Jarkko Kniivilä wrote:
> 
> ----- Original Message -----
> From: "Dan Hollis" <[EMAIL PROTECTED]>
> To: "Jarkko Kniivilä" <[EMAIL PROTECTED]>
> Sent: Thursday, July 20, 2000 1:20 AM
> Subject: Re: 3Ware Escalade 5000 & 6000, opinions?
> 
> > On Thu, 20 Jul 2000, [iso-8859-1] Jarkko Kniivilä wrote:
> > > between Linux (2.2.x) SW RAID and LVM. And then there's the idea that
> maybe
> > > the sophisticated elevator routines that make SCSI drives so much better
> in
> > > JBOD setups are not so much better in a RAID setup if the controller has
> its
> > > own elevator routines.
> >
> > The elevator routines are in the linux kernel.
> >
> 
> Sorry if I used the wrong term but I was referring the routines in the drive
> firmware that respond to command tag queueing and optimally rearrange the
> order of subcommands in the queue. Isn't this elevator term basically the
> same concept, or am I mistaken?
> 
Queue ordering can be done a number of ways to accomplish different ends.
The elevator algorithm describes a method to prevent request starvation.
The first solution to minimize average disk access is to order requests
according to least distance/time to access. However, a burst of requests
in the same area of the disk can constantly defer accesses to another
area. Therefore, the concept of sweeping was introduced i.e. the seek
direction does not change until the last request in that direction is
serviced. This is what the elevator algorithm does. Normally that is 
enough to ensure a reasonable bound on upper access limits. If the 
request rate is high, then requests may have to be serviced in sets to 
ensure timely response.


> > Don't fool yourself, most SCSI drives are the exact same hardware as the
> > IDE drive, just with a different physical interface. But the firmware is
> > largely the same.
> >
> 
> If thats the case why do people buy SCSI if not for the multi-tasking
> scalability in server conditions, ie. the rearrangement of IO requests? I
> had the impression that a SCSI drive has a certain electrical interface not
> much different from IDE (although more scalable) but the main difference was
> in controller PCB on the drive which is housing a more sophisticated
> firmware and possibly a more efficent CPU/DSP supporting the firmware.
> 

One out of three. Vendors are tending toward using the same HDA for both
their IDE and SCSI lines. The interface and interface firmware are quite
different.

> > -Dan
> >
> 
>   // Jarkko

-- 
Dan Jones, Manager, Storage Products          VA Linux Systems
V:(408)542-5737 F:(408)745-9911               1382 Bordeaux Drive
[EMAIL PROTECTED]                            Sunnyvale, CA 94089

Reply via email to