Btrfs RAID-10 performance

Wed Sep 9 18:52:33 EEST 2020

>>>>> "Miloslav" == Miloslav Hůla <miloslav.hula at gmail.com> writes:

Miloslav> Hi, thank you for your reply. I'll continue inline...

Me too... please look for further comments.  Esp about 'fio' and
Netapp useage.

Miloslav> Dne 09.09.2020 v 3:15 John Stoffel napsal(a):
Miloslav> Hello,
Miloslav> I sent this into the Linux Kernel Btrfs mailing list and I got reply:
Miloslav> "RAID-1 would be preferable"
Miloslav> (https://lore.kernel.org/linux-btrfs/7b364356-7041-7d18-bd77-f60e0e2e2112@lechevalier.se/T/).
Miloslav> May I ask you for the comments as from people around the Dovecot?
>> 
>> 
Miloslav> We are using btrfs RAID-10 (/data, 4.7TB) on a physical Supermicro
Miloslav> server with Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz and 125GB of RAM.
Miloslav> We run 'btrfs scrub start -B -d /data' every Sunday as a cron task. It
Miloslav> takes about 50 minutes to finish.
>> 
Miloslav> # uname -a
Miloslav> Linux imap 4.9.0-12-amd64 #1 SMP Debian 4.9.210-1 (2020-01-20) x86_64
Miloslav> GNU/Linux
>> 
Miloslav> RAID is a composition of 16 harddrives. Harddrives are connected via
Miloslav> AVAGO MegaRAID SAS 9361-8i as a RAID-0 devices. All harddrives are SAS
Miloslav> 2.5" 15k drives.
>> 
>> Can you post the output of "cat /proc/mdstat" or since you say you're
>> using btrfs, are you using their own RAID0 setup?  If so, please post
>> the output of 'btrfs stats' or whatever the command is you use to view
>> layout info?

Miloslav> There is a one PCIe RAID controller in a chasis. AVAGO
Miloslav> MegaRAID SAS 9361-8i. And 16x SAS 15k drives conneced to
Miloslav> it. Because the controller does not support pass-through for
Miloslav> the drives, we use 16x RAID-0 on controller. So, we get
Miloslav> /dev/sda ... /dev/sdp (roughly) in OS. And over that we have
Miloslav> single btrfs RAID-10, composed of 16 devices, mounted as
Miloslav> /data.

I will bet that this is one of your bottlenecks as well.  Get a secord
or third controller and split your disks across them evenly.  

Miloslav> We have chosen this wiring for severeal reasons:
Miloslav> - easy to increase a capacity
Miloslav> - easy to replace drives by larger ones
Miloslav> - due to checksuming, btrfs does not need fsck in case of power failure
Miloslav> - btrfs scrub discovers failing drive sooner than S.M.A.R.T. or RAID 
Miloslav> controller

Miloslav> Server serves as a IMAP with Dovecot 2.2.27-3+deb9u6, 4104 accounts,
Miloslav> Mailbox format, LMTP delivery.

>> How ofter are these accounts hitting the server?

Miloslav> IMAP serves for a univesity. So there are typical rush hours from 7AM to 
Miloslav> 3PM. Lowers during the evening, almost not used during the night.

I can understand this, I used to work at a Uni so I can understand the
population needs.  

Miloslav> We run 'rsync' to remote NAS daily. It takes about 6.5 hours to finish,
Miloslav> 12'265'387 files last night.
>> 
>> That's.... sucky.  So basically you're hitting the drives hard with
>> random IOPs and you're probably running out of performance.  How much
>> space are you using on the filesystem?

Miloslav> It's not so sucky how it seems. rsync runs during the
Miloslav> night. And even reading is high, server load stays low. We
Miloslav> have problems with writes.

Ok.  So putting in an SSD pair to cache things should help.  

>> And why not use brtfs send to ship off snapshots instead of using
>> rsync?  I'm sure that would be an improvement...

Miloslav> We run backup to external NAS (NetApp) for a disaster
Miloslav> recovery scenario.  Moreover NAS is spreaded across multiple
Miloslav> locations. Then we create NAS snapshot, tens days
Miloslav> backward. All snapshots easily available via NFS mount. And
Miloslav> NAS capacity is cheaper.

So why not run the backend storage on the Netapp, and just keep the
indexes and such local to the system?  I've run Netapps for many years
and they work really well.  And then you'd get automatic backups using
schedule snapshots.

