Follow Slashdot stories on Twitter

 


Forgot your password?
Close
typodupeerror
×
Data Storage Bug

New Samsung SSD 840 EVO Read Performance Fix Coming Later This Month 72

Posted by Soulskill from the slower-than-fastest-but-faster-than-slowest dept.
An anonymous reader writes: The Samsung SSD 840 EVO read performance bug has been on the table for over six months now. Initially Samsung acknowledged the issue fairly quickly and provided a fix only a month after the news hit the mainstream tech media, but reports of read performance degradation surfaced again a few weeks after the fix had been released, making it clear that the first fix didn't solve the issue for all users. Two months ago Samsung announced that a new fix is in the works and last week Samsung sent out the new firmware along with Magician 4.6 for testing, which will be available to the public later this month.
This discussion has been archived. No new comments can be posted.

New Samsung SSD 840 EVO Read Performance Fix Coming Later This Month More

New Samsung SSD 840 EVO Read Performance Fix Coming Later This Month

Comments Filter:

  • Too late; already sold my EVO's on eBay (Score:4, Insightful)

    by SplatMan_DK ( 1035528 ) on Wednesday April 15, 2015 @04:21AM (#49476717) Homepage Journal

    Six months is not an acceptable time to wait for a performance fix for an SSD drive. The very essence of an SSD is "speed".

    I offloaded the EVO's on eBay (being honest about the reason) and got myself a couple of Plextor Pro drives. Running in RAID0 they are a bit slower at random reads than the EVOs, but faster at sustained transfer rates.

    An SSD with slow/degraded performance is like a burger without the beef... something is missing...!

    • Re:Too late; already sold my EVO's on eBay (Score:5, Insightful)

      by davester666 ( 731373 ) on Wednesday April 15, 2015 @04:34AM (#49476751) Journal

      Good. Fast. Cheap.

      You bought Samsung, so you picked cheap.

      • Re:Too late; already sold my EVO's on eBay (Score:4, Insightful)

        by SplatMan_DK ( 1035528 ) on Wednesday April 15, 2015 @04:39AM (#49476765) Homepage Journal

        Not quite. The specs on the EVO 840's are actually very good ... if only they lived up to them!

        If this new fix actually works, the EVO's will be a very good buy. Pretty reliable too according to tests. I just couldn't wait (and had no idea if/when a fix was ever going to surface).

      • With SSDs it is more like: Fast / Cheap / Reliable (pick two).

        Samsung delivers the best reliability [techreport.com] (for sample size of one anyway) and is acceptably priced. It also appeared to be one of the fastest but this bug has proved it otherwise.

        So, if they can fix this performance bug they might just be all three.

      • Re: (Score:2)

        by Lumpy ( 12016 )

        No I skipped Samsung as they are pricey compared to the Crucial... Yet the Crucial outperforms the Samsung over and over again.

        The problem is Samsung screwed up and shipped drives that cant live up to the claims.

        • Actually they perform as advertised when they're new, or when the data is refreshed. So your claim is wrong.

          The degraded performance occurred when blocks of data remained unchanged for many months. And the fix was somewhat easy: Run the Samsung optimizer tool or alter the data.

          You can criticize them for the degradation-bug, sure. But to claim they shipped drives that did not live up to claimed performance is wrong.

          • Really? Just reset the drive and it goes fast again?

            Do you want to keep the Samsung SSD that's in this drawer?
            Yes, it's the fastest SSD I own.
            Why isn't it in one of your computers?
            Because once I use it, it goes slower than all the other SSDs I have.

            • The bug does not affect your data for at least a couple of months. And running the Samsung tool fixes it completely - as does moving the data to new blocks (though OS files are somewhat harder to move around if they're not in a virtual disk file).

              Not sure if you're trolling or just being contrary out of principle. But yeah, I'll take the EVO in your drawer if you insist it is unusable. Hell, I'll even make a donation to the EFF worth the drives second-hand value when I receive it. Let me know if you want my

    • I offloaded the EVO's on eBay (being honest about the reason) and got myself a couple of Plextor Pro drives. Running in RAID0 they are a bit slower at random reads than the EVOs, but faster at sustained transfer rates.

      Wow, so you replaced your existing drives with a performance "problem", with some new drives that you yourself admit are slower at random reads (BTW, guess what sort of reads are more important to the perception of speed). Do you really think you're a savvy consumer?

