Next Generation SSDs Delayed Due To Vista 600
PoliTech notes in a journal entry that "Vista is the gift that just keeps on giving." "Speaking during SanDisk's second-quarter earnings conference call, Chairman and [CEO] Eli Harari said that Windows Vista will present a special challenge for solid state drive makers. 'As soon as you get into Vista applications in notebook and desktop, you start running into very demanding applications because Vista is not optimized for flash memory solid state disk,' he said... 'The next generation controllers need to basically compensate for Vista shortfalls,' he said. 'Unfortunately, (SSDs) performance in the Vista environment falls short of what the market really needs and that is why we need to develop the next generation, which we'll start sampling end of this year, early next year.' Harari said this challenge alone is putting SanDisk behind schedule. "We have very good internal controller technology... That said, I'd say that we are now behind because we did not fully understand, frankly, the limitations in the Vista environment.'"
It is not just vista... (Score:5, Informative)
For some reason 'rpm' from mandrake is surprisingly inefficient on SSD's. It makes mandrake practically unusable for me on my eeepc. Yet dpkg/apt-get/aptitude on debian and ubuntu is just zippy.
--jeffk++
that's one way to look at it (Score:5, Informative)
Another way to look at it is that SSDs aren't optimized for Vista.
Here's a basic issue with NAND. NAND is most efficient when written in chunks of at least 128KB in size. Some NAND chips aren't even efficient until 256KB. Because this is the smallest unit that can be erased in NAND. If you write a smaller amount (say 8KB), it actually has to erase a new block, copy 120KB to the new block from the old, then write in the new 8KB. Then, if you write another 8KB, might have to do it again!
So these SSDs would be fastest if Vista would write in larger blocks. Unfortunately, 512B is the block size for ATA. There are extensions for 2KB, 4KB and 8KB blocks, but Vista doesn't implement them. And it doesn't have to, as they're optional.
Also notable is that even some regular magnetic hard drives now have native 2KB or 4KB blocks and it is written in 512B chunks, it might have to do a read-modify-write cycle to do it.
Anyway, if you know ATA until recently the LARGEST possible write was 128KB (256 blocks), to expect Vista to use writes this large or larger when many drives (like almost any under 137GB) doesn't even implement them is perhaps too optimistic. To expect it to use 2KB or 4KB blocks when 95% of drives don't implement them is perhaps too optimistic.
In the end, drive (including SSD) companies can't operate in a vacuum. They know they have to make what is useful for the customer, which means usable by the OS.
As an additional note, MacOS recently (10.4.something) added support for 2KB, 4KB, etc. blocks, but it still has difficulty using large writes too. I think when operating through the file system, it never generates a write larger than 256 blocks either (which is 128KB or more depending on block size).
Re:Funny how Sandisk is the only one with this pro (Score:5, Informative)
Sandisk SSD drives are poorly made and perform poorly (much worse than others..). This is just Sandisk trying to shift the blame elsewhere..
DailyTech's article [dailytech.com] (and others [techspot.com]) have also added opinions similar to yours. From the DT article:
In fact, it's not uncommon to see SanDisk SSDs rank last in testing in almost every benchmark and by a large margin -- even in Windows XP. Recent testing showed that MSI's Wind netbook was no faster with a SanDisk SATA 5000 SSD [laptopmag.com] than with the standard 80GB HDD -- an Eee PC 1000h featuring similar specifications was significantly faster with a competing SSD from Samsung [laptopmag.com].
While Vista may be a performance inhibitor compared to Windows XP for SSDs, it appears that most new, current-generation SSDs are having no problems performing well with the operating system. The problem appears to be SanDisk's low reads and writes (67 MB/sec and 50 MB/sec respectively) compared to the competition (i.e., OCZ's new Core Series SSDs [dailytech.com] which clock in at 120 to 143 MB/sec for reads and 80 to 93 MB/sec for writes)."
Re:what about linux? (Score:5, Informative)
Re:Unbelievable (Score:5, Informative)
Every time you boot into Vista, god kills a little kitten!
Re:I wonder what he means (Score:4, Informative)
Vista does lots and lots of writing - especially lots of small writes... Then again so does XP - just Vista does more.
