On 14/05/2012 9:47 AM, Percival P. Cassidy wrote:
> On 05/14/12 12:25 am, Yousuf Khan wrote:
>
>> I can see HDD's taking over the spot of current tape-backup operations.
>> It's already happening actually. Basically archival storage.
>
> I was talking this past weekend with a member of my extended family who
> works for an organization with huge server farms. Their backups use
> humungous (about the size of the living room where we were talking)
> robotic tape systems.

Yup, that used to be something I used to do myself: enterprise tape
backup operations. These days hard drives are cheap enough and large
enough to take over some of the medium-sized jobs that tape used to do.
Tapes usually hold several terabytes themselves, and they have a
sequential read/write speed advantage over hard drives still, especially
when you have multiple simultaneous tape drives being used
simultaneously. It's amazing how quickly you can backup an entire
department with multiple tape drives. And these days with a SAN serving
as the medium for data transfer, you can keep several tape drives fully
saturated and busy enough to justify the purchase of the robotic
library. You can backup an entire departmental data centre within a
couple of hours starting after midnight and going to the morning before
users come back into the office.

However, I am seeing a trend towards using removable hard drives in some
of the smaller departmental settings because they are easier to do
random accesses on, and they offer cheaper prices at the mid-level than
tapes do. You also see larger departments using HDD's as archival
storage in hierarchical storage management scenarios. Hierarchical
storage is where little used files are moved to long-term storage in a
slightly farther drive system, freeing up your local hard drive from
getting filled up. The archived files will still be easily accessible,
on demand.

Yousuf Khan

Percival P. Cassidy <(E-Mail Removed)> wrote
> Rod Speed wrote

>>> I can see HDD's taking over the spot of current tape-backup operations.

>> Hardly anyone uses tape for backup anymore, they use hard drives.

> I was talking this past weekend with a member of my extended family who
> works for an organization with huge server farms. Their backups use
> humungous (about the size of the living room where we were talking)
> robotic tape systems.

Sure, but we were discussing storage in personal computers, not huge server
farms.

There's only a few dinosaurs that still use tape for backup of personal
computers.

Even google doesn't.

Tom Del Rosso <(E-Mail Removed)> wrote
> Rod Speed wrote
>> Yousuf Khan <(E-Mail Removed)> wrote

>>> Hard disks don't follow the Moore's Law rate, they are much slower in
>>> density increases.

>> That’s just plain wrong. And says nothing useful about whether they
>> will ever be as cheap as hard drives with 10TB drives etc anyway.

> Moore's Law isn't in effect anyway. He said transistor count would double
> every year, and it did from 1959 to the mid-80's. Then it was every 18
> months or 2 years but people still call it Moore's Law.

Yeah, the detail has certainly varied since his original observation.
http://en.wikipedia.org/wiki/Moore%27s_law

>> Hardly anyone uses tape for backup anymore, they use hard drives.

> With one tape drive equal to the price of a server, they use tape where
> they have lots of servers.

Not always, google doesn’t.

Yousuf Khan <(E-Mail Removed)> wrote
> Rod Speed wrote
>> Yousuf Khan <(E-Mail Removed)> wrote

>>> One advantage of an old SSD is that it is still alive for reads even
>>> after death.

>> Depends on how it dies. That isnt how they die on a power failure when
>> writing.

> But that isn't really dying, that's just a power failure.

The power failure is the cause of the SSD dying in the sense that the data
is lost.

> The drive is still usable afterwards when the power is re-established.

Not necessarily. The data can be lost.

> Some data might get corrupted, but the drive will still be usable.

Just as true of many hard drive failures.

The point is that with that sort of SSD failure,
it doesn't just go read only with no data lost.

>>> With an HDD, when it dies -- it dies, for both reads and writes.

>> That's just plain wrong too. Quite a bit of the time they just have
>> some unreallocatable sectors and you can still read the rest fine.

>> Just had that with a Toshiba that died with just 2 weeks to go with
>> warranty it was possible to do a final backup before returning it fine.

> But that's not the definition of dying either.

It is failure of the hard drive in the sense that it has to be replaced.

> Dying means becoming unusable.

But isnt necessarily completely unusable in the sense
of being able to move the data to the replacement.

> A lot of bad sectors is just a major error, but still not dead: but it may
> be enough to get warranty replacement, sure.

Its dead in the sense that it needs to be replaced.

No different to an SSD that refuses to write any
more and so needs to be replaced and the data
moved to the replacement when that happens.

> Dead is when for example the controller dies, or the spindle seizes up.

That's just a more dramatic death. And SSDs have that sort of death too.

>>> As far as its maximum writes,

>> Which hard drives don't have.

> But a hard drive will usually run out of reads and writes at the same
> time.

That's just plain wrong. Quite often they degrade more gracefully.

And SSDs just die too, not being able to read or write.

