TLER needs special RAID controller? was: Xeon ...

[willbill]

TLER needs special RAID controller? was: Xeon ...
(Time Limited Error Recovery, a WD feature

Advanced or not, if TLER is going to send an error msg instead of a block
of data, the controller has to know what it means and the driver has to
know what to do instead of receiving the data.

did a bit more homework. there sure is a lot
of confusion out there about TLER, and WD isn't
doing much to help reduce it

still looks to me that a WD SATA drive that has
TLER will work just fine with non-RAID usage

TLER is intended for use with a *redundant* raid setup,
coz when an error is reported to the raid controller,
after 7 seconds, all that is going to happen is
recovery of that small bit of data from the good
part of the raid set. iow, it won't drop the
entire drive, which is what tends to happen when
the resonse takes more than 8 seconds

TLER is clearly NOT for use with raid-0 coz as soon
as it takes 7 seconds (or more) to respond to the raid
controller, an error gets sent to the controller (even
if it's only a slow response situation), and wham,
the raid-0 is gone

with a TLER drive that you are using as a normal (non raid)
disk, you might or might not lose a small bit of data

best answer i've seen so far is Eugene's #39, in the Q/A
response to his review of a SATA WD RE drive. see:
http://forums.storagereview.net/index.php?s=adfe4fc157537bb10c1e4cf9a3fec62a&showtopic=20980&st=25

imho, the rest of the 4 pages isn't worth the time

even Eugene's #39 response leaves a number of
unanswered questions

about the only special thing you might need
with the raid controller is that it not drop
the drive when the drive reports an error,
but instead do a minor rebuild

at this point, afaict, it seems fairly clear that TLER
can be turned on and off, probably on all of WD's SATA
drives that have it. what is still unclear (to me)
is how this is actually done

bill

 

[J. Clarke]

TLER needs special RAID controller? was: Xeon ...
(Time Limited Error Recovery, a WD feature
on some of their SATA hard drives)

Geez, this is ludicrous. TLER is Western Digital Marketing Speak for "we're
too butt-headed to admit that we screwed up". The whole point of it is
that WD set the timeout on their drives so long that a drive would be
dropped out of a RAID as nonresponsive and dead during normal operation,
and rather than FIXING THE DAMNED THINGS they decided to make a new model
and charge extra for a shorter timeout.

 

[willbill]

willbill wrote:

Geez, this is ludicrous. TLER is Western Digital Marketing Speak
for "we're too butt-headed to admit that we screwed up".

the fact that TLER has been on WD's RE/RE2 SATA
hard drives for at least the last 12+ months
suggests that you may be right, but my hunch is
that it's more complicated than that

The whole point of it is that WD set the timeout on their drives
so long that a drive would be dropped out of a RAID as
nonresponsive and dead during normal operation, and rather than
FIXING THE DAMNED THINGS they decided to make a new model
and charge extra for a shorter timeout.

SATA drives appear (to me, with my very limited
1st hand experience) to be a very valid choice
for single use RAID sets

are SATA drives now becoming a real alternative
to more expensive SCSI drives for server RAID sets?

as well as the still unanswered question of does
TLER help within usage in a *server* RAID set?

bill

 

[Arno Wagner]

J. Clarke wrote:


the fact that TLER has been on WD's RE/RE2 SATA
hard drives for at least the last 12+ months
suggests that you may be right, but my hunch is
that it's more complicated than that

It is not. WD has serous enough troubles with read operations
that they need long recovery times occasionally. A reason to
stay away from their drives. BTW, Linux software RAID will drop
a drive on read errors, so WD drives (both variants) are essentially
unusable with it.

SATA drives appear (to me, with my very limited
1st hand experience) to be a very valid choice
for single use RAID sets

are SATA drives now becoming a real alternative
to more expensive SCSI drives for server RAID sets?

SATA is not SCSI. SCSI is still high-end (speed and reliability),
while SATA is consumer-grade. Still, is speed is not a primary
concern, an SATA RAID1/5/6 should be superiour to a single
SCSI drive today. And it may be cheaper as well.

as well as the still unanswered question of does
TLER help within usage in a *server* RAID set?

Stay away from WD, especially in servers. Use SCSI wehn
you need high reliability and can pay for it. Use Seagate,
Samsung, Hitachi, when SCSI is too expensive.