Trust me, Vista is vicious to a hard drive. I got a new Quad6600 with 3GB and it felt slow... sometimes absolutely crawling because it had a slower 8MB cache 500GB drive installed. I finally figured out that the HD was the performance bottle neck. I just bought a WD Velociraptor (10K RPM 32MB cache) for $300 and my computer feels about twice as fast for daily usage.
Re:I wonder what he means (Score:5, Informative)
You can see even when you think your computer should be idle that Vista has anywhere from several dozen to over a hundred outstanding writes queued up to the hard drive at just about any time.
Re:File swapping destroys SSDs (Score:5, Informative)
Today flash hard drives levy on technology used in older embedded devices that relied on flash, called "wear leveling".
Because each write is spread out throughout the entire disk, you don't physically write to the same sector X thousands of times when updating a cache file or whatnot.
Even if you had something thrashing the SSD continuously, you would not destroy the drive within the reasonable lifespan of a comparable rotating media drive.
Re:Unbelievable (Score:4, Informative)
http://www.modaco.com/content/asus-eee-pc-http-www-eeeasy-com/261965/installing-vista-on-the-eee-ive-done-it-and-it-works/ [modaco.com]
So it looks like it is possible...
Not rushing to do it on my Eee though!
SANDISK has been caught in a lie here. (Score:5, Informative)
http://www.sandisk.com/Corporate/PressRoom/PressReleases/PressRelease.aspx?ID=3785 [sandisk.com]
Re:that's one way to look at it (Score:1, Informative)
Vista does support disks with sector size up to 4Kb. It can't go to 8Kb because it requires the ability to write an x86/amd64 page (4Kb) atomically.
Re:that's one way to look at it (Score:3, Informative)
Maybe, but it is insane to develop hardware to suit software. Far too many types of software will want to use the same hardware, you can't optimize for them all. It is far more logical, far more rational, to optimize the hardware for the task, and leave it to software on the device or drivers on the host machine to present a suitable view for the operating system. Any remaining problems are for the OS to take care of. If the OS doesn't, that's the OS' problem, not the hardware's.
In this case, let's take the thrashing problem. If the driver did not provide write-through, knew enough to distinguish data from indexes and had sufficient ramdisk to work with, it should be possible to eliminate writes until the last possible moment, and to maximize the number of whole-block writes. Ideally, you'd mirror the entire drive in RAM, do everything in a ramdrive, then reorganize and write only once at the very end. In practice, you won't have that much RAM, but you should have enough to absorb unnecessary or inefficient writing methods.
Comment removed (Score:4, Informative)
Vista is the same as XP in this respect (Score:4, Informative)
The problem SanDisk had is they expected the OS to batch writes to an erase block size (at least 128Kb) and were surprised to find this isn't how operating systems typically work. That's not specific to Vista; it applies to every previous release of Windows, and most other mainstream OSes.
On random writes, the performance of SSDs is terrible, since they need to perform read/modify/write on every small write. So sequential write performance looks fine, and random write performance looks bad.
What filesystem guarantees to write its metadata (directories, bitmaps, etc) in 128/256Kb chunks? None do. Every time the filesystem writes a small chunk of data, the disk has to work extra hard. Any app writing small, random chunks also performs badly (eg Outlook); this is true on XP and Vista (equally.)
Really, SanDisk would have been well advised to speak to OS developers (any) before releasing their first attempt at and SSD. Experience with removable flash (typically file copies) does not equate to experience with fixed disk scenarios (eg registry & log flushes.)
Re:Unbelievable (Score:3, Informative)
Ahh, but SSD's will consume less power.
Can I get a +1 backOnTopic please :-)
Re:Unbelievable (Score:5, Informative)
I'm running Ubuntu 8.04 on my desktop computer at the moment. That is,
P4 2.66GHz
512MB of RAM
GeForce2 MX400 graphics card
No overclocking, no tricks, running the latest version of Ubuntu with far more 3d eye-candy than Aero is capable of, every service on, a crap load of extra packages installed, including server software (such as mySQL and Apache) running in the background, running Firefox with 10 tabs on one desktop, Evolution on another, xChat on the third, and Rhythmbox, Skype and Pidgeon on the fourth, and it's still nice and responsive.
I'd be lucky to get Vista to even install, let alone run Aero and programs as well...
Re:Unbelievable (Score:5, Informative)
How can this be informative?