      This whole EVO performance problem is a crock of shit. People only noticed the problem running synthetic benchmarks. For normal usage, there'll be no perceptible performance issue with these drives. And yes, I have a couple of EVO drives, and ye

      • Hey, why the attitude man? Did it somehow offend you that I decided to replace the drives? Are you so emotionally attached to your own identity as an EVO customer that you must attack people who choose differently?

        I used the EVO 840 from virtually the same week they became available. Performance was great, but it slowly degraded with time passed. The decrease was very visible for large blocks of data which were read often but not refreshed or updated. Not just on synthetic benchmarks. My usage pattern fits

        • Re: (Score:3)

          by swb ( 14022 )

          It looks like a serious degredation of peformance from the perspective of the difference between what the drives should be capable of versus what the bug limits them to, but the GP poster sort has a point in that the drive's performance doesn't seem to be dropping even to the level of a USB3 flash disk let alone a mechanical SATA disk.

          Obviously nobody but the actual user of a specific setup would know whether or not it was burdensome, but I would be kind of surprised if it was generally noticable. I see a

          • I would be kind of surprised if it was generally noticable

            Amazon reviews are quite positive.

            • amazon reviews?

              other than the occasional smart guy there, amazon is a joke for reviews. I look at the q/a section and someone asks a valid question, some idiot replies "sorry, I don't know the answer to that, but I love this product so much!".

              sigh!!!

              much worse than AOL ever was. and that says something.

              what is it about amazon that draws the lowest iq's to 'respond' to questions saying 'I don't know'? is it some dialog box that pops up saying 'please answer this'? why spend time answering a question (tha

              • is it some dialog box that pops up saying 'please answer this'?

                I've received emails before from amazon saying "You bought this product, can you answer this question?"

                Taken with a grain of salt, Amazon reviews can sometimes show useful information:
                -Two similar products on Amazon, if one has a high rate of failure, that usually becomes apparent comparing average ratings, and distribution of ratings
                -Comments sometimes provide indication as far as common failure modes "This drive started slowing down after months", or other common problems, that you can then use for furthe

          • If this represents a real business workload, I would be kind of curious to know what kind of a workload you'd have to present to a stripe set of SSDs to see the effects of this performance bug. The linked article shows some kind of performance graph hitting a low of 100 MB/sec sustained read. A raid 0 stripe would be close to 200 MB/sec sustained read at worst. Maybe you'd notice it, but it seems like a pretty unusual workload that would expose this.

            I recently hit 600 MB/s disk throughput on our current test database (the interface limit). I was trying to do an incremental load of machine logs that have 16 datapoints per discrete item - and the machines process tens of millions of items per day. The datamodel wasnt as smart as it should have been, and the database had incorrect analytics, leading it to the conclusion that walking through the entire table (250GB data) row-by-row was a smart move...

            But that was on a small sample of the data, so when we l

            • And I should have added: We corrected the modeling mistakes. So now it's down to 50 MB/s at its peak, on test. Given the datamodel and the layout of the incoming data we don't expect it to get much higher even with bigger data loads.

          • Generally speaking most "Enterprise" VMs on slow SANs suck. Really - seriously - suck!

            And no, it does not represent real business workloads unless the company in question is either being cheap, or their infrastructure team has no clue what they're doing.

            Don't get me wrong; VMs on a SAN can be fine if the SAN is any good. The best SAN I have seen to date was at one of our clients where it was used for the primary servers running a complex ERP solution. It provided almost a full million IOPS when benchmarked

            • Re: (Score:2)

              by swb ( 14022 )

              We sell and install Compellents where I work and AFAIK the models with flash tiers are marketed at 300k IOPS max. 1M IOPS sounds like a benchmarking flaw.

              Maybe it would be possible to get 1M with multiple SSD enclosures uitlizing multilple SAS backend loops, but something tells me it would have to represent the sum of many workloads, not what would be possible for a single workload once you account for some latency associated with synchronizing dual controllers and the front end fabric connectivity limitat

              • Out of curiosity I contacted them and asked about the benchmark. Turns out they simulated 5 high-load servers running concurrently by started the bechmark tool on 5 different virtual servers on the SAN and running them at the same time, taking the average values for about 7 minutes of benchmarks. Each server reported around 192.000 IOS at a QD of 16, for a total of approx 960.000 IOPS.

