Subject: | |
From: | |
Reply To: | Tony B. Shepherd |
Date: | Fri, 25 Aug 1995 06:15:42 -0400 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
In article <[log in to unmask]>,
Paul Wang <[log in to unmask]> wrote:
] Subject: Re: sector atomicity
] Guy wrote:
] >Could someone elaborate on "sector atomicity" and why this feature
] >would confound the firmware in off-the-shelf SCSI discs?
]
] Sector atomicity means, as its name suggested, the whole sector is either
] completely written or nothing is written. This is particularly useful
] when the power fails!
]
] Without this feature, a sector could be partially written if power fails
] in the middle of an I/O. Subsequent read of this sector will result in a
] read error, since the ECC checksum won't match. The end result is data
] corruption! If this sector happens to be holding critical data, such as
] transaction manager log files, the whole volume set may be lost!
]
] Therefore I will never put my production data on non-HP disks. For crash
] and burn test or development data, I think non-HP disks are OK as long as
] reload/data lost is not an issue.
I understand your concern. Series II machines (early 70's) offered data
protection too.
I'm not a hardware ace, but it seems to me the issue is a matter of
maintaining power to the controller and drive long enough to be sure an
I/O operation is complete, internal (drive and controller) buffers are
written back and 'turn-off' orders are issued to the drive.
So what we're talking about is a mini-UPS? Something I'm missing?
--
Regards -- Tony B. Shepherd -- [log in to unmask]
This opinion is worth what you paid for it, and if it proves to be wrong,
all monies paid will be cheerfully refunded upon presentation of receipt.
|
|
|