I'm afraid some moderators have a sense of humour ;-)
Re:File swapping destroys SSDs (Score:2, Informative)
Only with SLC NAND Flash (with 100,000-millions of writes per sector).
Some of the cheaper ones coming out now that people might actually use are based on MLC NAND Flash, which still has a short (10,000 write) cycle. There could well be a reliability problem there.
Also, unfortunately, some of the wear levelling algorithms suck, and remember they don't perform well when the disk is nearly full. They could do a lot better than they do at the moment, and include a bank of spare blocks (and make that available over SMART using the reallocation count so you'd have plenty of warning when the disk was dying). There are improvements they need to make in that department.
However, yes - compared to a rotating hard disk which (in my experience) generally has a 1-5 year lifespan, I would expect MLC NAND Flash drives to fall in about the middle of that range (so not really being worse than a hard disk), and SLC to greatly exceed it. Besides, when they fail, it's writes that fail. You still won't lose the data.
At this time, I'm in no doubt - my next machine will have a Flash system disk.
Re:Unbelievable (Score:5, Informative)
Linux already runs just fine on Flash devices, and has done for many years - there are filesystems optimised for flash, and many embedded devices that use Linux on Flash, e.g. GPS devices (TomTom, Garmin), WiFi/DSL/Cable routers (most of them), etc, etc. There are also consumer distros that run really well from USB flash drives, e.g. Damn Small Linux, Slack, Puppy and many others.
Re:It is not just vista... (Score:4, Informative)
NTFS, Linux, and Modern Filesystems (Score:5, Informative)
While Linux has modern filesystems and gets optimized and fixed almost constantly, Windows Vista still uses the same basic NTFS layout and associated algorithms that were finalised around 10 years ago, and weren't even very good back then. There have been only very minor revisions to NTFS and virtually none of them have improved its performance or reduced its fragmentation.
I don't know if you're blatantly lying or just very misinformed.
Let's take age and revisions first. Ext2 was introduced to Linux in January 1993. NTFS was introduced to Windows in July 1993 (in NT 3.1). So your implication that NTFS is much older than ext is nonsense.
You say that there have been "only minor" revisions to NTFS in comparison to ext2. Ext2 has in fact had only one (stable) revision, ext3, and it introduced only one new feature, journalling (something NTFS has had from the start). Various new revisions of NTFS, on the other hand, have added: transparent compression, named streams, disk quotas, filesystem-level encryption, sparse files, reparse points, update sequence number journaling, $Extend, distributed link tracking, and atomic transactioning, among others.
Some of these features, such as sparse files, are things that ext2 has had from the start. But many, such as transparent compression and file-system level encryption, are not only not, but have even now not found their way into mainstream Linux. To take those two features as an example, the only filesystems even close to mainstream that have them are Resier4 and ZFS, neither of which are ready for widespread use in Linux.
You say "Vista still uses the same basic NTFS layout and associated algorithms that were finalised around 10 years ago" -- conventiently not mentioning that that that 'ten-year-old layout policy' uses a number of modern layout features, such as extents, that have also still not yet found their way into mainstream Linux (ext4 and Reiser4 both support them, but neither are yet out of beta; neither ext3 nor ReiserFS 3 do). Directory contents in NTFS, incidentally, is stored as a B+ tree, which is the same structure that ReiserFS uses due to its scalability.
Re:File swapping destroys SSDs (Score:3, Informative)
I'm very dubious about your statement that you get only 10K writes per "erase block" (e.g. 128 KB) on MLC - that would destroy its use for many applications, and I believe all flash devices are quoted per "block" e.g. 4 KB, not the erase block. Most analyses I've seen show that there is nothing to worry about with typical OS usage patterns on flash drives.
As for Unix/Linux writing the access time back all the time - this happens only every 5 seconds with ext3 (default config), and less often with ext2. You can disable this completely by mounting all filesystems with "noatime" to prevent these updates, which is recommended on hard disks as well to improve performance.
Re:Pointing fingers (Score:3, Informative)
WinFS was never a file system, it was a layer that got inserted on TOP of the existing NTFS filesystem and the technology is still used by Microsoft today, just for different applications.
Re:Ya, it is Vista's fault... (Score:5, Informative)
It is quite true that SanDisk's SSD are woefully subpar in performance when running Windows Vista. Numerous benchmarks from around the web have shown SanDisk SSDs getting outpaced by the competition.
