
 erns at the expense of the efficient ones\, and (ii) that operating system
  and database system should explicitly provide mechanisms to ensure that e
 fficient access patterns are favoured.  We thus advocate a co-design of S
 SD controllers\, operating system and database system with appropriate cro
 ss-layer optimisations.\nIn this talk\, I will report on the lessons we ha
 ve learnt so far in the project. In particular\, I will describe the SSD s
 imulation frameworks that we have developed to explore cross layer designs
 : EagleTree and LightNVM. I will discuss our findings on the importance of
  scheduling within an SSD. I will present our contribution to the re-desig
 n of the Linux block layer\, that makes it possible for Linux to keep up w
 ith SSD performance on multi-socket systems. Finally\, I will present prel
 iminary results on the co-design of file system and SSDs.\nCLyDE is a join
 t project between IT University of Copenhagen and INRIA Paris Rocquencourt
 \, started in 2012 and funded by the Danish Council for Independant Resear
 ch.
LOCATION:BC 410 https://plan.epfl.ch/?room==BC%20410
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR