If this is your first visit, be sure to check out the FAQ by clicking the link above. You may have to register before you can post: click the register link above to proceed. To start viewing messages, select the forum that you want to visit from the selection below. |
|
|
|
Thread Tools | Rate Thread | Display Modes |
#1
|
|||
|
|||
Can HDTune go slower where there is data, if something's set wrong?
I have a brand-new "1T" (of course thus only 931.5G) drive. (It is now
showing 158 hours on time.) When I run HDTune, I get a huge gouge from about 11-23%, and some in the first 3%, about which I'm somewhat concerned: surely a brand-new drive can't be this bad? (The SMART data actually showed 0 for hours the first time I looked, and I have stored looks that then showed 4 hours and now 15x; it came in a sealed bag; and the label on the drive itself says November 2017. I _imagine_ all these things can be faked, though I imagine the SMART one requires special knowledge/equipment; however, I've no real reason to suspect the trader I bought it from - and besides, 1T 2.5" haven't been around _that_ long, have they?) However, it has just occurred to me that the gouges correspond to where I actually have data: I have it partitioned into C:, which is 71.6G of 99.9G free, and D:, which is 723G of 831G free. [Is there a setting in explorer to show used rather than free?] Since the gouges correspond rather well to where my data is, I wonder if that's their reason. Also, the remainder of the HDTune graph is fairly flat, whereas the ones I've seen people post usually tail down towards the end, so I'm wondering if I've got HDTune - or some other setting in the computer (BIOS?) - set at sub-optimum, such that it's (a) limiting the maximum (the flat line is at about 170 MB/sec - is that low? - and the "gouge" is at about 90), and (b) making HDTune drop where there's actually something on the disc. I'm quite happy with the performance of the computer, and all other tests including "Very long test (~4 min)" from http://www.softwareok.com/?Microsoft/IsMyHdOK say it's OK. (I haven't run HDTune's non-quick error scan yet, as that clearly will take hours; I'll set it going tonight if I remember.) It's an HGST (I think that's Hitachi/Toshiba combined?) HTS541010B7E610, according to HDTune [I deliberately went for a "5400" rather than a "7200"], and (again according to HDTune) apparently supports UDMA Mode 6 (Ultra ATA/133), with Active being UDMA Mode 5 (Ultra ATA/100). The PC is a Toshiba Portégé R700-1F5. -- J. P. Gilliver. UMRA: 1960/1985 MB++G()AL-IS-Ch++(p)Ar@T+H+Sh0!:`)DNAf "I'm a self-made man, thereby demonstrating once again the perils of unskilled labor..." - Harlan Ellison |
Ads |
#2
|
|||
|
|||
Can HDTune go slower where there is data, if something's setwrong?
J. P. Gilliver (John) wrote:
I have a brand-new "1T" (of course thus only 931.5G) drive. (It is now showing 158 hours on time.) When I run HDTune, I get a huge gouge from about 11-23%, and some in the first 3%, about which I'm somewhat concerned: surely a brand-new drive can't be this bad? (The SMART data actually showed 0 for hours the first time I looked, and I have stored looks that then showed 4 hours and now 15x; it came in a sealed bag; and the label on the drive itself says November 2017. I _imagine_ all these things can be faked, though I imagine the SMART one requires special knowledge/equipment; however, I've no real reason to suspect the trader I bought it from - and besides, 1T 2.5" haven't been around _that_ long, have they?) However, it has just occurred to me that the gouges correspond to where I actually have data: I have it partitioned into C:, which is 71.6G of 99.9G free, and D:, which is 723G of 831G free. [Is there a setting in explorer to show used rather than free?] Since the gouges correspond rather well to where my data is, I wonder if that's their reason. Also, the remainder of the HDTune graph is fairly flat, whereas the ones I've seen people post usually tail down towards the end, so I'm wondering if I've got HDTune - or some other setting in the computer (BIOS?) - set at sub-optimum, such that it's (a) limiting the maximum (the flat line is at about 170 MB/sec - is that low? - and the "gouge" is at about 90), and (b) making HDTune drop where there's actually something on the disc. I'm quite happy with the performance of the computer, and all other tests including "Very long test (~4 min)" from http://www.softwareok.com/?Microsoft/IsMyHdOK say it's OK. (I haven't run HDTune's non-quick error scan yet, as that clearly will take hours; I'll set it going tonight if I remember.) It's an HGST (I think that's Hitachi/Toshiba combined?) HTS541010B7E610, according to HDTune [I deliberately went for a "5400" rather than a "7200"], and (again according to HDTune) apparently supports UDMA Mode 6 (Ultra ATA/133), with Active being UDMA Mode 5 (Ultra ATA/100). The PC is a Toshiba Portégé R700-1F5. I think that in your situation I'd run some MS utility before anything else. Firstly a simple check file structure. Secondly a pre-boot full scan. I'd decide what next from those results. If ok, look further in HDTune. If not, then back to the vendor. Have you got a good backup? Ed |
#3
|
|||
|
|||
Can HDTune go slower where there is data, if something's setwrong?
