xfs
[Top] [All Lists]

Re: Q: Filesystem block sizes available?

Subject: Re: Q: Filesystem block sizes available?
From: "D. Stimits" <stimits@xxxxxxxxxx>
Date: Tue, 13 Nov 2001 11:23:16 -0700
Cc: XFS Mailing list <linux-xfs@xxxxxxxxxxx>
References: <Pine.LNX.4.30.0111131903400.2151-100000@mustard.heime.net>
Reply-to: stimits@xxxxxxxxxx
Sender: owner-linux-xfs@xxxxxxxxxxx
Roy Sigurd Karlsbakk wrote:
> 
> > What chipset is the motherboard? And do you have IO-APIC enabled? See if
> > irq's appear under both cpu's in /proc/interrupts...if they are spread
> > out on both cpu's it is enabled.
> 
> Chipset:
>   Host bridge: ServerWorks CNB20HE + +

Known for 64 bit bus.

>   PCI Bridge Intel 80960 RP
> 
> Interrupts:
>            CPU0       CPU1
>   0:      37553      37575    IO-APIC-edge  timer
>   1:        674        624    IO-APIC-edge  keyboard
>   2:          0          0          XT-PIC  cascade
>   5:     109107     106066   IO-APIC-level  cciss0
>   9:          0          0   IO-APIC-level  acpi
>  11:        969        967   IO-APIC-level  e1000
>  12:         14         21    IO-APIC-edge  PS/2 Mouse
>  15:          1          0   IO-APIC-level  cpqphp.o
> NMI:          0          0
> LOC:      75032      75031
> ERR:          0
> MIS:          0

APIC is definitely enabled.

> 
> > Now you mention a high memory bandwidth, but is this still a standard
> > PCI slot (thus 33 MHz/32 bit)? If so, it'd be kind of like putting a
> > fast AGP graphics card in an ordinary PCI slot...it would be a
> > bottleneck between the higher bandwidth and the card.
> 
> It's using two sdram dimms in parallel on the memory bus. It's got one
> 33/32 PCI bus, one 33/64, and one 66/64 bus. I really can't find any
> bottleneck apart from the disks seeking like idiots.

It isn't unusual for any of these busses to try and autosense if
standard 32 bit/33 MHz cards are in a slot, and scale back the speed of
the entire bus to 33 MHz for anything on it. This certainly has the
ability to pump out a lot of bandwidth on any U160 or RAID controller,
probably the only thing I'd want to double check on is if a slower card
is cutting the whole bus back (I'm guessing dmesg startup will give that
information, but I'm not positive).

Now if these disks are thrashing around a lot, it seems that something
else must be going on. Have you tried checking performance and thrashing
behavior if the disks are formatted and used separately, versus some
form of RAID or logical volume management? I'm also curious what the
chunk size is? (chunk size != block size). If chunk size is too small,
it might cause thrashing. If it is rather large, you won't really be
splitting writes and reads across multiple striped disks, it'd be a
series of small writes alternating between disks in that case.

D. Stimits, stimits@xxxxxxxxxx

> 
> >
> > D. Stimits, stimits@xxxxxxxxxx
> >
> > >
> > > --
> > > Roy Sigurd Karlsbakk, MCSE, MCNE, CLS, LCA
> > >
> > > Computers are like air conditioners.
> > > They stop working when you open Windows.
> >
> 
> --
> Roy Sigurd Karlsbakk, MCSE, MCNE, CLS, LCA
> 
> Computers are like air conditioners.
> They stop working when you open Windows.


<Prev in Thread] Current Thread [Next in Thread>