100 --> 200 is 100% improvement, but 200 --> 100 is only 50% reduction.

> -----Original Message-----
> From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]
> Sent: Monday, July 12, 1999 5:49 AM
> To: [EMAIL PROTECTED]
> Cc: [EMAIL PROTECTED]
> Subject: Re: Two questions about performance
>
>
> "Roeland M.J. Meyer" wrote:
> >
> > > [mailto:[EMAIL PROTECTED]]On Behalf Of
> Juan Piernas
> > > C�novas
> > > Sent: Saturday, July 10, 1999 4:04 PM
> > >
> > > I have two questions:
> > >
> > > 1.- Why are RAID-1 writes slower than writes to one of the
> > > RAID-1 disks?. Software overhead?. Hardware problems?.
> >
> > For every write request, two physical writes are performed,
> under RAID1.
> > One write to each mirror. Under SCSI, with asynchronous
> access, this can
> > be overlapped to some degree, but not entirely in parallel.
> Under IDE,
> > parallel writes are next to impossible and even
> asynchronous writes are
> > difficult. This is somewhat ameliorated with multiple disk
> controllers,
> > evenso not entirely. The end result is that write times are
> 25% to 100%
> > longer in duration, depending on hardware configuration..
>
> I see.
>
> >
> > > 2.- Why are RAID-1 reads slightly better than reads from one
> > > of the RAID-1 disks?. If a RAID-1 is made up of two disks,
> > > I've got two copies of data and two request can be made in
> > > parallel. Why isn't read performance much better?. With
> > > RAID-0 and a single file I get nearly a two fold
> > > performance.
> >
> > RAID0 is faster becasue of overlapped IO, on a striped volume-set,
> > theoretically. This would not be true with two RAID0 disks,
> on the same
> > IDE bus, becasue IDE can not overlap IO. With RAID1, on
> SCSI or separate
> I know.
>
> > IDE controllers, there should be a 0% to 40% improvement, in read
> > access, due to over-lapped IO, depending on the exact hardware
> > configuration.
> Yes but, why don't you have a 70% to 80% improvement like with RAID0?.
> You also have overlapped IO, don't you?.
>
> >
> > I hope this helps.
> Yes, thanks.
> --
> D. Juan Piernas C�novas
> Departamento de Ingenier�a y Tecnolog�a de Computadores
> Facultad de Inform�tica. Universidad de Murcia
> Campus de Espinardo - 30080 Murcia (SPAIN)
> Tel.: +34968364633    Fax: +34968364151
> email: [EMAIL PROTECTED]
>

Reply via email to