J. P. Gilliver (John) wrote:
I have a brand-new "1T" (of course thus only 931.5G) drive. (It is now showing 158 hours on time.) When I run HDTune, I get a huge gouge from about 11-23%, and some in the first 3%, about which I'm somewhat concerned: surely a brand-new drive can't be this bad? (The SMART data actually showed 0 for hours the first time I looked, and I have stored looks that then showed 4 hours and now 15x; it came in a sealed bag; and the label on the drive itself says November 2017. I _imagine_ all these things can be faked, though I imagine the SMART one requires special knowledge/equipment; however, I've no real reason to suspect the trader I bought it from - and besides, 1T 2.5" haven't been around _that_ long, have they?) However, it has just occurred to me that the gouges correspond to where I actually have data: I have it partitioned into C:, which is 71.6G of 99.9G free, and D:, which is 723G of 831G free. [Is there a setting in explorer to show used rather than free?] Since the gouges correspond rather well to where my data is, I wonder if that's their reason. Also, the remainder of the HDTune graph is fairly flat, whereas the ones I've seen people post usually tail down towards the end, so I'm wondering if I've got HDTune - or some other setting in the computer (BIOS?) - set at sub-optimum, such that it's (a) limiting the maximum (the flat line is at about 170 MB/sec - is that low? - and the "gouge" is at about 90), and (b) making HDTune drop where there's actually something on the disc. I'm quite happy with the performance of the computer, and all other tests including "Very long test (~4 min)" from http://www.softwareok.com/?Microsoft/IsMyHdOK say it's OK. (I haven't run HDTune's non-quick error scan yet, as that clearly will take hours; I'll set it going tonight if I remember.) It's an HGST (I think that's Hitachi/Toshiba combined?) HTS541010B7E610, according to HDTune [I deliberately went for a "5400" rather than a "7200"], and (again according to HDTune) apparently supports UDMA Mode 6 (Ultra ATA/133), with Active being UDMA Mode 5 (Ultra ATA/100). The PC is a Toshiba Portégé R700-1F5. It's possible for a drive to have a "wear" pattern. The place where say, an OS lives, can cause a lot of reallocations just in that area. When HDTune runs over it, the reallocations take longer to read, and cause the visible "gouge" symptoms. I had to replace a drive here, just because the OS portion (72GB approx) had become too slow to use. Reallocations still read out as "0". Benchmarking is a more sensitive test than looking for Reallocation raw data of 0. You can have a gouge in a drive, the drive can "feel" slow, and Reallocations still reads 0 and the overall drive health will use a superlative. And yes, of course drives leave the factory with defects already patched. No drive leaves the factory without reallocations already on there. There is an "acceptance" value for defects. If the drive is below that limit, "ship it". For example, there might already be 100,000 reallocations as it slips out the door. When Reallocations data becomes non-zero, you're looking at some "slice" of ill health, rather than receiving a totally unbiased report. In the example here, we only find out about the N reallocations as the drive is approaching end of life. The value of N is high enough, the user runs out to the store and buys another. The drive is acceptable to the factory, anywhere in the "0" area. The drives I was watching here, the percent life said the N would span 0..5500 or so as a raw value. Above 5500, one would guess that visible CRC errors and retired NTFS clusters would be the result. Once the Service Area (Track -1) is unreadable, the drive should stop being detected. | | |------- 0 -------| "N" | "~dead..." |______________________________________ The first hard drives, the full height drives which were 3.5" high, they had "factory" and "grown" errors. I used to get drives, and dump the "factory" list and print it on a piece of paper (pasted inside the computer chassis). The fact there was a "factory" list tells you the "factory" knew about them. And during the testing/verifying phase at the factory, sectors were already being reallocated. On those drives, you could reset the "grown" defects and let the reallocation algo re-detect them. Which worked pretty well (I tested it, because that's the kind of guy I am). There weren't a lot of grown that weren't re-detected that way, so the ones it had detected, really were bad. Modern IDE drives do not support that interface and once a sector is reallocated on an IDE/SATA drive, it's like that forever (or