While Vista may be a performance inhibitor compared to Windows XP for SSDs, it appears that most new, current-generation SSDs are having no problems performing well with the operating system. The problem appears to be SanDisk's low reads and writes (67 MB/sec and 50 MB/sec respectively).
Re:Famous quote (Score:4, Informative)
> "The Wise adapts himself to the world. The Fool adapts the world to himself. Therefore, all progress depends on the Fool." -Unknown
Perhaps you are misquoting George Bernard Shaw:
"Reasonable people adapt themselves to the world. Unreasonable people attempt to adapt the world to themselves. All progress, therefore, depends on unreasonable people."
Comment removed (Score:5, Informative)
Re:File swapping destroys SSDs (Score:3, Informative)
No, this statement will not be put to bed, because it is based on facts - measured physical quantities. And here's one thing to ponder: if an application writes to the disk 100 times per second, how much will your 4GB SSD going to last? If you have only 1GB of space left, then wear leveling can only count on the blocks that don't contain data. And if the blocksize for the Flash RAM device is 128KB (which is typical, but there are also 256KB Flash RAMs), then the number of blocks you can spread out the writes is 8192. If the SSD is based on MLC Flash (as is, sadly, becoming typical) then you can write up to 10.000 times per block. Assuming perfect wear leveling, the device will last less than 819200 seconds which is 9 days and a few hours.
You fail to consider several things:
1. Static wear levelling/leveling rotate the blocks being written to so both "empty" and "full" blocks are being used, so the amount of free space on the filesystem doesn't matter.
2. The 100.000 writes often quoted are a guaranteed statistical _minimum_, not a average or a maximum. According to some sources the typical cell will endure 200K-1M writes.
http://www.solidkor.com/en/technology/414we.html [solidkor.com]
3. A typical SSD has spare blocks (just as HDD have spare blocks). So when a block is toast it is just marked as "bad" and a spare block is used instead.
4. Let us not forget ECC schemes that may extend the life of a block significantly.
All this adds up to a considerable lifespan for SSD's.
Let's for arguments sake say that that a SSD has 1 megabyte of spare blocks per 1 gigabyte storage. So if one were to read continuously to one 128 kilobyte block it would take:
(500k writes=assumed lifespan of a block)*(8 the number of spare blocks)=4M writes, and still not a single block lost in the sense that the filesystem/OS still sees 1 gigabyte of storage.
But read more:
http://www.storagesearch.com/ssdmyths-endurance.html [storagesearch.com]
--
Regards
Re:File swapping destroys SSDs (Score:2, Informative)
If you have only 1GB of space left, then wear leveling can only count on the blocks that don't contain data.
The wear levelling hardware does not contain drivers for your filesystem, or any filesystem, so it cannot know whether the block "contains data"
So your claim that it will only use 1GB and then wear it out is pure fallacy.
Re:Unbelievable (Score:3, Informative)
Re:Unbelievable (Score:5, Informative)
Hello, by 'reasonably modern hardware' do you mean those newfangled silent,fast SSD drive thingies? I think I read somewhere that Vista doesnt play nice with them.
Maybe you mean the latest CPUs comming out the Fabs, like the Atom and Via low power chips. I may have read a story about a hardware company (I think is was Asus) producing a low power device (the Eaaa PC?) that runs the latest Linux, but for the Windows version, they chose Windows XP over Vista for performance reasons.
Perhaps you mean new hardware designs like the Cell architecture and other SMP designs coming to a Blade Center near you. The NT base for Vista has a shitty scheduler, and appears to require 1 NIC per CPU for good performance, which is going to make 32-way CPUs rather expensive if you want to run Windows.
I was going to mock Windows for not being able to run on Cell based machines like the PS3, but it looks like somebody has managed it [engadget.com], pffft.
Re:Unbelievable (Score:2, Informative)
Ubuntu 7.10 vs Vista
Core 2 Duo E8400
2GB 800MHz DDR2
All SATA drives
7900 GS KO
In Ubuntu, I have compiz + beryl and every single sweet looking plugin you can imagine. My windows literally burst into flames when I close them, and everything has a neat little animation. I've got more apps that I can count on both hands running, and those are just the ones with the GUI representations. I have 4 workspaces, dual monitor each, and all 4 constantly have ZERO desktop showing. I regularly overlay a video, lowering opacity and using the ghost plugin to pass clicks through to windows underneath, while using eclipse and firefox with many many tabs. I never get a hiccup.