                There was no other activity on the SAN an the servers were small so it's a pretty fair bet that 100% of the load was on the

                • Re: (Score:2)

                  by swb ( 14022 )

                  Wow, that is impressive. I doubt I'll ever see that much write intensive flash in one place. We added a flash tier to a seed install (dumb customer only bought a single 15k tier and wondered why performance sucked) and I can never get over how fucking outrageously expensive the flash tier costs. I think it was pushing $100k.

                  I'd wager that mid range flash like the Samsung 850 Pros are getting cheap enough that double parity, double hot spares and replacing disks regularly due to burnout is probably cheaper

        • But motherboard Raid is usually fakeRAID.

          • No, that hasn't been true for quite a while. At least not on workstation or gaming-grade motherboards. They all come with an Intel controller and Rapid Storage Technology which performs quite well for simple raid volumes like 0 and 1. Probably because there is much less work for these volumes than for the complex ones, like RAID 5, 6, 50 and 60.

        • Yes, I believe I am a reasonably savvy consumer when it comes to SSD drives. I am a Business Intelligence specialist and I am quite confident in my ability to understand and evaluate disk read performance. It is part of my job. When analyzing large amounts of data or operating virtual servers (booting or resuming), sustained data transfer is very important.

          In which case, you shouldn't have picked the 840 EVO to start with, but rather the 840 Pro (or 850 Pro now).

          If you want true performance, you have to pay for it. I have a pair of 840 EVOs myself because I don't need the extra speed from faster drives, and the price was right.

          • For the 840 series, the EVO is almost as fast as the Pro. The difference is about 3-5% depending on sources. The price of the pro is almost double that of the EVO. So no, I'd say the Pro only makes sense if you want the added reliability and extended warranty.

            The difference was bigger for the previous series, and I haven't checked the new 850 drives. But if you're in the 840 series then the Pro actually makes very little sense unless you get it at a bargain price.

      • This whole EVO performance problem is a crock of shit. People only noticed the problem running synthetic benchmarks. For normal usage, there'll be no perceptible performance issue with these drives. And yes, I have a couple of EVO drives, and yes, they are also hit with the performance issue. I just don't notice because it's such a non-issue outside of synthetic benchmarks.

        Bullshit I have 2 EVO's and I did notice. Some regions were very very slow (50-60 MB/s).
        When your VM copy suddenly takes 5-10 time

        • This whole EVO performance problem is a crock of shit. People only noticed the problem running synthetic benchmarks. For normal usage, there'll be no perceptible performance issue with these drives. And yes, I have a couple of EVO drives, and yes, they are also hit with the performance issue. I just don't notice because it's such a non-issue outside of synthetic benchmarks.

          Bullshit I have 2 EVO's and I did notice. Some regions were very very slow (50-60 MB/s).

          When your VM copy suddenly takes 5-10 times as long as expected, you do notice.

          OK, 99.9% of the use cases won't notice. Entire VM copies are probably quite rare, and are generally not an interactive type of operation (you don't sit there watching it copy, waiting for it to finish.)

          I've got a 840 EVO machine that's been powered off for a couple of months. I can't wait to power it up again and see if I do indeed perceive performance issues.

          • I think your 99,9% assertion is flat out wrong.

            Any file regardless of size which is unchanged for several months will be subject to this issue. This is not isolated to "pro" stuff like databases and virtual servers, but also includes locally stored video files, games, photos, etc.

  • Yay (Score:2)

    by JoelKatz ( 46478 )

    I'm looking forward to pulling all my mSATA EVOs out of their RAID controllers, inserting them one at a time into a spare PC with one mSATA slot, and upgrading their firmware. The last update (which also rewrites all data) took over two hours per drive, and it looks like this next one is going to take just as long. Anybody want to spend a really boring weekend with me?

    The EVO's are still the only 1TB mSATA drive, so not a lot of choices.

    • Is the update that rewrites the data going to be a problem on a LUKS encrypted volume? From what I saw it looks like it only supports NTFS? I also have an NTFS partition on the drive though. I guess I'm just concerned about it borking the LUKS partition.

      I hadn't heard about the original firmware update but was wondering why my read performance had gotten so much worse over time. Here I was blaming it on btrfs...