until the factory refurbishes the drive or something). ******* HGST appeared out of no-where one day. There was some announcement of the consumer part of the IBM disk business, being part of the new venture. With it, went some IBM scientists. The IBM staff were responsible for the HGST web pages, the ones that explained how modern drives worked, and what wizmo material science was being used. I don't know what hand Hitachi had in all of it. Japanese companies don't typically have those sorts of web pages, explaining how they do stuff. It was the IBM people doing that. HGST was later bought by Western Digital. There was an article on Anandtech in the last couple years, where it listed who bought who, and it's possible the only three brands left are WDC, Seagate, and Toshiba. HGST drive may still keep their branding, but WDC is the corporate master now. I don't know whether the IBM staff go back to IBM or what happens to them. This article actually sums up what happened, in a paragraph or two. What these articles don't typically cover, is what happened to the people. https://en.wikipedia.org/wiki/HGST Paul |
#4
|
|||
|
|||
Can HDTune go slower where there is data, if something's set wrong?
In message , Paul
writes: J. P. Gilliver (John) wrote: I have a brand-new "1T" (of course thus only 931.5G) drive. (It is now showing 158 hours on time.) When I run HDTune, I get a huge gouge from about 11-23%, and some in the first 3%, about which I'm somewhat concerned: surely a brand-new drive can't be this bad? (The SMART data actually showed 0 for hours [] However, it has just occurred to me that the gouges correspond to where I actually have data: I have it partitioned into C:, which is 71.6G of 99.9G free, and D:, which is 723G of 831G free. [Is there a setting in explorer to show used rather than free?] [] It's possible for a drive to have a "wear" pattern. I'm not sure I understand what you mean by a wear pattern. (Would you expect one after only 173 hours from new anyway?) The place where say, an OS lives, can cause a lot of reallocations just in that area. When HDTune runs over it, the reallocations take longer to read, and cause the visible "gouge" symptoms. I had to replace a drive here, just because the OS portion (72GB approx) had become too slow to use. Reallocations still read out as "0". This drive "feels" fine. Benchmarking is a more sensitive test than looking for Reallocation raw data of 0. You can have a gouge in a drive, the drive can "feel" slow, and Reallocations still reads 0 and the overall drive health will use a superlative. And yes, of course drives leave the factory with defects already patched. No drive leaves the factory without reallocations already on there. There is an "acceptance" value for defects. If the drive is below that limit, "ship it". For example, there might already be 100,000 reallocations as it slips out the door. Yes, I realise that, and that they conceal the first few to prevent selective returns; and what you've said often about HDTune being a reasonable way to detect what the SMART won't tell you, by consistent dips on successive runs. But (a) there seem to be a lot more - i. e. about 12% or more - than I'd have expected on a new disc, and (b) they _do_ seem to correspond rather suspiciously to where I actually have data. [And (c) the HDTune line doesn't tail off at all towards 100%.] (If you email me, I'll send you the HDTune screenshots.) [] The first hard drives, the full height drives which were 3.5" high, they had "factory" and "grown" errors. I used to get drives, and dump the "factory" list and print it on a piece of paper (pasted inside the computer chassis). The fact there I remember seeing those with a provided list pasted to the drive. [] HGST appeared out of no-where one day. There was some announcement [] where it listed who bought who, and it's possible the only three brands left are WDC, Seagate, and Toshiba. HGST drive may still keep their branding, but WDC is the corporate master now. I don't know whether the IBM staff go back to IBM or what happens to them. This one says Nov 2017 on the label. This article actually sums up what happened, in a paragraph or two. What these articles don't typically cover, is what happened to the people. https://en.wikipedia.org/wiki/HGST Paul Thanks. So the H was originally Hitachi, Toshiba is nowhere in it, and in practice it's part of WD now. -- J. P. Gilliver. UMRA: 1960/1985 MB++G()AL-IS-Ch++(p)Ar@T+H+Sh0!:`)DNAf There's only so much you can do... with gravel. - Charlie Dimmock, RT 2016/7/9-15 |
#5
|
|||
|
|||
Can HDTune go slower where there is data, if something's set wrong?
