Intel 160 GB ssd drive now $425

[Lynn McGuire]

Intel 160 GB ssd drive now $425:
http://www.newegg.com/Product/Product.aspx?Item=N82E16820167017

OK, this is getting very close to my buy point. Very close.

Lynn

 

[Arno]

Hmmm.... Does anyone here know if SSDs can be RAIDed at level 0,
i.e. data striped for faster throughput? Would it make any sense to do so?

Yes. But they require special controllers or suitable software RAID
for this to make sense. Ordinary RAID controllers were designed with
relatively slow access times and will often perform badly with SSDs.

However, if throughput is your goal, conventional HDDs may perform
similar, at a fraction of the cost.

Incidentially, I have a HDD/SSD RAID1 running under Linux software
RAID, with the HDDs set to be typically only written. Gives SSD
read speeds in practive and HDD write speeds. Perfect for my
application and even better than different brand HDDs in a RAID1
for added reliability.

Arno

 

[DevilsPGD]

In message <[email protected]> "Timothy

That hybrid RAID system sounds intriguing (out of my league, but
still intriguing). Was all the software/firmware commonly available,
or did you author some of it yourself? Care to share pointers with
others here who might want to do the same or similar?

Personally, I'm really looking forward to when some RAID card vendor
grabs this idea and does it in hardware. Even better would be a
controller built to support RAID-0 across several SSDs along with 1-2
HDDs for redundancy.

I think you could probably have SSD write speeds too in most
circumstances if you're willing to use the SSD as a buffer, and complete
the write to the HDD as fast as that drive permits.

There would obviously be some risk to data, but in order to have data
loss you'd need to have an SSD failure mid-write and you'd only lose the
amount of buffered data. You could still honour requests from the OS to
flush the cache when needed.

All of that being said, so far SSDs seem to have a level of reliability
that exceeds traditional HDDs, so personally, I'm reasonably content
with nightly images of my SSD, at least for most purposes.

 

[Arno]

That hybrid RAID system sounds intriguing (out of my league, but
still intriguing). Was all the software/firmware commonly available,
or did you author some of it yourself? Care to share pointers with
others here who might want to do the same or similar?

Very simple: Do a Linux software RAID1 and set the --write-mostly
command line option when adding the HDDs/HDD partitions with mdadm.
Add the SSD/SSD patition without that option. If doing this for an
existing RAID, you will have to kick the elements one at a time,
re-add with --write-mostly and wait for the RAID to sync. I
did the latter, as I replaced one HDD partition in a 3-way
RAID1 with an SSD.

The software is just standard Linux kernel RAID via the md driver
and the standard mdadm administration software. Nothing special
at all needed.

What you get is that reads go almost exclusively to the SSD,
writes go to all (on RAID1). The option was originally intended
when using a remote mirror, e.g. with ATA over ethernet, to
reduce network traffic. All operations that need to read from
the HDDs as well (e.g. consistency checks) will still do so. The
RAID software just avoids reading from the member disks/partitions
flagged as "write mostly" whenever it has a choice.

Arno

 

[Lynn McGuire]

That hybrid RAID system sounds intriguing (out of my league, but

still intriguing). Was all the software/firmware commonly available,
or did you author some of it yourself? Care to share pointers with
others here who might want to do the same or similar?

Also see this new Seagate hybrid hard drive:
http://www.newegg.com/Product/Product.aspx?Item=N82E16822148591

Lynn

 

[Arno]

Seagate Momentus XT ST95005620AS 500GB 7200 RPM 32MB Cache
2.5" SATA 3.0Gb/s with NCQ Solid State Hybrid Drive for $135 plus
applicable state tax plus $6.30 shipping. Nice price point.

*TimDaniels*

$60 for a 4GB SSD. Nice racket Seagate has going there.
Also don't expect wonders with an SSD this small, this
is still fundamentally a HDD.

Arno

 

[DevilsPGD]

$60 for a 4GB SSD. Nice racket Seagate has going there.
Also don't expect wonders with an SSD this small, this
is still fundamentally a HDD.

You're not buying a 500GB 7200rpm drive and a 4GB SSD. Rather, you're
buying those components, plus Seagate's logic to give you the best
possible performance out of those components.

You might be paying more than the individual components would cost, but
from the reviews I've read, if you can't justify a SSD (or have a user
who can't grasp managing a SSD and 7200 rpm drive in the same system),
it looks like this drive is a reasonable compromise.

 

[Arno]

In message <[email protected]> Arno <[email protected]> was
claimed to have wrote:

You're not buying a 500GB 7200rpm drive and a 4GB SSD. Rather, you're
buying those components, plus Seagate's logic to give you the best
possible performance out of those components.

