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

Re: [Nbd] NBD Super Size io operations



Hi,

the purpose of my patch was to make the max_sectors_kb value
_adjustable_, not necessarily changing its default value although 4KB
aligned would be a bit better than 127KB...
Currently there is no way of raising the default as the
max_hw_sectors_kb is also 127KB...
Benchmarks results may vary and they do -- in my local env using my old
SSD I was getting better performance with max_sectors_kb=512 but my new
SSD shows better performance with max_sectors_kb=128 and the tests over
the network with much higher latencies may vary even more than that...

Paul, is there a chance for this and _the_ ERESTARTSYS patches got into
3.9?

On Wed, Apr 03, 2013 at 09:36:34PM +0200, folkert wrote:
> People,
> 
> I did it wrong.
> 
> This is correct:
> 
> previously version (stock 3.8 with signal patch) with 4KB nbd, 128KB iozone3:
>                                                             random  random    bkwd   record   stride
>               KB  reclen   write rewrite    read    reread    read   write    read  rewrite     read   fwrite frewrite   fread  freread
>          2097152     128   37035   38578   168336   142210  138569   36966  110692    31541   143964    53846    53114  122773   124531
> 
> current version with 4KB nbd, 128KB iozone3:
>                                                             random  random    bkwd   record   stride
>               KB  reclen   write rewrite    read    reread    read   write    read  rewrite     read   fwrite frewrite   fread  freread
>          2097152     128   33933   35063   116522   117882  114987   34607  116599    36000   116012    55279    55677  116960   116745
> 
> 
> current version with 128KB nbd, 128KB iozone3:
>                                                             random  random    bkwd   record   stride
>               KB  reclen   write rewrite    read    reread    read   write    read  rewrite     read   fwrite frewrite   fread  freread
>          2097152     128   35664   35674   123922   122845  126486   36151  112671    35884   118069    53315    59279  115894   117202
> 
> current version with 128KB nbd, 1024KB iozone3:
>                                                             random  random    bkwd   record   stride
>               KB  reclen   write rewrite    read    reread    read   write    read  rewrite     read   fwrite frewrite   fread  freread
>          2097152    1024   33705   35078   131637   136712  136685   35152  139128    38010   148841    59181    57372  121299   113432
> 
> current version with 1024KB nbd, 1024KB iozone3:
>                                                             random  random    bkwd   record   stride
>               KB  reclen   write rewrite    read    reread    read   write    read  rewrite     read   fwrite frewrite   fread  freread
>          2097152    1024   34208   33322   138010   136586  136461   35383  139139    36043   131382    55621    57448  116226   117450
> 
> so I don't see any logic in it
> 
> 
> On Tue, Apr 02, 2013 at 04:07:07PM +0200, folkert wrote:
> > I've applied the patch but I don't think I see any gain?
> > 
> > now (= with that patch) with block size 1024KB:
> >                                                             random  random    bkwd   record   stride
> >               KB  reclen   write rewrite    read    reread    read   write    read  rewrite     read   fwrite frewrite   fread  freread
> >          2097152    1024   35080   35989   135806   132713  134109   35615  133128    35345   139671    57022    58270  125159   131571
> > 
> > now (= with that patch) with block size 128KB:
> >                                                             random  random    bkwd   record   stride
> >               KB  reclen   write rewrite    read    reread    read   write    read  rewrite     read   fwrite frewrite   fread  freread
> >          2097152     128   36651   35484   112819   113962  114233   34810  113014    35782   119221    55876    54225  115214   117897
> > 
> > previously version (stock 3.8 with signal patch):
> >                                                             random  random    bkwd   record   stride
> >               KB  reclen   write rewrite    read    reread    read   write    read  rewrite     read   fwrite frewrite   fread  freread
> >          2097152     128   37035   38578   168336   142210  138569   36966  110692    31541   143964    53846    53114  122773   124531
> > 
> > 
> > Tested using nbd-server on ramdisk, client in a vm (kernel 3.8, see above), tested with iozone3 from debian testing:
> > 	iozone -j 1024 -s 2g -r 128k -I -f test -+w 7
> > 
> > Test system is a Linux system with kernel 2.6.38 and vmware workstation 9.
> > It has a Q9450 running at 2.1GHz and 8GB of ram with 512MB of ram for the vm.