Arno

 

[George Macdonald]

the fact that TLER has been on WD's RE/RE2 SATA
hard drives for at least the last 12+ months
suggests that you may be right, but my hunch is
that it's more complicated than that




SATA drives appear (to me, with my very limited
1st hand experience) to be a very valid choice
for single use RAID sets

are SATA drives now becoming a real alternative
to more expensive SCSI drives for server RAID sets?

That's an interesting question and Seagate now has their Nearline-class
SATA drives targeted at "entry level servers". Quite what the difference
is, is hard to tell but I'm sure the mfrs are concerned that high ASP
devices might be on the decline. OTOH, how do you tell the desktop sector
that you have a better SATA drive but it's really not targeted at their
systems?

 

[willbill]

It is not.

^^^^^^^^^

excuse me?

WD has serous enough troubles with read operations
that they need long recovery times occasionally. A reason to
stay away from their drives. BTW, Linux software RAID will drop
a drive on read errors, so WD drives (both variants) are essentially
unusable with it.



SATA is not SCSI. SCSI is still high-end (speed and reliability),
while SATA is consumer-grade. Still, is speed is not a primary
concern, an SATA RAID1/5/6 should be superiour to a single
SCSI drive today. And it may be cheaper as well.




Stay away from WD, especially in servers.

again, excuse me

do you actually support WD SATA drives
that are used in real server RAID setups?

Use SCSI wehn
you need high reliability and can pay for it. Use Seagate,
Samsung, Hitachi, when SCSI is too expensive.

do you have any clue as to how many *server*
RAID setups actually now use SATA HDD?

if so, kindly post a link.

bill

 

[willbill]

That's an interesting question

i'll take that as a compliment.

i suspect that SATA drives are now
starting to be used in real server
RAID sets, but i don't know it for sure

but i've also been out of IT for the
last 11 years (last was a very large
Sun setup). i mean, what does
www.newegg.com use for raid (assumming
that they use raid) in their servers?
SCSI or less expensive SATA? and what
does www.anandtech.com use for raid in
their servers? not to mention, what does
either of them use for the CPU? x86 or
dual/quad core or what?

the latest www.anandtech.com storage review
(on a WD RE16 drive 2 weeks ago), made
mention of doing a followup review of
raid issues (which might or might not
comment on WD TLER)

problem is i don't trust anandtech on an issue
like this, but their comments might well be
worth looking at and thinking about, when it
shows up (presumably in the next 4 weeks)

and Seagate now has their Nearline-class
SATA drives targeted at "entry level servers".

not to mention, how would one benchmark SATA
HDD "drop out" when used in a raid set?

i can't think of any easy way

to me, it's more of a seat of the pants issue for
those who actually use/support real *server* raid
arrays that actually use SATA drives; and i rather
suspect that this is still rare at this point

Quite what the difference
is, is hard to tell but I'm sure the mfrs are concerned that high ASP

"ASP" = ?

devices might be on the decline. OTOH, how do you tell the desktop sector
that you have a better SATA drive but it's really not targeted at their
systems?

this may be why WD is so confusing on the subject
of TLER with their current RE/RE2/Raptor drives

i mean hey, i ordered a normal 150 GB Raptor
(for non raid use), and i expect it to show up
with TLER turned off. and if not, that i can
turn it off (if it shows up turned on)

i've also got two WD RE drives, but IDE (not SATA),
and i rather like them better than my IDE 7200.9
drives (their physical construction, and how they
don't "ping" when i tap them with my fingertip.
i mean, they both come with 5 year warranties.)

(i have both IDE/SATA 7200.9 drives, which are,
from the bottom, difficult/impossible to tell
them apart)

bill

 

[daytripper]

i suspect that SATA drives are now
starting to be used in real server
RAID sets, but i don't know it for sure

[...]

They are. For one example, Stratus uses SATA drives in Raid 1+0 on their model
5700 - which is a fault tolerant 2P dual-core server, no less. A real
*serious* server...

http://www.stratus.com/products/ftserver/software.htm

/daytripper

 

[Arno Wagner]

excuse me?

I think your hunch is incorrect.

again, excuse me

do you actually support WD SATA drives
that are used in real server RAID setups?