Which is rather limited. Also I seriously doubt the "best possible".
It will be something that is showy on benchmarks (and optimized to
be) and will give somerather limited boost on actual usage. Of course,
if you always only use the same 2-3GB of the disk, it will be fast,
but then getting a small (say, 30GB) SSD would have been as fast or
faster at the same price, and that for all 30GB of it. For
large accesses or repetitive accesses that exceed a 4GB window,
the hybrid disk can basically do nothing and may even slow things
down due to trashing.

You might be paying more than the individual components would cost, but
from the reviews I've read, if you can't justify a SSD (or have a user
who can't grasp managing a SSD and 7200 rpm drive in the same system),
it looks like this drive is a reasonable compromise.

This is a laptop drive. For this special case only, it does make
some limited sense as you cannot put two drives into most
laptops. For the general case getting both a 500GB drive
and a 30GB SSD and then putting OS and often accessed data on
the 30GB SSD will cost about the same and be far more effective.

Given what the 4GB SSD costs Seagate and what they can charge
for it, it is quite clear why Seagate marketing is hyping this
thing.

Arno

 

[Percival P. Cassidy]

Personally, I'd get two (one for the page file).

It was suggested to me recently that, because of the finite number of
writes that SSDs will survive, the only thing for which they are really
suitable is something that will seldom be written but frequently read --
i.e., the OS itself.

Perce

 

[Lynn McGuire]

It was suggested to me recently that, because of the finite number of writes that SSDs will survive, the only thing for which they

are really suitable is something that will seldom be written but frequently read -- i.e., the OS itself.

And that number of finite writes is ?

Lynn

 

[David Brown]

And that number of finite writes is ?

For a modern SSD disk, more writes than you could get through in years
of continuous writing.

In other words, unless you are working with very old SSDs (or perhaps
very cheapo devices), or continuously (24 hours a day) writing at high
speed, then flash wearout is a thing of the past.

 

[DevilsPGD]

In message <[email protected]> "Percival P.

It was suggested to me recently that, because of the finite number of
writes that SSDs will survive, the only thing for which they are really
suitable is something that will seldom be written but frequently read --
i.e., the OS itself.

You probably should stop listening to whoever made that suggestion.
Modern SSDs have write cycles measured in the millions of writes.

http://www.storagesearch.com/ssdmyths-endurance.html gives a good
breakdown, but to save you some reading:

| 2 million (write endurance) x 64G (capacity) divided by 80M bytes / sec gives the endurance limited life in seconds.

Or to phrase that in a form that Google Calculator or Wolfram|Alpha can
understand:

2 million * 64GB / (80MB/s)

<http://www.google.ca/#hl=en&safe=off&q=2+million+*+64GB+/+(80MB/s)>
<http://www.wolframalpha.com/input/?i=2+million+*+64GB+/+(80MB/s)>

Assuming you write to your SSD 24/7 continually, without pause for any
reason, and without taking time to wipe into consideration, you'll want
to start stressing out in 40 years or so.

In practice, most SSDs have some form of wear leveling in play too,
either out of a desire to wear level or alternatively just a desire to
optimize writes to avoid having to flush blocks prematurely due to the
fact that blocks can only be flushed in larger chunks.

SSDs have their issues (mainly in the $/GB area, although the sector
alignment issue is a bit annoying on out of date operating systems, but
you can work around it), but write cycles isn't one of them unless you
have very early generation gear, or very very cheap parts.

In practice, most SSDs that I've seriously considered have had 3-5 year
warranties, which probably covers their useful life anyway.

 

[Rod Speed]

Timothy Daniels wrote

Lynn McGuire wrote

Personally, I'd get two (one for the page file).

Thats the last thing you should put on one unless you plan on replacing it often.

And it makes a lot more sense to replace the use of the page file with more system ram anyway.

 

[Lynn McGuire]

For a modern SSD disk, more writes than you could get through in years of continuous writing.

In other words, unless you are working with very old SSDs (or perhaps very cheapo devices), or continuously (24 hours a day) writing
at high speed, then flash wearout is a thing of the past.

That is what I thought.

Thanks,
Lynn

 

[DevilsPGD]

In message <[email protected]> "Rod Speed"

Timothy Daniels wrote


Thats the last thing you should put on one unless you plan on replacing it often.

Why?

The pagefile is a perfect example of something that can and should be
placed on a modern SSD. Pagefile performance is critical when you're in
a low-memory situation, and an SSD will help speed things up
sigificantly.

And it makes a lot more sense to replace the use of the page file
with more system ram anyway.

True, to a point.

If you've maxed out your hardware's capabilities or only need to catch
rare/occasional edge cases, a pagefile will do the job nicely. If
you're talking day to day in-use application memory being paged out, you
really need more RAM before bandaiding over the problem.

