"Dean" <dean.sun@gmail.com> schrieb im Newsbeitrag
news:a2efdbbd.0411112105.d210392@posting.google.co m...
> I did a write IO test in IBM P690 and HP SUPERDOME.
> When I tested the local disks in P690,use this:
> ----------------------------------------------------
> time dd if=/dev/zero of=/tmp/test.dat bs=1024 count=300000
> 300000+0 records in
> 300000+0 records out
> real 0m42.01s
> user 0m0.85s
> sys 0m6.18s
> The Speed should be 7MB/s almost,and the P690 is running none except
> this.
>
> When I tested the local disks in the SUPERDOME,
> ---------------------------------------------
> time dd if=/dev/zero of=/tmp/test.dat bs=1024 count=300000
> 300000+0 records in
> 300000+0 records out
>
> real 2.9
> user 0.2
> sys 2.5
> The Speed should be 100MB/s almost!!!!
> =======================================
> I used the same method to test the Storage of HDS 9980,which is
> connected to the P690 and SUPERDOME.
> -----------------------------
> the result of P690 is 56MB/s almost,and result of SUPERDOME is 90MB/s
> almost !!!
>
> I'll cry for explaining this to the customer.
> Anyone can help me to explain this?

Hallo Dean,

I would guess that the difference of the first comparison can be explained
by caching. 7MB/s sounds 'normal' for accessing one or two SCSI disks
depending on lv layout. 100MB/s can not be done with the same number of
physical disks.
Second comparison looks to me like you where writing to 7 or 8 physical
disks that would result in a throughput of 56MB/s (for serial I/O) while
90MB/s might have to do with caching again (i.e. different handling of this
I/O).

These values for I/O can differ to a great extent depending on lv layout and
caching. E.g. I have seen p690 writing to 6+p hdisks RAID5 devices 100% busy
and 2MB/s (!) I/O (because of some bad misconfiguration) as well as 75% busy
with 56MB/s on the same device after the config was fixed. This was almost
the maximum (constant) I/O the physical disks could handle. When the amount
of data fitted into the fastwrite cache the peak value almost doubled. Can
you see the picture?
Now, if I wanted to compare p690 with Superdome I would either use no
caching at all (neither on the operating system layer nor on the hardware
layer) for the comparison or have each system fine tuned by some dedicated
expert. The second variation seems more appropriate to me actually. Only
this way you can be sure that you do not compare apples with pears. Just my
5c.

Regards,
Andreas