Einstein on SSD's

Dana
Dana
Joined: 11 Sep 06
Posts: 44
Credit: 4303113
RAC: 0
Topic 194826

Hello,
I'm not certain this is a problem for me but I was hoping someone out there would know. I have an Intel gen 2 SSD that my operating system is installed on along with an overclocked i7 and the rest of the goodies. I was wondering if I may be impacting the life of my SSD to an extent that is unwise and unwarranted since I believe Einstein@Home to be writing to it all the time. I run the system 24/7 with Einstein running the whole time. In relation to the Windows 7 operating system, could Einstein@Home be using a significant percentage of the writes if not much else is being done on the system a lot of the time? If so, what would be the easiest way to move the whole thing to another drive without messing stuff up? I apologize if this topic has already been covered somewhere else.

Thank you for any help,
Dana

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1364
Credit: 3562358667
RAC: 1580

Einstein on SSD's

Quote:

Hello,
I'm not certain this is a problem for me but I was hoping someone out there would know. I have an Intel gen 2 SSD that my operating system is installed on along with an overclocked i7 and the rest of the goodies. I was wondering if I may be impacting the life of my SSD to an extent that is unwise and unwarranted since I believe Einstein@Home to be writing to it all the time. I run the system 24/7 with Einstein running the whole time. In relation to the Windows 7 operating system, could Einstein@Home be using a significant percentage of the writes if not much else is being done on the system a lot of the time? If so, what would be the easiest way to move the whole thing to another drive without messing stuff up? I apologize if this topic has already been covered somewhere else.

Thank you for any help,
Dana

Current generation consumer SSDs are good for 10k writes and the firmware is able to spread them out evenly. If you wrote 20GB/day, your SSD should last 4000days, or about 12 years. Under normal circumstances a typical user doesn't even hit 1GB/day. Short of thrashing your swap file to death there's nothing to worry about.

Even then, there are a few sites that have tested SSDs by constantly writing to them. Because of the way SSDs fail (each flash cell slowly but steadily increases in the amount of time it takes to write) the controller is able to ID failing cells and take them out of use before they go bad. At the offical 10k writes point the drives still had 90% usable capacity.

Realistically the only way to write a consumer SSD to death is to run a write intensive database off of it, and that's why server class SSDs use a higher grade flash that has only half as much capacity per chip but is good for 100k writes.

Dana
Dana
Joined: 11 Sep 06
Posts: 44
Credit: 4303113
RAC: 0

Hello, Thank you for your

Message 97372 in response to message 97371

Hello,

Thank you for your informative reply. It made me wonder if, to an equal extent, it is the number of writes that will degrade an SSD. What I have been doing is leaving a small (512Mb) swap file on the SSD and made a large (6GB to equal amount of RAM) swap file on a different drive. That was the best solution I could come up with which I thought would limit writes to the SSD swap file.

I did not understand your math very well to arrive at the 12 year figure. Is that 10,000 write figute you quoted per day, per year or lifetime? Could you help me understand that part a little better?

Thank you again,
Dana

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1364
Credit: 3562358667
RAC: 1580

RE: Hello, Thank you for

Message 97373 in response to message 97372

Quote:

Hello,

Thank you for your informative reply. It made me wonder if, to an equal extent, it is the number of writes that will degrade an SSD. What I have been doing is leaving a small (512Mb) swap file on the SSD and made a large (6GB to equal amount of RAM) swap file on a different drive. That was the best solution I could come up with which I thought would limit writes to the SSD swap file.

I did not understand your math very well to arrive at the 12 year figure. Is that 10,000 write figute you quoted per day, per year or lifetime? Could you help me understand that part a little better?

Thank you again,
Dana

Lifetime. Which means a 80GB SSD is good for 800,000GB of data writing.

[pre]
800,000GB of writes 1 day 1 year 4000 years
------------------ * ----------- * -------- = ---------- =
1 SDD lifetime 20GB writes 365 days 365 SSD lifetime
[/pre]

== 10.95 years to wear out the SSD, or that my mental math in the prior post was off by about 9% because I made a stupid mistake.