 

[Arno]

In message <[email protected]> "Rod Speed"


The pagefile is a perfect example of something that can and should be
placed on a modern SSD. Pagefile performance is critical when you're in
a low-memory situation, and an SSD will help speed things up
sigificantly.

I agree. In fact I have the page-file on my SSD. For traditional
FLASH that was a sure way to kill the device fast. But with modern
wear-leveling, it is going to take several years of constant
pageing to kill the device. And constant pageing is a
pain even with an SSD. What benefits most in my case is when I
switch focus between different Apps. That has gotten a lot
faster.

Note that this applies to MS OSes. On Linux you have sharing
for most things anywasy and a lot lower memory requirements
in general. It has been some time that I saw any swap usage
at all.

Arno

 

[Arno]

On 08/05/10 01:32 pm, Timothy Daniels wrote:

It was suggested to me recently that, because of the finite number of
writes that SSDs will survive, the only thing for which they are really
suitable is something that will seldom be written but frequently read --
i.e., the OS itself.

Historic information. Not correct anymore. Still relevant to a degree
for USB flash.

Arno

 

[Rod Speed]

DevilsPGD wrote

Why?

Because its used more than the rest of the drive for writes, in most real world situations.

The pagefile is a perfect example of something
that can and should be placed on a modern SSD.

Wrong, it makes a lot more sense to have more physical system ram instead
so the page file doesnt get used at all. MUCH cheaper and much faster too.

Pagefile performance is critical when you're in a low-memory situation,

So it makes a lot more sense to not have a low memory situation instead.

and an SSD will help speed things up sigificantly.

Enough system ram so the page file doesnt get used will speed it up MUCH more for a lot less money.

True, to a point.

Its always true unless it isnt possible to add more system ram for some reason.

If you've maxed out your hardware's capabilities

You hardly ever are in that situation and it makes a lot more
sense to replace the hardware so it doesnt have that situation
than to go for an expensive SSD for the page file anyway, because
enough system ram so the page file doesnt get used costs a lot
less even if the motherboard has to be changed to allow that.

or only need to catch rare/occasional edge cases,

Then the speed isnt a problem, so there is no point in an expensive SSD for just the page file.

a pagefile will do the job nicely.

Wrong, its either a completely ****ed kludge if the page file is
used a lot or a complete waste of money if its hardly ever used.

If you're talking day to day in-use application memory being paged
out, you really need more RAM before bandaiding over the problem.

What I said in a lot more words.

 

[David Brown]

DevilsPGD wrote


Because its used more than the rest of the drive for writes, in most
real world situations.

As has already been noted, modern SSDs have effectively unlimited write
lifetime, unless you are doing something truly exceptional. So please
stop worrying about writing to flash drives.

Wrong, it makes a lot more sense to have more physical system ram
instead so the page file doesnt get used at all. MUCH cheaper and
much faster too.

Physical ram is faster - no doubts there.

But cheaper? That's only true if you are talking about a GB or 2. But
since most people are running 32-bit OS's, upgrading beyond about 3.5 GB
means installing a whole new 64-bit OS - buying an SSD is going to be a
lot cheaper if you value your time. And once you already have 6 GB or 8
GB, a memory upgrade probably means buying a full new set of DIMMs for a
lot more than an SSD.

So it makes a lot more sense to not have a low memory situation
instead.

If you are living in a Windows world, that's the case - swap is only
used if the OS runs out of real memory. But with Linux, swap space is
used if the real memory can be used for other purposes. Thus
little-used programs may be moved from memory to swap to free up space
for more disk cache. And it can be particularly useful for using tmpfs
for temporary directories - the temporary files will then be in ram, and
moved into swap if ram space is getting low.

Enough system ram so the page file doesnt get used will speed it up
MUCH more for a lot less money.



Its always true unless it isnt possible to add more system ram for
some reason.

I find cost, memory slots, and 32-bit OS limitations to be pretty good
reasons. Roughly 3.5 GB is the limit for 32-bit OS (depending on the
graphics card memory space), and 8 or 12 GB is the practical and
economic limit for 64-bit OS unless you are talking about serious
workstations or servers.

You hardly ever are in that situation and it makes a lot more sense
to replace the hardware so it doesnt have that situation than to go
for an expensive SSD for the page file anyway, because enough system
ram so the page file doesnt get used costs a lot less even if the
motherboard has to be changed to allow that.

Changing motherboard is seldom going to be cheaper, unless you happen to
have a particularly limited motherboard and there's a simple upgrade
path while keeping other components the same. In most cases, changing
motherboards is a time-consuming and risky process, while adding an SSD
is generally quick, simple, and low risk.