>>> I don't see a big problem with write lifetimes

>> I do when hard drives don't have any write lifetime.

> No,

Yep.

> HDD's just have an overall lifetime.

No they don't. Most get replaced because they are too small.

Few actually die.

Even with modern drives, most systems get replaced for
various reasons and the hard drive does not die at all.

Just had one of those the other day too.

>>> as they make it sound like in most write-ups. For example, I've had my
>>> new SSD for less than a month, it's a 240GB drive,

>> I'm not silly enough to buy hard drives that small today.

> More than big enough for an operating system drive.

Few bother with more than one drive in a system today.

> You can still use big hard drives for data drives.

Makes a lot more sense to have everything on the much bigger drive.

> That's what I'm doing. Even when I was using an HDD for my operating
> system drive,

Hardly anyone is silly enough to have a
separate physical drive for the OS anymore.

> I had it partitioned down to 200GB for the OS, and the remaining for data.

But that's not a separate physical drive for the OS,

That's a much larger than 200GB drive.

> Keeping the OS partition small allows for much easier backup and restore
> of the OS.

That's just plain wrong with modern backup systems.

And is nothing like a separate physical drive for the OS anyway.

It makes absolutely no sense to be buying 240GB drives anymore.

>> In fact I don't buy hard drives smaller than 2TB today.

> Good for you.

>>> and I've already had about 1.09TB written to it (almost 5x more than its
>>> capacity). It's still showing 100% lifetime.

>> And it remains to be seen if that is a lie or not like hard drive
>> MTBFs always were.

> It's a separate SMART value only available in SSD's.

And it remains to be seen if that is a lie or not like hard drive MTBFs
always were.

> There's no reason to believe that SSD's don't know exactly how much is
> being written to them, based on block sizes and number of blocks written.

And it remains to be seen if they are lying about lifetimes.

> It's also safe to assume that the SSD's controller knows exactly which
> blocks are no longer writeable, so it may update the SSD lifetime value
> based on the number of unwriteable blocks accumulated.

And it remains to be seen if they are lying about lifetimes because
it isnt actually a linear function of what has already been seen.

>>> Of course I've attempted to keep the really heavy-duty and frequent
>>> write operations off of it, like the pagefile and file indexer.

>> And that isnt even possible if you don't have a hard drive.

> Sure, but even when I had a boot hard drive, I used to keep the pagefile
> on other hard drives.

And few were silly enough to have more than one hard drive
and those who cared about performance ensured that they
had enough physical ram so that the pagefile did not get
used enough so that the location of it mattered a damn.

> But a lot of people also keep their pagefiles on the SSD.

And are likely to discover the downsides of SSDs when they do that.

Yousuf Khan <(E-Mail Removed)> wrote
> Rod Speed wrote
>> Yousuf Khan <(E-Mail Removed)> wrote

>>> which is a factor of only 8x difference. A few years ago, the difference
>>> was more like 32x. I can see 1TB SSD's aren't too far around the corner
>>> (maybe by next year),

>> By which time you are likely to see 6TB hard drives too.

> Which actually just proves my point.

Nope. And your CLAIM cant be proven anyway until that time
actually shows up and we can see what actually happened.

> If 1TB SSD's are to be met by 6TB HDD's, then the capacity differential
> will only be 6X, which is a smaller differential.

The maximum size differential is irrelevant to what matters, $/TB.

We arent going to see that cross over any time soon.

Essentially because it doesn’t cost anymore to double
the density with hard drives, in fact it costs less.

You can never get around the fact that with with an SSD,
double the capacity means double the number of transistors.

> If the differential started out at 32X and is now around 8X, and then
> it'll be 6X, the trend is obvious, the SSD is catching upto the HDD in
> capacity.

Irrelevant to what matters, $/TB.

>>> When the HDD size stagnates,

>> Don’t believe that will ever happen with operations like google around.

> That's hardly a home usage scenario, that's an enterprise usage scenario.
> In fact, it's operations like Google that might reduce the need for
> internal storage capacity at home.

And so that just means more and more hard drives for google
so that will keep driving the increased capacity of hard drives.

And I doubt too many will keep all their irreplaceable stuff like
videos of their grandkids etc on something like google exclusively.

> They are the companies developing cloud storage, which might make large
> capacities at home unnecessary.

Not a chance, because the absolute vast bulk of
what people choose to record with their PVR wont
be available for free from operations like that.

I doubt that they will even be cheap enough so that no one
bothers to record anything from free to air TV anymore.

And they certainly arent going to do it with video of the grandkids etc.

>>> if network bandwidth allows them.

>> It handles hundreds of channels simultaneously fine with digital TV.

> With digital tv, you aren't getting all channels broadcast to you all of
> the time, just the ones that you're watching or recording.