> > 
> > 
> > On Mon, Apr 01, 2013 at 10:25:23PM +0200, Michal Belczyk wrote:
> > > This looks like my _other_ patch I mentioned last time (*), except that the
> > > suggested default value is different, doesn't it?
> > > 
> > > (*) http://www.mail-archive.com/nbd-general@lists.sourceforge.net/msg01314.html
> > > 
> > > On Mon, Apr 01, 2013 at 02:41:43PM -0400, Paul Clements wrote:
> > > > I'm looking at this right now. I just need to test and decide what the
> > > > default should be, and I'll send this upstream.
> > > > 
> > > > Thanks,
> > > > Paul
> > > > 
> > > > On Mon, Apr 1, 2013 at 1:45 PM, Rohan Sharma <rickyrohan14@...58...> wrote:
> > > > > My last message didn?t show up properly due to some formatting errors.
> > > > > So, I am sending it again
> > > > >
> > > > > I am using nbd over high latency network (100 ms ).
> > > > > Currently maximum size of nbd io operation(amount of data fetched in one
> > > > > request)
> > > > > I can achieve is 130kb and it gives very poor disk performance.
> > > > > I tried recompiling nbd driver code so as to increase max_sector_kb property
> > > > > of disk queue.
> > > > > I added following two lines to nbd_init function
> > > > >               blk_queue_max_hw_sectors(disk->queue, 65536);
> > > > >               disk->queue->limits.max_sectors = 4096;
> > > > > But these modifications changed nothing.
> > > > > Do you have any idea what can I do to increase size of nbd io operations?
> > > > > I would like it to go as high as 4MB.
> > > > >
> > > > > Thanks
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > ------------------------------------------------------------------------------
> > > > > Own the Future-Intel? Level Up Game Demo Contest 2013
> > > > > Rise to greatness in Intel's independent game demo contest.
> > > > > Compete for recognition, cash, and the chance to get your game
> > > > > on Steam. $5K grand prize plus 10 genre and skill prizes.
> > > > > Submit your demo by 6/6/13. http://p.sf.net/sfu/intel_levelupd2d
> > > > > _______________________________________________
> > > > > Nbd-general mailing list
> > > > > Nbd-general@lists.sourceforge.net
> > > > > https://lists.sourceforge.net/lists/listinfo/nbd-general
> > > > >
> > > > 
> > > > ------------------------------------------------------------------------------
> > > > Own the Future-Intel&reg; Level Up Game Demo Contest 2013
> > > > Rise to greatness in Intel's independent game demo contest.
> > > > Compete for recognition, cash, and the chance to get your game 
> > > > on Steam. $5K grand prize plus 10 genre and skill prizes. 
> > > > Submit your demo by 6/6/13. http://p.sf.net/sfu/intel_levelupd2d
> > > > _______________________________________________
> > > > Nbd-general mailing list
> > > > Nbd-general@lists.sourceforge.net
> > > > https://lists.sourceforge.net/lists/listinfo/nbd-general
> > > 
> > > -- 
> > > Michal Belczyk Sr.
> > > 
> > > ------------------------------------------------------------------------------
> > > Own the Future-Intel&reg; Level Up Game Demo Contest 2013
> > > Rise to greatness in Intel's independent game demo contest.
> > > Compete for recognition, cash, and the chance to get your game 
> > > on Steam. $5K grand prize plus 10 genre and skill prizes. 
> > > Submit your demo by 6/6/13. http://p.sf.net/sfu/intel_levelupd2d
> > > _______________________________________________
> > > Nbd-general mailing list
> > > Nbd-general@lists.sourceforge.net
> > > https://lists.sourceforge.net/lists/listinfo/nbd-general
> > 
> > 
> > Folkert van Heusden
> > 
> > -- 
> > www.vanheusden.com/multitail - multitail is tail on steroids. multiple
> >                windows, filtering, coloring, anything you can think of
> > ----------------------------------------------------------------------
> > Phone: +31-6-41278122, PGP-key: 1F28D8AE, www.vanheusden.com
> > 
> > ------------------------------------------------------------------------------
> > Own the Future-Intel(R) Level Up Game Demo Contest 2013
> > Rise to greatness in Intel's independent game demo contest. Compete 
> > for recognition, cash, and the chance to get your game on Steam. 
> > $5K grand prize plus 10 genre and skill prizes. Submit your demo 
> > by 6/6/13. http://altfarm.mediaplex.com/ad/ck/12124-176961-30367-2
> > _______________________________________________
> > Nbd-general mailing list
> > Nbd-general@lists.sourceforge.net
> > https://lists.sourceforge.net/lists/listinfo/nbd-general
> 
> 
> Folkert van Heusden
> 
> -- 
> Multi tail barnamaj mowahib li mora9abat attasjilat wa nataij awamir
> al 7asoub. damj, talwin, mora9abat attarchi7 wa ila akhirih.
> http://www.vanheusden.com/multitail/
> ----------------------------------------------------------------------
> Phone: +31-6-41278122, PGP-key: 1F28D8AE, www.vanheusden.com
> 

-- 
Michal Belczyk Sr.



Reply to: