SSDs - State of the Product?

[udaman]

shipping in volume? But I can't find them anywhere yet.

Given the Portege R600 laptop premium is $1400, don't think well see much of a price break on these for a good long time

Solid-state rivalry sizzles: Toshiba ships 512GB SSD
http://news.cnet.com/8301-13924_3-10303815-64.html

 

[LunarMist]

I vaguely recall something about a special formatting tool to align the logical clusters with the SSDs physical blocks. It was supposed to help, but I've never bothered with it as I am completely happy with the performance of my SSDs.

So you don't feel that the "alignment" mumbo jumbo is needed for the X25-E?

 

[LunarMist]

Possibly here, but I have not tested or verified legitimacy.

Holy cow, a bootable floppy is needed to maintain an SSD! Technology is out of whack.

 

[LunarMist]

Anyone found a test that puts SCSI 15k'S, in raided configuration, against the SSD's?

Oh Greg, I see you are not giving up. Which Cheetahs are you looking at, 15K.7, or something older?

 

[sechs]

I vaguely recall something about a special formatting tool to align the logical clusters with the SSDs physical blocks. It was supposed to help, but I've never bothered with it as I am completely happy with the performance of my SSDs.

Perhaps something called diskpart, which comes with Windows?

The theory is that you want to align the file system clusters with the erase blocks. Otherwise you end up tripping more read-write-erase cycles, since one cluster could span two blocks. So, part one is that the drive is faster after you've filled every block, as you won't need to do two rewrites; part two is that you wear the drive less because you don't do as much erasing and writing.

As I understand it, alignment isn't an issue with Windows Vista or 7. With second generation controllers, there doesn't seem to be nearly the performance hit for misalignment.

 

[mubs]

Holy cow, a bootable floppy is needed to maintain an SSD! Technology is out of whack.

A few years ago, I paid good bucks to buy a YE-Data External USB floppy drive. Works perfectly and is great for situations like this and for using floppies (when you have to) with laptops that don't have a FDD.

 

[mubs]

I seem to have caught the "random capitalization of words" disease from a co-worker :cyclopsani:

 

[Fushigi]

Anyone found a test that puts SCSI 15k'S, in raided configuration, against the SSD's?

IBM did. The white paper is about SSDs v. 15K v. 7.2K disks RAIDed in a SAN. Another example with more detail is Performance Impacts of Flash SSDs Upon IBM Power Systems (PDF). IO/s is stressed more heavily than MB/s, as it should be in a server environment.

IBM specs the Power system's RAID card (266MHz PCI-X, 1.6GB read cache, 3Gb SAS interface) as supporting up to 60 HDDs but only up to 8 SSDs. That right there should tell you about the ability of an SSD to saturate the host interface.

BTW, while we know Intel and the others present a formatted capacity of 10-20% smaller than the actual drive size, reserving 8+GB for internal overhead to maintain performance, the IBM SSDs under-report by closer to 50%. Unfortunately, Intel X25-E drives look really, really super duper cheap in comparison to the price IBM charges for their SSDs. Try an approximate $10K per 69GB drive. :colors:

 

[Pradeep]

Yes those IBM SSDs are massively over provisioned like the other enterprise drives. When you need a device to keep running you do what it takes.

Frankly the X25-E is enterprise in name only. I was just thinking what the hell TRIM or periods of poor write performance does to say an array of "E" drives running in a RAID. Nothing good that's for sure.

For RAID, dependability and consistency is a key factor (I would consider it the primary goal).

 

[LunarMist]

A few years ago, I paid good bucks to buy a YE-Data External USB floppy drive. Works perfectly and is great for situations like this and for using floppies (when you have to) with laptops that don't have a FDD.

I probably have a 2003 Sony external floppy drive around somewhere, but that Mickey Mouse alignment procedure is not reasonable for routine use. I'm having zero issues with the X25-E slowing down after 9 months of operation. I do not trust the Vortex drives to be as trouble free. I don't think the X25-M G2 has what it takes to handle the heavy page file/scratch/temp file workload. Meanwhile, it’s awful to keep restoring images every time I switch back and forth between XP32 and XP64. I really need two SSDs.

 

[Stereodude]

The alignment process isn't that hard. You don't need a floppy either. You just put the drive in another machine (where it's not the primary) to set it up.

 

[LunarMist]

It states that HDDErase must be run from a bootable DOS environment. Are you using a different app?

 

[Stereodude]

I don't think we're talking about the same thing. I was talking about using Diskpart to align the partition with the sectors of the drive. It can be used under Windows.

 

[sdbardwick]

HDDErase addresses a different issue than alignment; an inexact old-school analogy would be low-level format (remember when you needed to change sector interleave to match the HDD controller) vs. creating partitions.