But you can clearly record a hell of a lot more than the 2 you claimed
that PVRs can only do, so there is no point in using a MUCH more
expensive and much smaller SSD for the storage on a PVR for the
alleged increased performance which you don’t actually get at all.

>>> and faster and smoother search response (forward and backwards).

>> That’s just plain wrong. Its done from the file
>> structure and searching is instant even with a
>> hard drive because the file structure is cached.

> Searching forwards is easy, but searching backwards has always been jerky.

Not with digital TV it aint. Backwards is identical to forwards.

>>> I can see HDD's taking over the spot of current tape-backup operations.

>> Hardly anyone uses tape for backup anymore, they use hard drives.

> Never worked in a data server centre have you?

Ran them thanks. And we were discussing storage
on personal computers, not data server centers anyway.

> When you're backing up petabytes of data, you're using tape right now.

Google doesn’t.

> We're talking about robotic tape libraries equipped with multiple DLT or
> LTO tape drives.

Not with the PERSONAL COMPUTERS being discussed we don’t.

We use hard drives, not tape.

>>> It's already happening actually. Basically archival storage.

>> Not with SSDs it aint, because they don’t last
>> as long as hard drives for archival storage.

> I was talking about HDD's taking over from tape.

You were claiming that SSDs would take over from tape with
PERSONAL COMPUTERS. Not a chance, hard drives will be
whats used for backups and have been for a long time now.

Only a fool would use SSDs for archival storage of PERSONAL
COMPUTERS with so little experience with long term archival
storage on SSDs yet. Some might try they in ADDITION to
hard drives to see how they pan out over time.

On 12/05/2012 2:16 AM, Rod Speed wrote:
> Tom Del Rosso <(E-Mail Removed)> wrote
>> And does it really matter if they have the same parent company? It's
>> still a different production line.
>
> It remains to be seen how long they continue to develop new products for
> each production line tho.
> Cant see that continuing forever, even with Seagate that does tend to
> keep the old operation going for a long time sometimes.

Seagate probably keeps the factories going a long time, as this
increases their production capacity, but as these factories are upgraded
it probably eventually consolidates design towards one stream of products.

Yousuf Khan

On 14/05/2012 3:24 PM, Rod Speed wrote:
> Percival P. Cassidy <(E-Mail Removed)> wrote
>> I was talking this past weekend with a member of my extended family
>> who works for an organization with huge server farms. Their backups
>> use humungous (about the size of the living room where we were
>> talking) robotic tape systems.
>
> Sure, but we were discussing storage in personal computers, not huge
> server farms.
>
> There's only a few dinosaurs that still use tape for backup of personal
> computers.
>
> Even google doesn't.

But google isn't about personal computers either, it's a huge server
farm. In fact it's several server farms, highly spread out over the world.

As for Google not using tape drives to backup its data, that maybe true
of its online data servers, as that data is highly redundant anyways,
and in some cases it's too big to backup. However we don't know if it's
true for Google's internal servers where it keeps its in-house data such
as accounting, CRM, RDBMS, etc. Most likely it isn't true, and Google
does have tape backups of its in-house data, if for no other reason than
regulatory reasons. There's entire businesses, like Iron Mountain, setup
specifically for managing off-site archival tape data. These businesses
exist because many other businesses require maintenance of years-old or
even decades-old data.

Yousuf Khan

On 14/05/2012 4:09 PM, Rod Speed wrote:
> Yousuf Khan <(E-Mail Removed)> wrote
>> Rod Speed wrote
>>> Yousuf Khan <(E-Mail Removed)> wrote
>
>>>> which is a factor of only 8x difference. A few years ago, the
>>>> difference was more like 32x. I can see 1TB SSD's aren't too far
>>>> around the corner (maybe by next year),
>
>>> By which time you are likely to see 6TB hard drives too.
>
>> Which actually just proves my point.
>
> Nope. And your CLAIM cant be proven anyway until that time
> actually shows up and we can see what actually happened.

And since it was *your* example, then by extension your claim can't be
proven either.

>> If 1TB SSD's are to be met by 6TB HDD's, then the capacity
>> differential will only be 6X, which is a smaller differential.
>
> The maximum size differential is irrelevant to what matters, $/TB.

In a business environment that can be justified, based on performance.

> We arent going to see that cross over any time soon.

The performance/$ is already higher with SSD, that crossed over a long
time ago. Even if capacity/$ isn't fully matching HDD's yet, some
businesses may be looking at the performance/$ ratio and saying that the
capacity is good enough for certain performance critical apps. I'm
thinking of HPC environments like science projects (e.g. Large Hadron
Collider) that expect to see data streaming rates in the TB/s range.

> Essentially because it doesn’t cost anymore to double
> the density with hard drives, in fact it costs less.

That can be said for flash too. Currently flash is low-priority
semiconductors, which is produced on older process nodes, such as 45nm,
if the demand for flash increases, then they will get produced on the
newest process nodes, such as 32nm and 22nm. You can achieve huge
density doublings just by going to the latest process nodes.