EDIT: fixed formatting

Dana
Dana
Joined: 11 Sep 06
Posts: 44
Credit: 4303113
RAC: 0

Conversely, 10,000 writes

Conversely, 10,000 writes divided by 100 writes per day = 100 days. Perhaps I'm confused about what the definition of a "write" is.

transient
transient
Joined: 3 Jun 05
Posts: 62
Credit: 115835369
RAC: 0

A write is not defined as

A write is not defined as changing the harddrive as a whole. It is changing a small part of it, say: a single byte. I don't know the actual definition. It could be less; a bit. Or it coukd be more; a group of bytes. I don't know what constitutes a single unit on a SSD.

hotze33
hotze33
Joined: 10 Nov 04
Posts: 100
Credit: 368387400
RAC: 1355

I have also the 80GB G2 Intel

I have also the 80GB G2 Intel SSDs on my computers for 1/2 a year now. So far no problems. But I have changed BOINC to write only every 5 minutes to the disk instead of every minute.
In my opinion there are major problems that arise.
First the SSD got 80GB (74GB real useable) but after the installation of the OS and some programs there are only 30 GB left. So 11 years become more like 4 years.
Second thing: What truly kills a SSD are the random writes. I think for the 80GB models that is 3.5TB. If you use a tool to benchmark the random writes, you can kill the SSD in under a day (because the random writes are so high 50MB/s).
So after scaring everybody: The SSD needs unused space to compensate the broken blocks. It doesn´t use the free space in a partition but the space, which is unpartitioned. So to enhance your drives lifespan, you have to fully erease your SSD and instead of building a 74 GB partition you create a 65GB partition. So the spare space is doubled and so the lifespan (for typical usage). If you know, that you don´t need the full space on the drive and don´t partition the full disk. (This adds to the costs of a SSD).
The good thing about SSDs is, even if you can´t write to the disk anymore, you can still read the data.
For me I don´t care about this problem. If the disk holds for 2 years in my work pc it is ok, because then there will be much faster (SATA 3 drives) on the market for the same price.
And I would never go back to mechanical disk for the OS drive.
good source for SSDs: anandtech.com->storage and http://www.anandtech.com/storage/showdoc.aspx?i=3531
OT: @ Dana: What is the speed of your Core i7? We are fighting for the 1st place of the fastest single socket system and I just want to know;)

Ver Greeneyes
Ver Greeneyes
Joined: 26 Mar 09
Posts: 140
Credit: 9562235
RAC: 0

RE: What I have been doing

Message 97377 in response to message 97372

Quote:
What I have been doing is leaving a small (512Mb) swap file on the SSD and made a large (6GB to equal amount of RAM) swap file on a different drive.


The swap file should only -ever- be used if you run out of RAM. If that happens - let's face it - you're likely to need a lot more than 512MiB to help out whatever application is using up so much RAM. It is, however, a good idea to look up how to stop Windows (if that's what you're using) from using the swap file for system services - this will make your system more responsive, and stop it needlessly causing I/O activity.

Dana
Dana
Joined: 11 Sep 06
Posts: 44
Credit: 4303113
RAC: 0

Hello Hotzee33, I was

Message 97378 in response to message 97376

Hello Hotzee33,

I was very much hoping to run in to you at some point. I know this violates rules of changing the thread but I must since I can't figure out how to send you a message. I was looking at the stats for the top Einstein RAC systems (http://www.statsnstones.com/Hosts.aspx?sort=1&projid=12&page=0) and found you near the top in the number 4 overall spot. These rankings change quite often but at this particular moment I'm only a couple spots behind you with my (air cooled) OC'd i7. Somehow you are in the top five overall with a Q6600. I have a 3.48Mhz (low end liquid cooled) Q6600 and manage around a 3600 RAC compared to your 6700 RAC . I guess I won't ask you for all your secrets but I'm quite impressed and bewildered at the same time that I can't catch you with my i7 that has the second overall RAC of all i7's. If you are at all feeling like sharing system building and OC experiences with each other please send me a PM (I can't figure out how to send you a message). And thank you very much for your helpful, as well as the others, input regarding the SSD situation. I will read the article you posted today. Since we have the exact same model I'd like to share SSD experiences too.