      • Their first fix came out as a windows program and later as a live cd. The Windows program used free ntfs space to shuffle data around.
        The live cd did not need a NTFS system and it did not eat my data. (ext4 + LVM2 + md-raid)
        I suspect this fix will get a corresponding live cd as well, so we'll just have to wait a little bit longer.

        • Thanks for the info. I was contemplating booting into Windows to run the fix, but it sounds like I might be better off using the live CD. I hadn't run the initial fix, so I'm debating if it's worth it to run it now and then run it again later when they release this fix, or just wait for the new fix.

    • Is there a specific reason you're usuing a RAID controller with mSATA sockets instead of plain old SATA drives? Is this a uSFF system or something?

  • Raises a point about tech reviews (Score:5, Interesting)

    by RogueyWon ( 735973 ) on Wednesday April 15, 2015 @05:40AM (#49476929) Journal
    I shifted to an SSD for my OS and core applications (plus a few disk-speed sensitive games - a fast-growing category) last year. I'd been planning to buy the 500gb 840 EVO, but, by some small miracle, Amazon had a special on the 840 Pro on the weekend I made my purchase, putting its price very close to the EVO, so I bought that instead. The 840 Pro is apparently not affected by this. Phew, bullet dodged.

    But it's interesting that the issue is picked up in so few reviews. Indeed, there's a veiled apology for this in an ExtremeTech article about the bug from October [extremetech.com]. Reviews are generally carried out on the basis of a short but intensive testing period and hence don't pick up serious issues that take a bit of time to show up.

    That's obviously been particularly important in this case, due to the specific nature of this bug. But when it comes to expensive bits of hardware like SSDs and high-end graphics cards, I'd be interested in reviews which came out a bit later but gave a better reflection of failure rates and longer-term issues. I've been stung before by buying a well-reviewed graphics card which turned out to have a horrible failure rate over time.

    • Re:Raises a point about tech reviews (Score:4, Insightful)

      by Anonymous Coward on Wednesday April 15, 2015 @07:04AM (#49477139)

      Unfortunately, most of the sites that do reviews survive thanks to their readers. Which means they can't afford to wait 6 months after product release to publish. By that time, most of their competition has taken their views.

    • This is a fantasy that will never happen in the US, but might in the EU: require computer hardware manufacturers and those that sell said merchandise to clearly and explicitly state the Revision Level of the actual piece being offered for sale, and what changed in the new revision. A lot of these folks treat that info like the family jewels, with no way of knowing what version you will end up with until it is actually in your hands. Why? So they don't have to recall / scrap / take a bath on the Rev 0 cra

      • as evidenced by them moving on the Rev 1.

        That's not evidence. It's like saying any new product release is an indication that the old one is faulty, it isn't. Nearly all revision updates are the result of either a cheaper way of producing something, or a subtle change in the system due to something like a part supplier not pulling their weight, or a cheaper part becoming available.

    • > But when it comes to expensive bits of hardware like SSDs and high-end graphics cards, I'd be interested in reviews which came out a bit later but gave a better reflection of failure rates and longer-term issues. I've been stung before by buying a well-reviewed graphics card which turned out to have a horrible failure rate over time.

      You may very well be waiting for the product to go EOL and be superseded by a new model. In technology that is the ongoing story....

  • [warning: armchair comment] Looks like a leakage problem, degrading the cells, so reads must be retried sub-optimally. The fix would be for the drive to re-write/re-allocate old cells, which could become a maintenance task that does not noticeably degrade live performance nor lifespan. However, this does limit the drive's use as a portable or offline drive, where this maintenance cannot be performed routinely.

  • Just buy Crucial or Intel (Score:1)

    by Anonymous Coward

    And have it just work.

    • Re: (Score:2)

      by jez9999 ( 618189 )

      Yup, I almost always go with Intel for my SSDs now. Not sure what I'm going to go to replace my 1TB Seagate backup drive, though; I don't know whether there are any SSDs that big that are of decent quality and affordable.

      • But Intel does something awful when you reach the wear leveling limit. After a power cycle it closes your drive !
  • Forget the 840 EVO users, what about the plain old vanilla 840 TLC users? Jury's still out on the 850 EVO I suppose. Now their Pro line is great and is all I will use.

Slashdot Top Deals

"May your future be limited only by your dreams." -- Christa McAuliffe

Close