In message , "J. P. Gilliver (John)"
writes: In message , Paul writes: [] This drive "feels" fine. Benchmarking is a more sensitive test than looking for Reallocation raw data of 0. You can have a gouge in a drive, the drive can "feel" slow, and Reallocations still reads 0 and the overall drive health will use a superlative. And yes, of course drives leave the factory with defects already patched. No drive leaves the factory without reallocations already on there. There is an "acceptance" value for defects. If the drive is below that limit, "ship it". For example, there might already be 100,000 reallocations as it slips out the door. Yes, I realise that, and that they conceal the first few to prevent selective returns; and what you've said often about HDTune being a reasonable way to detect what the SMART won't tell you, by consistent dips on successive runs. But (a) there seem to be a lot more - i. e. about 12% or more - than I'd have expected on a new disc, and (b) they _do_ seem to correspond rather suspiciously to where I actually have data. [And (c) the HDTune line doesn't tail off at all towards 100%.] (If you email me, I'll send you the HDTune screenshots.) [] I've now done the non-"Quick Scan" scan (about 155 minutes), and it didn't find and damaged sectors. I've also played with the settings for Benchmark. There are two - one slider for speed versus accuracy, with five levels, and one for block size. What do these mean? I can see that the speed one affects how fast the test runs, but it doesn't seem to affect whether the gouges are there or not. But the block size does! All the tests I'd done so far - that show gouges (wide dips) near the beginning and at around 12-23% (corresponding rather suspiciously to where there is data on the disc) were with a block size of 64 KB, which was how HDTune came up. They gave - speed 2: flat at about 160 MB/sec, gouge roughly 90 (3 runs). Then I did a speed 1 (slowest) run with block size of 512 B (smallest): this gives a flat trace - no gouges! - but at only about 2.75 MB/sec. Then I did a speed 1 but back at 64 KB block size: gouges are back - flat is about 140 MB/s, gouge about 90. Then speed 2, 512 B block: flat, no gouges, about 2.8 MB/s. Then speed 5, 64 KB: gouges, flat ~140M/s, gouge 90ish. Finally speed 2, block size 8 MB: flat way up at about 230 MB/sec, gouge bottoming still at about 90. In short, a small block size gives a flat line with no gouges, but at a very low level; a larger one gives a graph with the gouges (which seem to correspond to where there is data), with varying flat level. So what does the block size parameter mean/do? -- J. P. Gilliver. UMRA: 1960/1985 MB++G()AL-IS-Ch++(p)Ar@T+H+Sh0!:`)DNAf Advertising is legalized lying. - H.G. Wells |
#6
|
|||
|
|||
Can HDTune go slower where there is data, if something's setwrong?