Keep the index files local on disk/SSDs and put the maildirs out to
NFSv3 volume(s) on the Netapp(s).  Should do wonders.  And you'll stop
needing to do rsync at night. 

Miloslav> Last half year, we encoutered into performace
Miloslav> troubles. Server load grows up to 30 in rush hours, due to
Miloslav> IO waits. We tried to attach next harddrives (the 838G ones
Miloslav> in a list below) and increase a free space by rebalace. I
Miloslav> think, it helped a little bit, not not so rapidly.

>> If you're IOPs bound, but not space bound, then you *really* want to
>> get an SSD in there for the indexes and such.  Basically the stuff
>> that gets written/read from all the time no matter what, but which
>> isn't large in terms of space.

Miloslav> Yes. We are now on 66% capacity. Adding SSD for indexes is
Miloslav> our next step.

This *should* give you a boost in performance.  But finding a way to
take before and after latency/performance measurements is key.  I
would look into using 'fio' to test your latency numbers.  You might
also want to try using XFS or even ext4 as your filesystem.  I
understand not wanting to 'fsck', so that might be right out.

Which leads me back to suggesting you use the Netapp as your primary
storage, assuming the Netapp isn't bogged down with other work.
Again, use 'fio' to run some tests and see how things look.

>> Also, adding in another controller card or two would also probably
>> help spread the load across more PCI channels, and reduce contention
>> on the SATA/SAS bus as well.

Miloslav> Probably we will wait how SSD helps first, but as you wrote, it is 
Miloslav> possible next step.

Miloslav> Is this a reasonable setup and use case for btrfs RAID-10?
Miloslav> If so, are there some recommendations to achieve better
Miloslav> performance?
>> 
>> 1. move HOT data to SSD based volume RAID 1 pair.  On a seperate
>> controller.

Miloslav> OK

>> 2. add more controllers, which also means you're more redundant in
>> case one controller fails.

Miloslav> OK

>> 3. Clone the system and put Dovecot IMAP director in from of the
>> setup.

Miloslav> I still hope that one server can handle 4105 accounts.

>> 4. Stop using rsync for copying to your DR site, use the btrfs snap
>> send, or whatever the commands are.

Miloslav> I hope it is not needed in our scenario.

>> 5. check which dovecot backend you're using and think about moving to
>> one which doesn't involve nearly as many files.

Miloslav> Maildir is comfortable for us. Time to time, users call us with: "I 
Miloslav> accidentally deleted the folder" and it is super easy to copy it back 
Miloslav> from backup.

>> 6. Find out who your biggest users are, in terms of emails and move
>> them to SSDs if step 1 is too hard to do at first.

Miloslav> OK

>> Can you also grab some 'iostat -dhm 30 60'  output, which is 30
>> minutes of data over 30 second intervals?  That should help you narrow
>> down which (if any) disk is your hotspot.

Miloslav> OK, thanks for the tip.

>> It's not clear to me if you have one big btrfs filesystem, or a bunch
>> of smaller ones stiched together.  In any case, it should be very easy
>> to get better performance here.

Miloslav> I hope I've made it clear above.

>> I think someone else mentioned that you should look at your dovecot
>> backend, and you should move to the fastest one you can find.
>> 
>> Good luck!
>> John

Miloslav> Thank you for your time and advices!