Flash also has the MLC (multi-level cell) feature which allows them to
store two bits per cell currently. This is already higher than the 1 bit
per cell available on SLC (single-level cell) flash. Next generation,
this can likely be increased to 3 bits per cell, and then 4 bits, etc.
So even without a process node change, you can still achieve doublings
of capacity on flash.

> You can never get around the fact that with with an SSD,
> double the capacity means double the number of transistors.

See above, about MLC flash.

> And I doubt too many will keep all their irreplaceable stuff like
> videos of their grandkids etc on something like google exclusively.

Well, I don't see that happening yet either, as the average broadband
bandwidth isn't high enough yet, and people find it difficult to keep
storage locations straight in their minds. However, videos of grandkids
are hardly straining anyone's personal storage anymore. Again, we're
approaching a point of diminishing returns, where capacity is good
enough. Many people are finding even a 500GB drive good enough for all
of their needs. For me, it would never be enough of course, but I'm not
typical.

>> They are the companies developing cloud storage, which might make
>> large capacities at home unnecessary.
>
> Not a chance, because the absolute vast bulk of
> what people choose to record with their PVR wont
> be available for free from operations like that.

That's not the market that the cloud storage people are after anyways.
They are after more of the Facebook-style crowd, sharing and storing
photos, maybe a few video clips, etc. Usually, they may give upto 10GB
for free on these services.

If you stretch the definition of cloud a bit, then even the PVR crowd is
accommodated somewhat by services like Netflix.

>>>> if network bandwidth allows them.
>
>>> It handles hundreds of channels simultaneously fine with digital TV.
>
>> With digital tv, you aren't getting all channels broadcast to you all
>> of the time, just the ones that you're watching or recording.
>
> But you can clearly record a hell of a lot more than the 2 you claimed
> that PVRs can only do, so there is no point in using a MUCH more
> expensive and much smaller SSD for the storage on a PVR for the
> alleged increased performance which you don’t actually get at all.

Well, all cable/satellite provider's PVR's I've seen so far are limited
to recording upto 2 HD or SD video streams. Whether that's a limitation
of hard disk speed, or network speed, I can't say for sure. It maybe a
network speed limitation then.

>>>> I can see HDD's taking over the spot of current tape-backup operations.
>
>>> Hardly anyone uses tape for backup anymore, they use hard drives.
>
>> Never worked in a data server centre have you?
>
> Ran them thanks. And we were discussing storage
> on personal computers, not data server centers anyway.

Well, now we're talking about enterprise stuff too.

>>> Not with SSDs it aint, because they don’t last
>>> as long as hard drives for archival storage.
>
>> I was talking about HDD's taking over from tape.
>
> You were claiming that SSDs would take over from tape with
> PERSONAL COMPUTERS. Not a chance, hard drives will be
> whats used for backups and have been for a long time now.
>
> Only a fool would use SSDs for archival storage of PERSONAL
> COMPUTERS with so little experience with long term archival
> storage on SSDs yet. Some might try they in ADDITION to
> hard drives to see how they pan out over time.

This part has nothing to do with SSD's, we're talking about enterprises
replacing tape backup with hard drive backup instead.

Yousuf Khan

Yousuf Khan <(E-Mail Removed)> wrote:

>Well, all cable/satellite provider's PVR's I've seen so far are limited
>to recording upto 2 HD or SD video streams.

I have a dish 722k satellite receiver with an OTA (Over The Air)
adapter. It can record two programs off the satellite feed and two
programs off the OTA adapter for a total of four.

The new Dish Hopper system has three tuners. During prime time one
of those tuners can record all four networks (ABC, CBS, FOX and
NBC). So this receiver can record up to six different programs
simultaneously.
--
When a cat sits in a human's lap both the human and the cat are usually
happy. The human is happy because he thinks the cat is sitting on him/her
because it loves her/him. The cat is happy because it thinks that by sitting
on the human it is dominant over the human.

Yousuf Khan <(E-Mail Removed)> wrote
> Rod Speed wrote
>> Tom Del Rosso <(E-Mail Removed)> wrote

>>> And does it really matter if they have the same parent company? It's
>>> still a different production line.

>> It remains to be seen how long they continue to develop new products for
>> each production line tho.

>> Cant see that continuing forever, even with Seagate that does tend to
>> keep the old operation going for a long time sometimes.

> Seagate probably keeps the factories going a long time, as this increases
> their production capacity,

But they are one of the few that has chosen to move
production to china so its possible that they may
choose to do even more of that with the operations
they have taken over, presumably in a hard nosed
approach to wiping out the competition.

> but as these factories are upgraded it probably eventually consolidates
> design towards one stream of products.

Yeah, that’s what I meant in different words.