J. P. Gilliver (John) wrote:
In message , "J. P. Gilliver (John)" writes: In message , Paul writes: [] This drive "feels" fine. Benchmarking is a more sensitive test than looking for Reallocation raw data of 0. You can have a gouge in a drive, the drive can "feel" slow, and Reallocations still reads 0 and the overall drive health will use a superlative. And yes, of course drives leave the factory with defects already patched. No drive leaves the factory without reallocations already on there. There is an "acceptance" value for defects. If the drive is below that limit, "ship it". For example, there might already be 100,000 reallocations as it slips out the door. Yes, I realise that, and that they conceal the first few to prevent selective returns; and what you've said often about HDTune being a reasonable way to detect what the SMART won't tell you, by consistent dips on successive runs. But (a) there seem to be a lot more - i. e. about 12% or more - than I'd have expected on a new disc, and (b) they _do_ seem to correspond rather suspiciously to where I actually have data. [And (c) the HDTune line doesn't tail off at all towards 100%.] (If you email me, I'll send you the HDTune screenshots.) [] I've now done the non-"Quick Scan" scan (about 155 minutes), and it didn't find and damaged sectors. I've also played with the settings for Benchmark. There are two - one slider for speed versus accuracy, with five levels, and one for block size. What do these mean? I can see that the speed one affects how fast the test runs, but it doesn't seem to affect whether the gouges are there or not. But the block size does! All the tests I'd done so far - that show gouges (wide dips) near the beginning and at around 12-23% (corresponding rather suspiciously to where there is data on the disc) were with a block size of 64 KB, which was how HDTune came up. They gave - speed 2: flat at about 160 MB/sec, gouge roughly 90 (3 runs). Then I did a speed 1 (slowest) run with block size of 512 B (smallest): this gives a flat trace - no gouges! - but at only about 2.75 MB/sec. Then I did a speed 1 but back at 64 KB block size: gouges are back - flat is about 140 MB/s, gouge about 90. Then speed 2, 512 B block: flat, no gouges, about 2.8 MB/s. Then speed 5, 64 KB: gouges, flat ~140M/s, gouge 90ish. Finally speed 2, block size 8 MB: flat way up at about 230 MB/sec, gouge bottoming still at about 90. In short, a small block size gives a flat line with no gouges, but at a very low level; a larger one gives a graph with the gouges (which seem to correspond to where there is data), with varying flat level. So what does the block size parameter mean/do? I would think a too-small block size choice, would result in not getting any "interesting" information from the hard drive. To get "gouges" to show up, you need to be asking for at least a handful of sectors, so the one or more bad sectors show abnormal response time (or equivalent bandwidth). If you wanted to know the details (stride and block size actually used), you could use Sysinternals Process Monitor and log what it is doing. It doesn't read every sector. It strides along the disk and takes samples of "blocksize" size. But you can verify that theory with ProcMon. I don't want to spoil your fun :-) Paul |
#7
|
|||
|
|||
Can HDTune go slower where there is data, if something's set wrong?
In message , Paul
writes: J. P. Gilliver (John) wrote: [] So what does the block size parameter mean/do? I would think a too-small block size choice, would result in not getting any "interesting" information from the hard drive. To get "gouges" to show up, you need to be asking for at least a handful of sectors, so the one or more bad sectors show abnormal response time (or equivalent bandwidth). If you wanted to know the details (stride and block size actually used), you could use Sysinternals Process Monitor and log what it is doing. It doesn't read every sector. It strides along the disk and takes samples of "blocksize" size. But you can verify that theory with ProcMon. I don't want to spoil your fun :-) Paul It isn't fun any more! I've just experimented: the gouge shows up with a block size of 32 KB, but not with one of 16 KB. As it "strides along the disc" (presumably at a speed determined by the speed slider setting), when it takes samples, does it find empty gaps of that size and use those? If it is looking at a part of the disc where there is data rather than empty space, what does it do - move aside the data in order to do its tests? That might take longer for larger block sizes maybe? I've put some screenshots in http://255soft.uk/temp/HDTune.zip (filenames giving speed and blocksize settings). -- J. P. Gilliver. UMRA: 1960/1985 MB++G()AL-IS-Ch++(p)Ar@T+H+Sh0!:`)DNAf Someone once said that scientists and prostitutes get paid for doing what they enjoy. - Prof Stepehen Hawking in RT 2013/12/7-13 |
#8
|
|||
|
|||
Can HDTune go slower where there is data, if something's setwrong?
