Hi Jason, As Shaohua discussed earlier in another thread, in this write cache approach, we are trying to solve two problems together: 1. RAID-5/6 write hole; 2. Slow sync writes to RAID-5/6 (due to read-modify-write). For other RAID types, there is no write hole issue. To get better sync write performance on RAID-10, you can use other caching solutions, like bcache or flashcache. The write cache can also be a MD device. Thanks, Song > -----Original Message----- > From: Jason Keltz [mailto:jas@xxxxxxxxxxxx] > Sent: Thursday, May 14, 2015 4:31 AM > To: Song Liu; linux-raid@xxxxxxxxxxxxxxx > Subject: Re: [PATCH 0/7] mdadm support for caching layer in raid 5/6 > > Hi.. > I'm curious - why not support write cache for other raid levels? If the write > cache becomes a fast SSD (with high endurance) for a pool that is all standard > HD then wouldn't the write cache benefit there as well? (Eg. > Raid10) . Can the write cache also be an MD device? I would think you could > use that to improve performance and redundancy of the cache? > > I'm using RHEL 7.1 on an NFS server for a virtualization back end. Because of > sync writes the write performance is poor. Its really no surprise.. I've got a 22 > disk raid10. At the moment I could try dm-cache in rhel7 in an effort to > improve write performance but its a technology preview and not sure I'd want > to rely on that for production. It's not really just a simple write cache either > which is all that I'm really after. I like simple! I could also enable async writes > on my NFS export and not sleep at night (only 3 virtualization nodes.. UPS > protection everywhere protected.. > But what if the file server crashes? You want eerie? I just got an email that > power is off on one node of my test setup....UPS failed... Aie! That could just as > easily have been the FS.) ..I could stop using MD for NFS backend storage > altogether and rely on a raid card with battery backed cache as well and then > when there's some kind of card firmware issue wait for the hardware vendor to > close my support ticket due to inactivity before they even respond.... which has > happened several times with a common 3 letter storage company that has > more than 3 letters in their name now... I could wait for someone to create a > magical hardware device that would do for MD software raid what the battery > backed cache does for hardware raid. > To this day I really don't understand why that hasn't been done. It would be a > whole lot cheaper than SSD. . it would have higher longevity.... I'd buy it... And I > suspect a billion other people as well... But I guess simple write caching will > help as well.... > > Since I don't have any of the above goodies I've decided to try a switch > from ext4 to xfs. Then test with HD based external log... Then Ill try > SSD based external log ... And finally potentially external log to an MD > raid1 SSD log device.... Though I can't find too many details, I've got a hunch > that if the log fails I'd be in trouble... > > J. > > Sent with AquaMail for Android > http://www.aqua-mail.com > > > On May 14, 2015 2:44:04 AM Song Liu <songliubraving@xxxxxx> wrote: > > > Hi, > > > > These are mdadm patches to support cache layer in raid 5/6. Shaohua > > has sent the kernel patch earlier with subject "a caching layer for > > raid 5/6". > > > > These patches add write cache support for the following commands: > > > > mdadm --detail > > mdadm --create > > mdadm --assemble > > mdadm --incremental > > mdadm --examine > > mdadm --zero-super > > > > Cache device is assigned with dev_role 0xFFFD (where 0xFFFF is for > > spare and 0xFFFE is for failed). Note that there is compatibility > > issue that older mdadm will show cache device as spare in --detail: > > > > Number Major Minor RaidDevice State > > 0 8 32 0 active sync /dev/sdc > > 1 8 48 1 active sync /dev/sdd > > 2 8 64 2 active sync /dev/sde > > 3 8 80 3 active sync /dev/sdf > > > > 4 8 17 - spare /dev/sdb1 > > > > Also, older mdadm will show cache device as "Active device 65533" > > in --examine: > > > > Device Role : Active device 65533 > > Array State : AAAA ('A' == active, '.' == missing, 'R' == > > replacing) > > > > > > Song Liu (7): > > Show device as cache in --detail > > Enable create array with write cache (--write-cache DEVICE). > > Create write-cache superblock in mdadm --create > > Assemble array with writecache > > Check write cache in incremental > > Zero write-cache superblock in --zero-super > > Add information about write-cache superblock to --examine > > > > Assemble.c | 52 +++++++++++++---- > > Create.c | 21 +++++-- > > Detail.c | 3 +- > > Examine.c | 29 +++++++++- > > Incremental.c | 37 ++++++++++-- > > Kill.c | 19 ++++++ > > ReadMe.c | 1 + > > md_p.h | 74 ++++++++++++++++++++++++ > > mdadm.c | 19 ++++++ > > mdadm.h | 11 +++- > > super1.c | 182 > ++++++++++++++++++++++++++++++++++++++++++++++++++++++++-- > > util.c | 3 +- > > 12 files changed, 422 insertions(+), 29 deletions(-) > > > > -- > > 1.8.1 > > > > -- > > To unsubscribe from this list: send the line "unsubscribe linux-raid" > > in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo > > info at http://vger.kernel.org/majordomo-info.html > -- To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