And if you already have 4 GB installed and a 32-bit OS, getting more
memory is very costly.

Then the speed isnt a problem, so there is no point in an expensive
SSD for just the page file.

You wouldn't use an SSD /just/ for swap. Even if you don't want to go
through the process of moving the OS or program directories over to the
SSD, you can easily move things like temporary directories and browser
caches to speed things up (these are read and written more than swap in
most cases). Data files are also typically painless to move, and may
benefit from the extra speed.

Wrong, its either a completely ****ed kludge if the page file is used
a lot or a complete waste of money if its hardly ever used.

You're a windows man to the core, aren't you? Historically, swap was -
as you say - a kludge when it was added to windows. For other systems,
swap has always been seen as a layer in the hierarchical memory system
and it is still useful today.

It is true that swap is not as relevant now (especially with windows) on
most systems as it used to be - the key reason being that the speed
difference between ram and disk has increased dramatically making swap
even slower (relatively speaking).

 

[Rod Speed]

David Brown wrote

Rod Speed wrote

As has already been noted, modern SSDs have effectively unlimited write lifetime,

Thats a lie with a system that is using the page file a lot because it doesnt have enough system ram.

unless you are doing something truly exceptional.

Thats overstating it.

So please stop worrying about writing to flash drives.

No thanks, its a lie.

Physical ram is faster - no doubts there.

So it makes absolutely no sense to be using a SSD instead
of more system ram if the system is using the page file a lot.

But cheaper? That's only true if you are talking about a GB or 2.
Nope.

But since most people are running 32-bit OS's, upgrading beyond about 3.5 GB means installing a whole new 64-bit OS -
buying an SSD is going to be a lot cheaper if you value your time.

Mindlessly silly.

And once you already have 6 GB or 8 GB,

You arent likely to be using the page file enough to matter if you have that much system ram.

a memory upgrade probably means buying a full new set of DIMMs for a lot more than an SSD.

Wrong, because with that much ram, you arent likely to be using the page file
much at all, so a SSD just for the page file will be a complete waste of money.

If you are living in a Windows world, that's the case - swap is only
used if the OS runs out of real memory. But with Linux, swap space is used if the real memory can be used for other
purposes.

Still makes a lot more sense to have real memory instead of a
SSD for the page file, if only because real memory is a LOT faster.

Thus little-used programs may be moved from memory to swap to free up space for more disk cache.

Not if they are properly designed and there is enough real memory.

And it can be particularly useful for using tmpfs for temporary directories - the temporary files will then be in ram,
and moved into swap if ram space is getting low.

Still makes a lot more sense to design the app properly or use one thats designed properly.

I find cost, memory slots, and 32-bit OS limitations to be pretty good reasons.

More fool you, its a monster kludge.

Roughly 3.5 GB is the limit for 32-bit OS (depending on the graphics card memory space), and 8 or 12 GB is the
practical and economic limit for 64-bit OS unless you are talking about serious workstations or servers.

And **** all thats well designed with 8-12GB uses the page file much
and what does needs a serious workstation or server ANYWAY.

Changing motherboard is seldom going to be cheaper,

Oh bullshit.

unless you happen to have a particularly limited motherboard

You must do if you cant have a decent amount of real memory.

and there's a simple upgrade path while keeping other components the same. In most cases, changing motherboards is a
time-consuming and risky process,

Complete and utter drivel.

while adding an SSD is generally quick, simple, and low risk.

More drivel.

And if you already have 4 GB installed and a 32-bit OS, getting more memory is very costly.

More drivel.

You wouldn't use an SSD /just/ for swap.

THATS the scenario being discussed, still visible right at the top.

Even if you don't want to go through the process of moving the OS or program directories over to the SSD, you can
easily move things like temporary directories and browser caches to speed things up (these are read and written more
than swap in most cases).

Makes a hell of a lot more sense to have enough real memory instead.

Data files are also typically painless to move, and may benefit from the extra speed.

Not compared with enough real memory SO THE PAGE FILE DOESNT GET USED.

You're a windows man to the core, aren't you?
Nope.

Historically, swap was - as you say - a kludge when it was added to windows. For other systems, swap has always been
seen as a layer in the hierarchical memory system and it is still useful today.

Pigs arse it is compared with ENOUGH PHYSICAL RAM SO THE PAGE FILE ISNT USED.

And the OP is discussing Win ANYWAY.

It is true that swap is not as relevant now (especially with windows)
on most systems as it used to be - the key reason being that the speed difference between ram and disk has increased
dramatically making swap even slower (relatively speaking).

So it makes a hell of a lot more sense to use the fastest physical
ram instead of farting around with the MUCH slower SSD even now.

You've got done like a ****ing dinner, yet again.