[DRBD-user] deadlock? NFS export DRBD on loop device on tmpfs

[HIROSE, Masaaki]

hello,

in "Re: [DRBD-user] deadlock? NFS export DRBD on loop device on tmpfs"
   <yf11x58lf1l.wl%hirose31 at t3.rim.or.jp>
at Fri, 05 Aug 2005 21:54:46 +0900,
   hirose31 at t3.rim.or.jp wrote:

> Additionaly question and ask advice.
> 
> 1 Why DRBD deadlock on loopback device? Does fsync() cause deadlock?
> 
> 2 Is there any idea that DRBD on huge (1G-12GB) memory base device?
> 
>   NG tmpfs   - tmpfs is not block device.
>   NG ramfs   - ramfs is not block device.
>   NG ramdisk - too small. (up to 512MB x 16 devices?)
> 
>   NG loseup + tmpfs - cause deadlock.
>   NG loseup + ramfs - cause deadlock.

I tried following.

>   ?? LVM + many ramdisk - create big block device with join many
>                           ramdisk(/dev/ramX).

NG: cannot pvcreate /dev/ramX.

  # pvcreate -vd /dev/ram2
      Device /dev/ram2 not found.

>   ?? MD  + many ramdisk - ditto.

NG: System down when write big data to /dev/md0.

  boot with ramdisk_size=524288.
  mdadm -C /dev/md0 -l0 -n2 /dev/ram2 /dev/ram3
  mdadm -D /dev/md0
    (seems to be 512MB+512MB=1GB RAID0 device.)
  mkfs.xfs -f -s size=1024 -b size=1024 -i size=512,maxpct=0 /dev/md0
  mount /dev/md0 /mnt/md0
    (mount device successfully)
  dd if=/dev/zero of=/mnt/md0 bs=1M couint=700
    (no problem)
  dd if=/dev/zero of=/mnt/md0 bs=1M couint=1500
    (system go down...)

>   ?? other idea??

maybe OK: DRBD on /dev/ram1 (512MB)

  I has done stress-test (repeat dd and bonnie++) over night. It seems
  good. 
  Is there a possibility of serious problem on DRBD on /dev/ramX?



> in "Re: [DRBD-user] deadlock? NFS export DRBD on loop device on tmpfs"
>    <20050804053619.GA13037 at barkeeper1.linbit>
> at Thu, 4 Aug 2005 07:36:19 +0200,
>    Lars.Ellenberg at linbit.com wrote:
> 
> > > I want memory storage NFS server, so using NFS server that export XFS
> > > on DRBD on loop device on tmpfs.
> > > 
> > > This system looks like working well, but sometimes load average
> > > increase too high (up to 20) suddenly and system is hung up.
> > > 
> > > [question]
> > > 1 Someone (on Internet) say that DRBD on loop device cause
> > >   deadlock. Is it true?
> > 
> > yes.
> > 
> > > 
> > > 2 What mean "top of loop device" on DRBD wiki page? loop block device?
> > >   or network loop device?
> > >     DataRedundancyByDrbd / Some Do's and Don'ts
> > >     <URL:http://wiki.linux-ha.org/DataRedundancyByDrbd#head-25da1eb596fac86b9b39073f3a9566007cc6a505>
> > 
> > from the context, it should be clear that loop block dev is meant.
> > 
> > > 3 Does DRBD work well on loop device on tmpfs?
> > 
> > no
> > 
> > > Is there a possibility of deadlock or serious problem?
> > 
> > yes

-- 
HIROSE, Masaaki