Here's a link to OCZ's forum that shows an alignment method.

 

[LunarMist]

Oh. I don't want to align anything. I just want to improve performance at the hardware level every few months, or is that not necessary with the Intel X25-M G2? OCZ SSDs are too complicated and too much hassle. A 32GB X25-E would solve short-term needs (remainder of 2009), but it is too small for medium or long term use. I could live with an 80GB X25-M G2 as a second SSD and then repurpose it for another system when there are larger, cheaper and/or faster X25-E drives.

 

[udaman]

Oh. I don't want to align anything. I just want to improve performance at the hardware level every few months, or is that not necessary with the Intel X25-M G2? OCZ SSDs are too complicated and too much hassle. A 32GB X25-E would solve short-term needs (remainder of 2009), but it is too small for medium or long term use. I could live with an 80GB X25-M G2 as a second SSD and then repurpose it for another system when there are larger, cheaper and/or faster X25-E drives.

Not complicated, just use the new firmware (fully automatic, you do *nothing*):

OCZ and Indilinx Collaborate On New SSD Garbage Collection Scheme



http://hothardware.com/News/OCZ-and-Indilinx-Collaborate-On-New-SSD-Garbage-Collection-Scheme/

 

[ddrueding]

I've bought around 20 of the Vertex drives, with no issues (performance on reliability) so far.

 

[ddrueding]

Not complicated, just use the new firmware (fully automatic, you do *nothing*):

OCZ and Indilinx Collaborate On New SSD Garbage Collection Scheme



http://hothardware.com/News/OCZ-and-Indilinx-Collaborate-On-New-SSD-Garbage-Collection-Scheme/

Very interesting.

 

[ddrueding]

Interesting explanation of SSDs, block erasing, and TRIM.

 

[CityK]

The review was crud, but the (currently short) ensuing discussion is interesting as it points out a number of areas/things readers should be aware of in regards to reading SSD performance comparisons and determining whether the review was, more or less, on an apples-to-apples basis or bunk.

There is also some interesting info about Linux filesystem inclusion of TRIM support etc.

 

[udaman]

with 512GB SSD's currently insanely expensive compared to slower HD tech, can't be better news than this...except they need to hurry up, and lower prices *more*:


Intel, Micron Achieve Industry's Most Efficient NAND Product Using 3-Bit-Per-Cell Technology
http://www.nanowerk.com/news/newsid=12099.php

Intel Corporation and Micron Technology Inc. today announced the development of a new 3-bit-per-cell (3bpc) multi-level cell (MLC) NAND technology, leveraging their award-winning 34-nanometer (nm) NAND process. The chips are typically used in consumer storage devices such as flash cards and USB drives, where high density and cost-efficiency are paramount. Designed and manufactured by IM Flash Technologies (IMFT), their NAND flash joint venture, the new 3bpc NAND technology produces the industry’s smallest and most cost-effective 32-gigabit (Gb) chip that is currently available on the market. The 32Gb 3bpc NAND chip is 126mm². Micron is currently sampling and will be in mass production in the fourth quarter 2009. With the companies continuing to focus on the next process shrink, 3bpc NAND technology is an important piece of their product strategy and is an effective approach in serving key market segments. “We see 3bpc NAND technology as an important piece of our roadmap,” said Brian Shirley, vice president of Micron’s memory group. “We also continue to move forward on further shrinks in NAND that will provide our customers with a world-leading portfolio of products for many years to come. Today’s announcement further highlights that Micron and Intel have made great strides in 34-nanometer NAND, and we look forward to introducing our 2xnm technology later this year.” “The move to 3bpc is yet another proof point to the remarkable progress Intel and Micron have made in 34-nm NAND development,” said Randy Wilhelm, Intel vice president and general manager, Intel NAND Solutions Group. “This milestone sets the stage for continued silicon leadership on 2xnm process that will help decrease costs and increase the capabilities of our NAND solutions for our customers.” Source: Micron Technologies (press release)

 

[Bozo]

I just read at HardOCP that Intel has released a firmware update for it's SSD drives. Fixes a password problem?

 

[Gilbo]

Indilinx and OCZ's garbage collection firmware seems superficially effective.


The big problem for me, is that I can't conceive of how it works. A drive doesn't know anything about the data written to it, and it certainly can't know whether data previously written has been subsequently deleted, since all modern filesystems make that change in the filesystem tables, not by actually going out and scrubbing the file off the disk.

So, is Indilinx' firmware somehow going into the filesystem table and tracking metadata updates to allow it to do this? Sounds complicated... And if so, which filesystems does it support?

 

[sechs]

I agree. The drive can only do so much without knowledge of the file system on it.

 

[LunarMist]

I just want a Windows app that will wipe out all that stuff, even if it takes an hour.