In Vista, I can run Aero, yes, but it hiccups occasionally. Not only that, but while opening programs, the entire system locks up quite regularly. Games in Vista vs. games in XP, XP smokes it. Playing media files (in WMP) puts significant stress on the processor (more than a couple percentages is significant, on my machine).
I'm not just saying windows sucks, because I can do about 70% of what I can do in Ubuntu in XP. Vista is absolutely terrible.
I don't have bleeding edge, but my hardware should be more than adequate in describing "modern hardware". Given that, relative to every other OS, Vista still runs like shit.
Why wait (Score:2, Informative)
Re:Unbelievable (Score:2, Informative)
Yeah... it's much better to have RAM in your system that isn't storing anything at all (or providing any benefit) and just consuming power. Caching is very cheap and if something else needs the RAM, the cached data can just be repurposed with little overhead (as long as it wasn't written to).
I don't get this from people... Whether I'm using Linux or Windows, I'd prefer for *all* of my RAM to be used *all* of the time, even if it is just caching stuff that it thinks I'll need but I won't. A cache should be able to drop untouched memory and give it quickly to any application that needs it. You should be pissed at an OS that *doesn't* use all your RAM all the time because it isn't doing its primary job... efficient use of resources.
Re:Unbelievable (Score:3, Informative)
Meh.
I run a Core 2 Duo E6550, 2GB, SATA, and a slower ATI card (Radeon HD 2400), and aero doesn't hiccup. In fact, even with 3-6 copies of Visual Studio open (2008), several PowerShell windows, and Outlook (bleh) on 2x 1680x1050 monitors, aero doesn't hiccup. I even open WoW as well once in a while just to see if it'll slow down, and it doesn't.
So much for anecdotal evidence.
Re:Why does NT need a pagefile? (Score:3, Informative)
This is certainly not true for VISTA and many other modern OS's. VISTA performa aggressive background prefetching of commonly used applications. It also builds a prefetch file in the background. This takes a big chunk of memory and a big of HD space. In cases like this where your OS is constantly doing lots of stuff with any of your extra memory in the background, sometimes it makes sense to swap out infrequently used memory to increase performance even if your entire working set fits into RAM. VISTA will run faster with a swap file even if you have 3GB of RAM and do little other than browsing the web. The extra memory from swapping out pages is used as file cache and for background operations that can be sped up by extra memory.
Oh, and why does NTFS need to be defragmented?
For the most part NTFS doesn't need to be defragmented. I've gone long periods of time without defragging on XP. However, if your disk starts getting over 80% full or you get more than 20% fragmentation on the HD, then your performance begins to suffer due to seeking. VISTA defragmentation is actually "smarter" internally than XP even though the interface has been dumbed down. It realizes that some fragments aren't a bad thing as long as the ratio of fragments to file size is reasonable. I believe it tries to make fragments be at least 64MB in size. Therefore a 2GB VOB file could be split into as many as 32 fragments before the file would be considered fragmented. This makes the defragmenter run much faster than XP (which tries to coalesce the whole 2GB file) with very little penalty in performance on the final defragged image. It also makes finding free space for coalescing fragments much easier.
It's a terrible shame that MS dumbed down the VISTA defragmenter interface [lifespy.com] and makes it hard for you to exclude drives (like SSD's which don't need defragging) from the automatic defrag schedule. It would have been much smarter for them to have the "dumb" interface with a single button to go to "advanced" or "power user" mode that had more options like drive exclusion.
Linux does. (Score:3, Informative)
In fact, Linux has at least one filesystem designed for flash.
So why doesn't Microsoft? Obviously, it was more important for them to meet once a week to debate the structure of "shut down" in the start menu.
I'm not making either of these things up, but I can't verify them right now. No time.
Re:Linux does. (Score:4, Informative)
Because you've put it in quotes, I suspect it's sarcasm, right?
Because I wasn't exaggerating. There actually is a filesystem, for Linux, deliberately written for flash: JFFS2 [wikipedia.org]. It actually will not work on a hard drive without an additional layer of emulation, and it wouldn't perform as well or be as reliable as on a real flash drive.