New CU/Mobo, Same System: Any Hope ?

Per Paul:
Ouch!

I especially liked the drive that reported a
"burst rate of 5MB/sec". Um, OK.

We are talking about the second link, right?

The first link did have one bad drive in it - whose rare earth magnet is
now holding up tools on my workbench.... (fourth from left on the top
row)

The *Second* link's graphs are all from the same drive.

As you suggest towards the end, I think it's time to run the same test
on the same drive on a couple different machines.

But, just to be sure, does the "Ouch" apply the graphs in the *second*
line (i.e. http://tinyurl.com/h43jkcs) ??
--
Pete Cresswell

(PeteCresswell) wrote:
Per Paul:
Ouch!

I especially liked the drive that reported a
"burst rate of 5MB/sec". Um, OK.

We are talking about the second link, right?

The first link did have one bad drive in it - whose rare earth magnet is
now holding up tools on my workbench.... (fourth from left on the top
row)

The *Second* link's graphs are all from the same drive.

As you suggest towards the end, I think it's time to run the same test
on the same drive on a couple different machines.

But, just to be sure, does the "Ouch" apply the graphs in the *second*
line (i.e. http://tinyurl.com/h43jkcs) ??

https://s24.postimg.org/vip6fj745/St2000_DL003x6.jpg

Top-left - graph has way-too-low bandwidth.
- sprinkle of anomalously high seek time on the left (yellow dots)

Bottom-left - yellow seek dots are falling well out of band
- burst rate is just as wrong as the one on top-left

Bottom-middle - reallocations etc in first percentage of disk
- yellow seek dots are well out of band

Generally, too much squiggle in graph line.

You can sometimes adjust the settings and change the block
size, and that will affect the appearance of the graph.

But I doubt that would result in the several cases
of bad symptoms.

I would test on another machine, anyway.

Like, what if all the drives did that. You'd
probably panic if they did.

Paul

Per Paul:
I would test on another machine, anyway.

Done.

Three observations and two questions.

The Observations:

- I ran a half-dozen benchmarks back-to-back on my DriveBender
server and got visibly-different results:
http://tinyurl.com/zhgrcwd I guess this implicates either
the entire system or the SATA sled what the drive was in
when I ran the tests on my 24-7 PC.

- During the process of benchmarking all my drives, I found
two obvious baddies on my 24-7 PC (even *I* could see that the error
scans were not good) in addition to the questionable drive that I
have been going on about.

- Since removing/replacing those 3 drives, my 24-7 PC has gotten
noticeably faster/snappier.... not just a little; quite a bit - enough
that I'm saying to myself "Geeze, this is great! I had forgotten how
it was supposed to be.


The Questions:

- Can you aim me at a paper or supply some terms-of-art for Googling so
I can learn how to read these graphs? A little bit of what you say
is soaking in, but I have a ways to go.

- What would be your comfort level with adding the questionable drive
to the DriveBender PC's pool? (benchmarks done on the DriveBender
PC: http://tinyurl.com/zhgrcwd)
--
Pete Cresswell

(PeteCresswell) wrote:
Per Paul:
I would test on another machine, anyway.

Done.

Three observations and two questions.

The Observations:

- I ran a half-dozen benchmarks back-to-back on my DriveBender
server and got visibly-different results:
http://tinyurl.com/zhgrcwd I guess this implicates either
the entire system or the SATA sled what the drive was in
when I ran the tests on my 24-7 PC.

- During the process of benchmarking all my drives, I found
two obvious baddies on my 24-7 PC (even *I* could see that the error
scans were not good) in addition to the questionable drive that I
have been going on about.

- Since removing/replacing those 3 drives, my 24-7 PC has gotten
noticeably faster/snappier.... not just a little; quite a bit - enough
that I'm saying to myself "Geeze, this is great! I had forgotten how
it was supposed to be.


The Questions:

- Can you aim me at a paper or supply some terms-of-art for Googling so
I can learn how to read these graphs? A little bit of what you say
is soaking in, but I have a ways to go.

- What would be your comfort level with adding the questionable drive
to the DriveBender PC's pool? (benchmarks done on the DriveBender
PC: http://tinyurl.com/zhgrcwd)

Usage of the benchmark for this purpose, is an
attempt to fill in for the lack of honesty in the
disk drive SMART info.

In the case of the old SCSI disks, you could
get an actual list (Grown Defect List). Based on
the LBA value of each, you could see whether they
were randomly distributed, or there was a "bad patch"
in the disk.

But that's not available in the ATA spec.

So we need some kind of scheme, to guestimate where
the defects may have been hiding.

SMART works to some extent, if the problems are
smoothly distributed over the disk. The forecasts
of trouble probably work OK in that case. It's when
the defects cluster in a 50GB wide swath, that the
SMART pretends "100% health", but we can see that
the disk isn't healthy. And if your OS partition
happens to fall on top of that bad spot, the
effect is perceptible enough to force you to
replace it, just to get the speed back.

Doing the analysis this (HDTune) way, is just a "made up
method", forced by the lack of good data from SMART.
You can read what you want into it. In the case of this
particular freebie program (HDTune), the trick is
to reject situations where the program is putting
spikes in the output, and not the disk drive.

I'd hoped that some day Linux would have one
of these kinds of programs. I was interested
in seeing if Linux was any more "quiet" than
Windows, and whether the spikes would disappear.

If you see some really thin spikes, they're
probably artifacts. Even a few yellow seek dots
outside the main band, isn't a sure sign of
failure. It's a judgment call, as to "how many
is too many".

So I've had the one drive, where it made OS operation
miserable (slow). And yet the two critical SMART
parameters were both zero. In the case of that
drive, the HDTune curve immediately showed the nice
alignment between where the OS partition sits, and
where the bad spot was (a nice wide one). And that
cinched the call for replacement. I know that some
people would look at the curve, and just move the
partition 100GB to the right, but that's not me :-)

One weird aspect about these drives, is they still
haven't outright "died" on me. They still run.
I use them as scratch disks. For example, I wanted
to run a RAID test, and I had two 500GB ones I could
erase and used for the setup. So they still get
used - just not for "production" OS usage. I don't
trust them enough for that.

Even the drive that let out the "squeaking" sound
one day, still runs. Really really weird stuff.
Usually when you get sound effects from a drive,
it should be dead the next day. And yet, after
a rest on the shelf for a day, that drive
was fine again.

Back in the Maxtor 40GB era, the end came quickly
for those drives. From "symptoms" to "toast"
took only 24 hours. That's why the behavior of
modern drives is so annoying. You're expecting
"toast" and... it doesn't happen.

Paul