 

[sdbardwick]

Indilinx and OCZ's garbage collection firmware seems superficially effective.


The big problem for me, is that I can't conceive of how it works. A drive doesn't know anything about the data written to it, and it certainly can't know whether data previously written has been subsequently deleted, since all modern filesystems make that change in the filesystem tables, not by actually going out and scrubbing the file off the disk.

So, is Indilinx' firmware somehow going into the filesystem table and tracking metadata updates to allow it to do this? Sounds complicated... And if so, which filesystems does it support?

My impression is that it functions similar to bad sector reallocation. Instead of bad sectors, it finds partially written blocks, relocates them to fill blocks, and then updates the sector id and erases the partially written block so they can be reused. Since nothing changes from the file system's point of view, it is independent of the OS/filesystem.

 

[P5-133XL]

My impression is that it functions similar to bad sector reallocation. Instead of bad sectors, it finds partially written blocks, relocates them to fill blocks, and then updates the sector id and erases the partially written block so they can be reused. Since nothing changes from the file system's point of view, it is independent of the OS/filesystem.

I think that is a good analogy.

 

[LunarMist]

I just read at HardOCP that Intel has released a firmware update for it's SSD drives. Fixes a password problem?

Maybe we will see them before 8/28?

 

[Gilbo]

My impression is that it functions similar to bad sector reallocation. Instead of bad sectors, it finds partially written blocks, relocates them to fill blocks, and then updates the sector id and erases the partially written block so they can be reused. Since nothing changes from the file system's point of view, it is independent of the OS/filesystem.

Thanks sdbardwick, that does make sense. From the review I linked too, it seems to function very well. I'd like to see a more detailed look from someone like Anandtech or Techreport though.

It certainly makes the Vertex more attractive. Of course the Intel drives are the same price now...

 

[ddrueding]

...but the Intel drives still suffer from low write speeds.

 

[LunarMist]

...but the Intel drives still suffer from low write speeds.

I recall in one of those umpteen links a test showed that the X25-M G2 was better at writes than similar drives, 30MB/sec. vs. 10MB/sec. Maybe it was in a dream.

 

[ddrueding]

I recall in one of those umpteen links a test showed that the X25-M G2 was better at writes than similar drives, 30MB/sec. vs. 10MB/sec. Maybe it was in a dream.

No, if you are talking about small files the G2 is much faster. If you are talking about large files (STR) than it is much slower.

Intel X-25M

• Avg Read = 229.8
• Avg Write = 77.6

OCZ Vertex

• Avg Read = 223.1
• Avg Write = 199.2

 

[LunarMist]

Well of course I meant the random writes. Question is always how sequential are the large writes as the drive fills or ages. Have you compared both?

 

[ddrueding]

I have. And I see similar numbers. Is there a particular test you would like run? I can do X25-Es and Vertex drives over the weekend. I'm out of X25-Ms at the moment (waiting for the high-capacity G2s).

 

[sechs]

My impression is that it functions similar to bad sector reallocation. Instead of bad sectors, it finds partially written blocks, relocates them to fill blocks, and then updates the sector id and erases the partially written block so they can be reused. Since nothing changes from the file system's point of view, it is independent of the OS/filesystem.

So, how does it know that a block is partially written if it doesn't know that some of the data has been deleted?

Recall that clusters are simply marked deleted in the file system metadata -- no actual change to the data is made. Unless the drive reads the metadata, then it doesn't know that a block is partially written -- unless it was never fully written. I imagine that's helpful, but all it's really doing is compacting the data, wearing the disk more in the process.

I read somewhere that the garbage collection scheme used by some Samsung controllers only works on NTFS drives. Perhaps they are actually using file system data to flush dirty blocks.

 

[Fushigi]

Regarding the wear on the drive, the MTBF is estimated to be 1,500,000 hours. That's 62,500 days, which is 8,928 weeks, which is 171 years. Even if the engineering (or marketing...) estimate was off by an order of magnitude that would still peg the average failure as occurring after 17 years.

I'm not concerned one bit about wearing out the drive.

 

[Mercutio]

You know that's not how MBTF works, right?

 

[ddrueding]

Interesting

Samsung has implemented a feature they call "auto-TRIM", which peeks at the NTFS $bitmap to clear up unallocated areas during idle time. The advantage is these drives will recover performance given sufficient idle time, and will do so on any platform. The catch is it only works with NTFS partitions, meaning Mac and Linux users will not share the benefit seen by Windows users.

 

[timwhit]

Interesting

I use some NTFS disks in Linux. Don't know if auto-TRIM would work though.

 

[sechs]

If it's looking at the system bitmap to check what areas of the disk are marked deleted, then its only dependent on having a compliant file system. The drive doesn't know anything about your operating system.