On Sat, Nov 13, 1999 at 02:54:33PM +0100, Miquel van Smoorenburg wrote:
> In article <[EMAIL PROTECTED]>,
> Brian D. Haymore <[EMAIL PROTECTED]> wrote:
> >On Thu, 11 Nov 1999, Matthew Clark wrote:
> >> Which gives better performance on Linux???
> >Mylex does from my tests. Mylex does very well in fact.
I've got some figures testing both mylex and ami megaraid.
The mylex is slightly faster. (Figures on the way).
> Also, very important, the mylex driver can be controlled through
> the /proc filesystem. The MegaRaid comes with a utility that AFAICS
That's a plus (about the mylex). But I can do anything with the
ami using ami-ioctls, thus not using their mgr.
> simply hooks into the BIOS to run the BIOS setup screen. And while
> it somewhat works, it's dead slow when not run on the console and
> you can 't script it or use it for automatic diagnostics.
Nope, it does not hook into the bios. But it is the same program.
And it is not well designed:
1) Uses obsolete SCSI_SEND_COMMAND
2) uses /dev/sdx instead of /dev/sgx. Very nice if you don't have
disks.
3) You can't do the things you like (init array)
4) You need sunglasses.
BUT the driver size is only 15k...
Anyway: using software raid-5 on a PII-450 is 5 to 10 times as fast
as using raid-1 on these controllers.
Linux averages about 20 MB/s reading and writing raid 5 with 3
scsi-disks using the ncr53c8xx driver for the sym53c895.
raid-1 on the ami and the mega averages about 2.5MB/s to 3MB/s.
The myles is not much faster.
> Also, we've seen several hangs with the MegaRaid when one disk
> is malfunctioning in a RAID5 array which is unforgivable since
> that defeats the whole purpose of a RAID5 array.
>
> [specifically we've seen that some SCSI disks crash in such a way
> that they claim to have a 0GB size - they get taken offline nicely,
> but after a reboot it takes ages to get through the BIOS and
> the Linux driver hangs indefinitely]
>
> Also I've seen the MegaRaid get stuck in an endless SCSI reset loop.
> Increasing the Linux SCSI layer timeout (got included with 2.2.13)
> seems to have fixed it but I still think it's strange
1) These seem to be fixed in their latest firmware upgrade. I've witnessed
it too. Would not boot with a disk that fell on the ground from 1 meter...
2) Turning P&Pray OS on the BIOS made my configuration BARF before even
being able to load the BOOT sector!
The reason for ami: you can issue ami_ioctl commands from within solaris,
since there is an ami device to control. From linux you need at least one
generic device (usually a SAF-te would suffice...).
--
intel1: 6:39pm up 10 days, 20:05, 4 users, load average: 0.00, 0.00, 0.00
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to [EMAIL PROTECTED]
