[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Bug#462617: debian-installer: need RAID bitmap support (mdadm -b internal)



On Sat, Jan 26, 2008 at 06:44:59PM +0700, martin f krafft wrote:
> also sprach Otavio Salvador <otavio@debian.org> [2008.01.26.1834 +0700]:
> > Please Martin, could you comment on that?
> 
> I have: http://lists.debian.org/debian-devel/2008/01/msg00929.html

Joerg Jaspert did some tests to measure the impact of write
intent bitmaps on performances.  Quoting his blog entry: [1]  

   So, having turned them on I just waited to see what happens with the
   machines under normal load. Not doing any stresstests, as they don’t
   show normal load situations anyway. And a little performance impact
   was expected. A little, not the thing that happened…

   I’m impressed, haven’t seen that much IO wait ever before. Even the
   old packages.debian.org code, which was really bad IO-wise (thanks to
   multiple greps per search), didn’t have such a bad impact. I mean - a
   Dual Core Athlon 3800, running a few Xen domains, usually without any
   noticable load, now having a load of 20 and more, while IO wait is at
   some 90% (as shown by top for both cores), constantly. (2 RAID1
   devices).

   Or a Desktop System with a RAID5 (mad of 4 SATA discs), which
   normally unpacks a kernel tarball in some 10 seconds - now needs
   about 2 minutes to get it unpacked. Removing that tree now uses 5
   minutes.

   No, bitmap internal is nothing one ever wants to use. One possibly
   might want to use that feature with a file for the bitmaps which is
   stored on a different device. That might actually make sense and wont
   hurt as much as internal does.

   I have one positive thing to say about them: Resyncing an array is
   indeed much much faster. But I go with longer resync times and
   slightly less performance during that time, the cost of having the
   bitmaps all the time massively impacts, a recovery of even a day
   doesn’t cost as much in total. 

I would be in favor of trusting Joerg's comments and tagging this bug as
"wontfix" in order to document the issue.  Any other opinions?

[1] http://blog.ganneff.de/blog/2008/01/30#mdraid_bitmap_internal_bad

Cheers,
-- 
Jérémy Bobbio                        .''`. 
lunar@debian.org                    : :Ⓐ  :  # apt-get install anarchism
                                    `. `'` 
                                      `-   

Attachment: signature.asc
Description: Digital signature


Reply to: