Intel Responds To X25-M Fragmentation Issue 111
Vigile writes "In mid-February, news broke about a potential issue with Intel's X25-M mainstream solid state drives involving fragmentation and performance slow-downs. At that time, after having the news picked up by everyone from CNet to the Wall Street Journal, Intel stated that it had not seen any of these issues but was working with the source to replicate the problem and find a fix if at all possible. Today Intel has essentially admitted to the problem by releasing a new firmware for the X25-M line that not only fixes the flaws found in the drive initially, but also increases write performance across the board."
Please, no more smokescreens. (Score:4, Interesting)
This was forseen: Intel will ultimately be forced to redesign their flash write algorithms [kerneltrap.org]
The point of this is, please please please if you are an engineering manager, when you make a collective booboo, no smoke screen please! It is unlikely to go unnoticed, and nothing positive will be achieved for you, your company, your potential customers or your tech audience. Instead, just come clean, admit the problem and get busy on the fix. Down that path lies increased trust, whereas the doublespeak path only erodes credibility. I certainly will be double checking any future claims, because of how this played out.
Anyway, big props to the team for implementing what appears to be a superior solution. Hey, how about just open sourcing that firmware and let everybody help make it even better? Just a thought.
Anandtech (Score:5, Interesting)
On this subject: I finally got around to reading Anandtech's very long article [anandtech.com] about the current crop of SSD drives. I feel like it was pretty educational, which is good because it took a long time to digest.
In its discussion of performance degradation as drives are used, the article explains that individual pages of NAND memory can't be rewritten. Early in a drive's life, page are remapped when they are rewritten by the OS. As the drive is used, the drive runs out of pages to remap and is forced to copy a block (typically a 512KiB collection of 4KiB pages) to cache, erase the block and then rewrite the block with the new pages. That explains pretty well why write performance degrades, since writing to a block that has data must perform a read and erase operation in addition to the write. However, that explanation also leaves open the question of how the drive prevents data loss if it loses power. Worst case, the OS issues a write and the drive copies a 512KiB block to cache and erases the block, and then loses power. Due to remapping, literally anything could be in that half a MiB. The data loss could corrupt the file that was being modified, obviously, but also any other file on the drive, or parts of the filesystem itself.
I figure there's got to be protection against data loss built-in, but I'm not able to find details regarding any individual drive or manufacturer's approach to solving that problem. Does anyone know more about this subject?
Every SSD has this problem (Score:5, Interesting)
We've all read this by now, right?
http://www.anandtech.com/storage/showdoc.aspx?i=3531 [anandtech.com]
The X25 has the same problem as all the other flash drives due to the need to erase in big chunks. Post-slowdown, the X25 is still faster than almost any other SSD that's brand new, and given the same usage, the X25 maintains the huge performance advantage it has from the start. I doubt Intel can really do much to improve this behavior without using TRIM.
I assume their "fix" will be slight tweaking of writing patterns done mostly to fool the mainstream press that had already been acting foolish by picked up this story without noticing the subtleties (such as the problem being present in all SSDs)
Re:Every SSD has this problem (Score:3, Interesting)
The problem Intel fixed is not the same thing you're thinking of. Anand's methodology was flawed, in that he was writing the OS back to the drive in sector-by-sector mode, which is effectively a large sequential write. This acts to heal drives that write combine and is not in line with how that OS would have got there in reality. The subsequent writes he did accomplished nothing more than seeing how far that particular drive could fill the 'holes' in the partition (i.e. how fast it can perform small random writes).
Allyn Malventano Theory - It Can Write Faster! (Score:4, Interesting)
The most interesting thing is the last section on the last page.
PC Perspective: Intel Responds to Fragmentation with New X25-M Firmware - My Theory - It Can Write Faster [pcper.com]
by Allyn Malventano 2009-04-13
I think that Allyn is onto something because if you look at the graph for write speed of the X25-M (MLC) it seems utterly perfect at 80 MB/s, almost like there is an artificial cap on the speed, while the one from the X25-E (SLC) series it produces a standard waveform, like Allyn pointed out, and not an artificial flat line.
I too believe that Intel is artificially capping the performance of this drive and they might decide to uncap it sometime in the future once the competitors start snapping at their heels or if enough time goes by and they decide to introduce a new SSD MLC based performance/server oriented product line and remove the cap then. This is very similar to the situation with processor multiplier locks that they remove in their performance oriented Extreme processor lines.
I frankly don't like this kind of behavior from Intel since they know that they have the upper hand so they are just doling out enough performance to beat the competitors and to satisfy the current customers but at the same time holding back to create a market for their X25-E product line with slightly higher performance.
I think the other shoe will drop sooner or later on the 80 MB/s cap.
Research
I've been doing research into Solid State Disks in the last few weeks and this article is yet another one of those for Required Reading in the course of learning about SSD. I've even wrote a detailed post with links to reviews and articles. You can read up on the linked articles to get a good primer on things.
Solid State Disk Benchmarks [slashdot.org]