Miloslav> Kind regards
Miloslav> Milo

Miloslav> # megaclisas-status
Miloslav> -- Controller information --
Miloslav> -- ID | H/W Model                  | RAM    | Temp | BBU    | Firmware
Miloslav> c0    | AVAGO MegaRAID SAS 9361-8i | 1024MB | 72C  | Good   | FW:
Miloslav> 24.16.0-0082
>> 
Miloslav> -- Array information --
Miloslav> -- ID | Type   |    Size |  Strpsz | Flags | DskCache |   Status |  OS
Miloslav> Path | CacheCade |InProgress
Miloslav> c0u0  | RAID-0 |    838G |  256 KB | RA,WB |  Enabled |  Optimal |
Miloslav> /dev/sdq | None      |None
Miloslav> c0u1  | RAID-0 |    558G |  256 KB | RA,WB |  Enabled |  Optimal |
Miloslav> /dev/sda | None      |None
Miloslav> c0u2  | RAID-0 |    558G |  256 KB | RA,WB |  Enabled |  Optimal |
Miloslav> /dev/sdb | None      |None
Miloslav> c0u3  | RAID-0 |    558G |  256 KB | RA,WB |  Enabled |  Optimal |
Miloslav> /dev/sdc | None      |None
Miloslav> c0u4  | RAID-0 |    558G |  256 KB | RA,WB |  Enabled |  Optimal |
Miloslav> /dev/sdd | None      |None
Miloslav> c0u5  | RAID-0 |    558G |  256 KB | RA,WB |  Enabled |  Optimal |
Miloslav> /dev/sde | None      |None
Miloslav> c0u6  | RAID-0 |    558G |  256 KB | RA,WB |  Enabled |  Optimal |
Miloslav> /dev/sdf | None      |None
Miloslav> c0u7  | RAID-0 |    558G |  256 KB | RA,WB |  Enabled |  Optimal |
Miloslav> /dev/sdg | None      |None
Miloslav> c0u8  | RAID-0 |    558G |  256 KB | RA,WB |  Enabled |  Optimal |
Miloslav> /dev/sdh | None      |None
Miloslav> c0u9  | RAID-0 |    558G |  256 KB | RA,WB |  Enabled |  Optimal |
Miloslav> /dev/sdi | None      |None
Miloslav> c0u10 | RAID-0 |    558G |  256 KB | RA,WB |  Enabled |  Optimal |
Miloslav> /dev/sdj | None      |None
Miloslav> c0u11 | RAID-0 |    558G |  256 KB | RA,WB |  Enabled |  Optimal |
Miloslav> /dev/sdk | None      |None
Miloslav> c0u12 | RAID-0 |    558G |  256 KB | RA,WB |  Enabled |  Optimal |
Miloslav> /dev/sdl | None      |None
Miloslav> c0u13 | RAID-0 |    558G |  256 KB | RA,WB |  Enabled |  Optimal |
Miloslav> /dev/sdm | None      |None
Miloslav> c0u14 | RAID-0 |    558G |  256 KB | RA,WB |  Enabled |  Optimal |
Miloslav> /dev/sdn | None      |None
Miloslav> c0u15 | RAID-0 |    838G |  256 KB | RA,WB |  Enabled |  Optimal |
Miloslav> /dev/sdr | None      |None
>> 
Miloslav> -- Disk information --
Miloslav> -- ID   | Type | Drive Model                       | Size     | Status
Miloslav> | Speed    | Temp | Slot ID  | LSI ID
Miloslav> c0u0p0  | HDD  | SEAGATE ST900MP0006 N003WAG0Q3S3  | 837.8 Gb | Online,
Miloslav> Spun Up | 12.0Gb/s | 53C  | [8:14]   | 32
Miloslav> c0u1p0  | HDD  | HGST HUC156060CSS200 A3800XV250TJ | 558.4 Gb | Online,
Miloslav> Spun Up | 12.0Gb/s | 38C  | [8:0]    | 12
Miloslav> c0u2p0  | HDD  | HGST HUC156060CSS200 A3800XV3XT4J | 558.4 Gb | Online,
Miloslav> Spun Up | 12.0Gb/s | 43C  | [8:1]    | 11
Miloslav> c0u3p0  | HDD  | HGST HUC156060CSS200 ADB05ZG4XLZU | 558.4 Gb | Online,
Miloslav> Spun Up | 12.0Gb/s | 46C  | [8:2]    | 25
Miloslav> c0u4p0  | HDD  | HGST HUC156060CSS200 A3800XV3DWRL | 558.4 Gb | Online,
Miloslav> Spun Up | 12.0Gb/s | 48C  | [8:3]    | 14
Miloslav> c0u5p0  | HDD  | HGST HUC156060CSS200 A3800XV3XZTL | 558.4 Gb | Online,
Miloslav> Spun Up | 12.0Gb/s | 52C  | [8:4]    | 18