No. I think WD is behind the other manufacturers both
in reliability and in design. The only thing they have going
for them is speed. Speed is not enough to make up for their
other problems.

do you have any clue as to how many *server*
RAID setups actually now use SATA HDD?

Oh, I know. I have some servers with a few TBs in
SATA and ATA myself. But these are medium reliability,
medium speed set-ups, and I do not have the money for SCSI.

I am just cautioning you to not consider SATA the equal of
SCSI. I know that many vendors will happily claim that
today SATA is equal to SCSI at a lower price. They are lying
to boost sales.

if so, kindly post a link.

Link? To what?

Arno

 

[Tony Hill]

SATA drives appear (to me, with my very limited
1st hand experience) to be a very valid choice
for single use RAID sets

are SATA drives now becoming a real alternative
to more expensive SCSI drives for server RAID sets?

Given that virtually all servers are now offered with the option of
SATA drives rather, I would say that yes, it is a real alternative to
SCSI for servers for operations that aren't really disk intensive.
SCSI does still seem better able to handle the really disk intensive
stuff better, particularly with regards to having LOTS of small,
independent disk reads/writes. However there are plenty of server
setups where this isn't such a big worry, especially given that you
usually want your server running almost entirely out of RAM instead of
disks anyway.

Also, the real key difference between SATA and SCSI has very little to
do with the interface and a LOT to do with target markets. In much
the same way that there's no technical reason why SATA drives are
almost all 7200 rpm while SCSI drives are almost all 10K or 15K rpm,
there's also absolutely nothing in SCSI that would make the drive any
more reliable and even the performance differences, at equal RPMs and
cache, are as much to do with different optimizations in the firmware
than differences in the interface. This is why SCSI drives are almost
always slower than similar-spec SATA drives for "desktop" style
applications (larger sequential read/writes), but the reverse is true
for "server" style applications (smaller non-sequential read/writes).

The new thing these days from the big three (or is it only two now?)
hard drive companies is their "Enterprise" class SATA drives. The
theory being that these are SATA drives designed for servers, though
the difference at the moment seems to be more marketing than anything
else. Still I would expect the trend to continue. Basically these
drives are going to replace all the server roles that might otherwise
been served by a 7200rpm SCSI drive, or possibly even 10Krpm SCSI
drives. There will still be a market for 15Krpm SCSI and Serial
Attached SCSI drives, but the low-end will probably move towards SATA.

 

[Folkert Rienstra]

It is not. WD has serous enough troubles with read operations
that they need long recovery times occasionally. A reason to
stay away from their drives.

BTW, Linux software RAID will drop a drive on read errors,

so WD drives (both variants) are essentially unusable with it.

Oh goodie, that must be really bad, right?
Like that won't happen with drives of other makes when they exhibit
read errors. Not to mention that one would even want them to not be
dropped if they exhibited read errors.

SATA is not SCSI.

No kidding. Maybe that's why they name it differently then.

SCSI is still high-end (speed and reliability), while SATA is consumer-grade.

Clueless, 'as always'.

Still, is speed is not a primary concern, an SATA RAID1/5/6 should be su-
periour to a single SCSI drive today.

Utter nonsense on the access aspect of 'speed'.

 

[George Macdonald]

i'll take that as a compliment.

i suspect that SATA drives are now
starting to be used in real server
RAID sets, but i don't know it for sure

I think that's the point of the Caviar RE line and the Seagate Nearline
class. Obviously there's a market for an uprated SATA drive; remember the
Deskstars which were rated for a max hours per monthly cycle a couple of
years back? On top of the click of death saga, finished them off.

but i've also been out of IT for the
last 11 years (last was a very large
Sun setup). i mean, what does
www.newegg.com use for raid (assumming
that they use raid) in their servers?
SCSI or less expensive SATA? and what
does www.anandtech.com use for raid in
their servers? not to mention, what does
either of them use for the CPU? x86 or
dual/quad core or what?

the latest www.anandtech.com storage review
(on a WD RE16 drive 2 weeks ago), made
mention of doing a followup review of
raid issues (which might or might not
comment on WD TLER)

problem is i don't trust anandtech on an issue
like this, but their comments might well be
worth looking at and thinking about, when it
shows up (presumably in the next 4 weeks)




not to mention, how would one benchmark SATA
HDD "drop out" when used in a raid set?