I again apologize for inserting this post. Hope I can be forgiven, or told a better way to accomplish the same goal,
Dana

P.S. the straight, honest answer is that my i7 is overclocked to exactly 3.923 GHz as measured by cpuz (19x206.5). But like I suggested, we really need to get together.

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1364
Credit: 3562358667
RAC: 1580

RE: A write is not defined

Message 97379 in response to message 97375

Quote:
A write is not defined as changing the harddrive as a whole. It is changing a small part of it, say: a single byte. I don't know the actual definition. It could be less; a bit. Or it coukd be more; a group of bytes. I don't know what constitutes a single unit on a SSD.

It's 10,000 writes for each cell of flash memory. In typical current SSD's the smallest chunk of disk space that can be written at a time is 4kb, and the smallest chuck that can be erased (neccesary prior to a write) is around 512k.

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1364
Credit: 3562358667
RAC: 1580

RE: In my opinion there

Message 97380 in response to message 97376

Quote:

In my opinion there are major problems that arise.
First the SSD got 80GB (74GB real useable) but after the installation of the OS and some programs there are only 30 GB left. So 11 years become more like 4 years.
Quote:

Not true. The controllers are smarter than you give them credit for. They implement wear leveling which occasionally moves static data around so that all of the flash is written at about the same time. On an SSD the write amplification (difference between how much you write and how much it writes) is only a few percent. A large part of how they do this is the use of scratch space. The drive might look full to your OS but it still has some spare space, typically 7% on consumer models so even when the drive is full and you make a small write it's only writing 4k of data somewhere, not erasing and rewriting 512k.


Second thing: What truly kills a SSD are the random writes. I think for the 80GB models that is 3.5TB. If you use a tool to benchmark the random writes, you can kill the SSD in under a day (because the random writes are so high 50MB/s).
Quote:

Again not true. Your random writes will be spread out over the entire drive. The fastest way to kill it would be with sequential writes at ~75MB/sec for an intel drive (~175MB/sec for an indilinx barefoot based drive). For the intel drive that's 6.4PB of writes/day or 81 complete writes to the drive. This would require ~4 months to hit the 10k writes the drive is rated for. People have done these sort of exercises, and at the 10k point the drives lost about 10% of their rated capacity.


So after scaring everybody: The SSD needs unused space to compensate the broken blocks. It doesn´t use the free space in a partition but the space, which is unpartitioned. So to enhance your drives lifespan, you have to fully erease your SSD and instead of building a 74 GB partition you create a 65GB partition. So the spare space is doubled and so the lifespan (for typical usage). If you know, that you don´t need the full space on the drive and don´t partition the full disk. (This adds to the costs of a SSD). There've been people who've done this, and at the 10k writes point the drives were still at ~90% of capacity because they removed the chunks of the flash chips that were dieing fastest from service before any data was lost. Unfortunately I haven't seen any tests that continued the run for some more months to see how quickly the remainder of the flash died. I'm curious about this, but not curious enough to destroy a $100+ device to find out.

That sort of activity is massively outside the bounds of normal consumer computer use. Unless you spend your time compiling very large software programs over and over (while forcing new writes of all the intermediate files each time) or doing video editing with huge numbers of saves, you'll only do a few to a few dozen GB/day of writes; at which point the flash chips will die of old age in a decadish, long before you reach their write lifetime (my post above contained a math error, under the circumstances described the drive would need 109 years, not 10.9 to reach the write limit).

The only real world situation where you would have anything approaching the theoretical maximum write rate of an SSD would on some types of servers. To protect against that they use flash chips that only have 50% of the capacity of what's used in consumer models but which good for 100k writes, which is sufficient to give the SSDs a longer useful lifetime than the servers they're installed in.

The only sort of flash you might be able to kill in a few days would be a memory card or a USB stick, because they use the cheapest, crappiest flash available, and their bottom spec controllers are less capable of minimizing the write amplification penalty.

The rest of your advise about how to paritition the SSD is nonsense because you'd have to be writing hundreds of gigabytes/day to write the chips to death before they died of old age.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.