Miloslav> c0u6p0  | HDD  | HGST HUC156060CSS200 A3800XV3VSKJ | 558.4 Gb | Online,
Miloslav> Spun Up | 12.0Gb/s | 55C  | [8:5]    | 15
Miloslav> c0u7p0  | HDD  | SEAGATE ST600MP0006 N003WAF1LWKE  | 558.4 Gb | Online,
Miloslav> Spun Up | 12.0Gb/s | 56C  | [8:6]    | 28
Miloslav> c0u8p0  | HDD  | HGST HUC156060CSS200 A3800XV3XTDJ | 558.4 Gb | Online,
Miloslav> Spun Up | 12.0Gb/s | 55C  | [8:7]    | 20
Miloslav> c0u9p0  | HDD  | HGST HUC156060CSS200 A3800XV3T8XL | 558.4 Gb | Online,
Miloslav> Spun Up | 12.0Gb/s | 57C  | [8:8]    | 19
Miloslav> c0u10p0 | HDD  | HGST HUC156060CSS200 A7030XHL0ZYP | 558.4 Gb | Online,
Miloslav> Spun Up | 12.0Gb/s | 61C  | [8:9]    | 23
Miloslav> c0u11p0 | HDD  | HGST HUC156060CSS200 ADB05ZG4VR3P | 558.4 Gb | Online,
Miloslav> Spun Up | 12.0Gb/s | 60C  | [8:10]   | 24
Miloslav> c0u12p0 | HDD  | SEAGATE ST600MP0006 N003WAF195KA  | 558.4 Gb | Online,
Miloslav> Spun Up | 12.0Gb/s | 60C  | [8:11]   | 29
Miloslav> c0u13p0 | HDD  | SEAGATE ST600MP0006 N003WAF1LTZW  | 558.4 Gb | Online,
Miloslav> Spun Up | 12.0Gb/s | 56C  | [8:12]   | 26
Miloslav> c0u14p0 | HDD  | SEAGATE ST600MP0006 N003WAF1LWH6  | 558.4 Gb | Online,
Miloslav> Spun Up | 12.0Gb/s | 55C  | [8:13]   | 27
Miloslav> c0u15p0 | HDD  | SEAGATE ST900MP0006 N003WAG0Q414  | 837.8 Gb | Online,
Miloslav> Spun Up | 12.0Gb/s | 47C  | [8:15]   | 33
>> 
>> 
>> 
Miloslav> # btrfs --version
Miloslav> btrfs-progs v4.7.3
>> 
>> 
>> 
Miloslav> # btrfs fi show
Miloslav> Label: 'DATA'  uuid: 5b285a46-e55d-4191-924f-0884fa06edd8
Miloslav> Total devices 16 FS bytes used 3.49TiB
Miloslav> devid    1 size 558.41GiB used 448.66GiB path /dev/sda
Miloslav> devid    2 size 558.41GiB used 448.66GiB path /dev/sdb
Miloslav> devid    4 size 558.41GiB used 448.66GiB path /dev/sdd
Miloslav> devid    5 size 558.41GiB used 448.66GiB path /dev/sde
Miloslav> devid    7 size 558.41GiB used 448.66GiB path /dev/sdg
Miloslav> devid    8 size 558.41GiB used 448.66GiB path /dev/sdh
Miloslav> devid    9 size 558.41GiB used 448.66GiB path /dev/sdf
Miloslav> devid   10 size 558.41GiB used 448.66GiB path /dev/sdi
Miloslav> devid   11 size 558.41GiB used 448.66GiB path /dev/sdj
Miloslav> devid   13 size 558.41GiB used 448.66GiB path /dev/sdk
Miloslav> devid   14 size 558.41GiB used 448.66GiB path /dev/sdc
Miloslav> devid   15 size 558.41GiB used 448.66GiB path /dev/sdl
Miloslav> devid   16 size 558.41GiB used 448.66GiB path /dev/sdm
Miloslav> devid   17 size 558.41GiB used 448.66GiB path /dev/sdn
Miloslav> devid   18 size 837.84GiB used 448.66GiB path /dev/sdr
Miloslav> devid   19 size 837.84GiB used 448.66GiB path /dev/sdq
>> 
>> 
>> 
Miloslav> # btrfs fi df /data/
Miloslav> Data, RAID10: total=3.48TiB, used=3.47TiB
Miloslav> System, RAID10: total=256.00MiB, used=320.00KiB
Miloslav> Metadata, RAID10: total=21.00GiB, used=18.17GiB
Miloslav> GlobalReserve, single: total=512.00MiB, used=0.00B
>> 
>> 
>> 
Miloslav> I do not attach whole dmesg log. It is almost empty, without errors.
Miloslav> Only lines about BTRFS are about relocations, like:
>> 
Miloslav> BTRFS info (device sda): relocating block group 29435663220736 flags 65
Miloslav> BTRFS info (device sda): found 54460 extents
Miloslav> BTRFS info (device sda): found 54459 extents
>> 