05.11.2013 Views

Slide 1/17 - Thomas-Krenn.AG

Slide 1/17 - Thomas-Krenn.AG

Slide 1/17 - Thomas-Krenn.AG

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

<strong>Slide</strong> 1/<strong>17</strong>


SSD Caches<br />

Festplatten-RAIDs mit Flashspeicher<br />

beschleunigen<br />

TK Roadshow 2012<br />

<strong>Slide</strong> 2/<strong>17</strong>


Agenda<br />

1) Wann cachen?<br />

2) Definitionen<br />

3) Feature-Vergleich<br />

4) Funktionsweise<br />

5) Performance Tests<br />

6) Fazit<br />

<strong>Slide</strong> 3/<strong>17</strong>


1) Wann cachen?<br />

if(dataSet == known &&<br />

ramAvailable != enough &&<br />

appAccess == analyzed &&<br />

perfTests == available)<br />

CheckForCacheTechnologies();<br />

else<br />

NeedMoreInfo();<br />

<strong>Slide</strong> 4/<strong>17</strong>


2) Definitionen<br />

<strong>Slide</strong> 5/<strong>17</strong>


3) Feature-Vergleich<br />

WB WT WA/read-only<br />

FlashCache<br />

CacheCade<br />

1<br />

MaxCache<br />

2<br />

!<br />

1<br />

Including ForcedWB<br />

! Achtung: Derzeit kein redundanter Cache mit mehreren SSDs möglich<br />

2<br />

Including InstantWB<br />

<strong>Slide</strong> 6/<strong>17</strong>


3) Feature-Vergleich<br />

FlashCache 1<br />

Skip sequential I/O<br />

CacheCade 2<br />

MaxCache 3<br />

●<br />

Das Festplatten-RAID kann für sequentiellen<br />

I/O schneller sein als die SSD<br />

1 Konfigurierbar via sysctl (threshold)<br />

2<br />

Keine weiteren Details bekannt<br />

3<br />

Nicht konfigurierbar, immer ausgelassen<br />

<strong>Slide</strong> 7/<strong>17</strong>


3) Feature-Vergleich<br />

Cache persistence<br />

FlashCache !<br />

1<br />

CacheCade<br />

2<br />

MaxCache<br />

3<br />

●<br />

Das Festplatten-RAID kann für sequentiellen<br />

I/O schneller sein als die SSD<br />

1 Nur WB Cache ist persistent<br />

2<br />

Persistent in alle Cache Modus<br />

3<br />

Dirty Shutdown: WT – discarded, WB – LV failed<br />

<strong>Slide</strong> 8/<strong>17</strong>


3) Feature-Vergleich<br />

FlashCache<br />

Hot spot detection<br />

CacheCade<br />

MaxCache<br />

<strong>Slide</strong> 9/<strong>17</strong>


FlashCache<br />

CacheCade<br />

MaxCache


5) MB/s - read<br />

<strong>Slide</strong> 11/<strong>17</strong>


5) MB/s - write<br />

<strong>Slide</strong> 12/<strong>17</strong>


5) IOPS - randread<br />

<strong>Slide</strong> 13/<strong>17</strong>


5) IOPS - randwrite<br />

<strong>Slide</strong> 14/<strong>17</strong>


6) Fazit<br />

Applikation 2x Erkenntnisse testen<br />

Ergebnisse überprüfen<br />

umsetzen<br />

Yes<br />

No<br />

No<br />

Establish a<br />

Baseline<br />

Know<br />

application<br />

RAM<br />

HDD is<br />

bottleneck<br />

SSD alone<br />

to small<br />

Test<br />

caching<br />

WB<br />

WT<br />

<strong>Slide</strong> 15/<strong>17</strong>


Backup: Test system<br />

●<br />

●<br />

●<br />

SSDs<br />

●<br />

●<br />

Intel Series 320 160GB<br />

Via HPA reduced to 32GB<br />

RAID Controller<br />

● LSI MegaRAID SAS 9260-4i<br />

●<br />

Adaptec 6805Q<br />

Software<br />

● Fio 2.0.7<br />

● Ubuntu 12.04<br />

– Updates from Release Day<br />

<strong>Slide</strong> 16/<strong>17</strong>


Backup: Test script<br />

●<br />

●<br />

●<br />

●<br />

●<br />

●<br />

●<br />

●<br />

●<br />

flashcache_create<br />

●<br />

Create a WB caching device<br />

cache_all=0<br />

●<br />

Don't cache ext4 initialization<br />

mkfs.ext4 -q -E lazy_itable_init=0,<br />

lazy_journal_init=0 /dev/mapper/fc-root<br />

mount /dev/mapper/fc-root<br />

cache_all=1<br />

Call fio<br />

umount /dev/mapper/fc-root<br />

dmsetup remove<br />

flashcache_destroy /dev/sdd<br />

<strong>Slide</strong> <strong>17</strong>/<strong>17</strong>

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!