i can't think of any easy way

to me, it's more of a seat of the pants issue for
those who actually use/support real *server* raid
arrays that actually use SATA drives; and i rather
suspect that this is still rare at this point

I'm not sure on that - corporate computing is moving into the glasshouse,
departmental servers are verboten in the brave new, security oriented
computing world so enterprise-class drives will have their place. OTOH
there are many more SMB servers than enterprise ones and for a less than
critical application, like say an Intranet Web Server, many smaller
businesses will be content to have an uprated drive like Caviar RE or
Seagate Nearline SATA RAID-1 array for "reliable enough".

"ASP" = ?

Average Selling Price.

this may be why WD is so confusing on the subject
of TLER with their current RE/RE2/Raptor drives

i mean hey, i ordered a normal 150 GB Raptor
(for non raid use), and i expect it to show up
with TLER turned off. and if not, that i can
turn it off (if it shows up turned on)

i've also got two WD RE drives, but IDE (not SATA),
and i rather like them better than my IDE 7200.9
drives (their physical construction, and how they
don't "ping" when i tap them with my fingertip.
i mean, they both come with 5 year warranties.)

I got a Caviar SE in an emergency from a local store a while back and
couldn't wait to get it outa there, when the replacement Seagate arrived -
horrible throbbing drone which about drove me nuts. OTOH, it's taken
Seagate a while to realize that SMT components should be on the inside of
the PCB. Nasty story there where, after swapping drives in and out of
drive cages a few times, I got things all put back together, only to find a
SM capacitor lying on the desk.<gack>

 

[J. Clarke]

the fact that TLER has been on WD's RE/RE2 SATA
hard drives for at least the last 12+ months
suggests that you may be right, but my hunch is
that it's more complicated than that

Nope. The history is that WD IDE drives attached to various RAID
controllers were failing right and left. It turned out that the problem
was that the error recovery timeout on the WD drives was so long that the
controller timed out on attempted access before the error recovery timed
out and so the controller marked the drive as offline.

The fix was to shorten the timeout, but instead of doing this across the
board WD instead make "RAID Edition" drives and charged extra for them.

Classic case of "if life hands you a lemon make lemonade".

SATA drives appear (to me, with my very limited
1st hand experience) to be a very valid choice
for single use RAID sets

are SATA drives now becoming a real alternative
to more expensive SCSI drives for server RAID sets?

as well as the still unanswered question of does
TLER help within usage in a *server* RAID set?

It's very simple. If you are using a WD drive in a RAID then you need TLER
unless the manufacturer of the RAID controller states _specifically_ that
you do _not_ need it. The safer choice is to simply avoid WD for mission
critical systems.

As for SATA being acceptable for mission-critical RAID, it's really a matter
of finding a host adapter that you trust and that has the performance you
need. The whole point of RAID is that drive failure is a given and so the
reliability of individual drives has little effect on system
reliability--the lifecycle cost for SATA might be higher, lower, or the
same as SCSI--I don't know of any direct comparisons that have been run.

It can be argued that it is potentially superior to SCSI in that a properly
designed SATA RAID controller has each drive on a separate channel and so a
drive failure that affects the interface (which can happen, I've seen it
with both SCSI and IDE systems) remains isolated with SATA while with SCSI
it takes down the whole channel until the defective drive is removed. In
practice however the RAID controllers for SATA have for the most part been
relatively "lightweight" units that either lacked some reliability features
such as battery-backed cache or offered relatively poor performance.

 

[J. Clarke]

I think that's the point of the Caviar RE line

Actually the point of the Caviar RE line is to make a buck by changing a
couple of bits in a ROM.

and the Seagate Nearline
class.

Not clear what Seagate is doing there, but the NL35 series seems to allow
higher nonoperating shock than the 7200.10, so clearly there's _some_
difference.

Obviously there's a market for an uprated SATA drive; remember the
Deskstars which were rated for a max hours per monthly cycle a couple of
years back? On top of the click of death saga, finished them off.

Best Buy and CompUSA and Staples and the rest seem to have shelves full of
Deskstars so maybe you need to inform them that the product was "finished
off" several years ago.