J. P. Gilliver (John) wrote:
In message , Paul writes: J. P. Gilliver (John) wrote: [] So what does the block size parameter mean/do? I would think a too-small block size choice, would result in not getting any "interesting" information from the hard drive. To get "gouges" to show up, you need to be asking for at least a handful of sectors, so the one or more bad sectors show abnormal response time (or equivalent bandwidth). If you wanted to know the details (stride and block size actually used), you could use Sysinternals Process Monitor and log what it is doing. It doesn't read every sector. It strides along the disk and takes samples of "blocksize" size. But you can verify that theory with ProcMon. I don't want to spoil your fun :-) Paul It isn't fun any more! I've just experimented: the gouge shows up with a block size of 32 KB, but not with one of 16 KB. As it "strides along the disc" (presumably at a speed determined by the speed slider setting), when it takes samples, does it find empty gaps of that size and use those? If it is looking at a part of the disc where there is data rather than empty space, what does it do - move aside the data in order to do its tests? That might take longer for larger block sizes maybe? I've put some screenshots in http://255soft.uk/temp/HDTune.zip (filenames giving speed and blocksize settings). I think I'd pull the drive from the laptop and test it on another computer. It's a SATA III drive, with a 128MB cache, 1040Mbit/sec media rate. If we take the media rate and divide by 10, the disk should do about 104MB/sec reads. If we take the media rate and divide by 8 (unlikely), the rate would be about 125MB/sec. Yet, some of your benches show 225MB/sec (like the drive is on a SATA II port and all the data is coming from a controller cache). Media rate is a shabby way of stating performance, as it requires guys like me to estimate the coding loss. Sustained read rate is a better measure, as it requires no "secret knowledge" of the thing. HDTune should be doing a read-only test that makes no state changes to the disk whatsoever. It's doing raw disk reads while ignoring the file system (it's like it is reading /dev/sda). You could do an error scan while screen recording the bandwidth recorded as a function of position, if you wanted to continue using your existing hardware setup for this. Those two numbers in the corner, can be used to chart disk performance in a pinch. With the assurance the test is reading the entire 1TB of data from the disk. https://s10.postimg.org/x1zpk616x/hdtune_error_scan.gif I've recorded such an HDTune run with a screen recorder, then applied OCR to the screen dump, but it was far from "fun". The OCR of course, mixes up 0 and O with the usual comedic results. The log generated required significant manual correction. ******* https://www.hgst.com/sites/default/f..._datasheet.pdf Your transfer curves are flat. That's unrealistic. HDDs have a declining curve. Your access time is 2.8ms. That's unrealistic. One of your scans shows 225MB/sec sustained, when the data sheet allows at most 125MB/sec. That's unrealistic. Maybe your machine has a Robson cache (20GB Flash that goes with certain chipsets). Maybe you've loaded a third-party caching software that doesn't listen to "Do-Not-Cache" requests. But as it stands now, your scans lack a certain amount of realism. The drive isn't as fast as an SSD. But it's too fast for an HDD in certain ways. Paul |
#9
|
|||
|
|||
Can HDTune go slower where there is data, if something's set wrong?
In message , Paul
writes: J. P. Gilliver (John) wrote: [] I've put some screenshots in http://255soft.uk/temp/HDTune.zip (filenames giving speed and blocksize settings). I think I'd pull the drive from the laptop and test it on another computer. It's a SATA III drive, with a 128MB cache, 1040Mbit/sec media rate. If we take Unfortunately, I suspect this machine is SATA II (Toshiba Portégé R700-1F5); I'm pretty sure any others I have will not be above SATA II, and I'd almost certainly have to test it via USB2 anyway. the media rate and divide by 10, the disk should do about 104MB/sec reads. If we take the media rate and divide by 8 (unlikely), the rate would be about 125MB/sec. Yet, some of your benches show 225MB/sec (like the drive is on a SATA II port and all the data is coming from a controller cache). Mostly 140 or 170 - only the one with 8 MB "block size" had the 225. But yes, suspiciously fast. Media rate is a shabby way of stating performance, as it requires guys like me to estimate the coding loss. Sustained read rate is a better measure, as it requires no "secret knowledge" of the thing. HDTune should be doing a read-only test that makes no state changes to the disk whatsoever. It's doing raw disk reads while ignoring the file system That I understand, I think (it's like it is reading /dev/sda). (that I don't, but I assume it's UNIX/Linuxspeak). I see (I think) - it's doing reads, but ignoring the data that is actually read. You could do an error scan while screen recording the bandwidth recorded as a function of position, if you wanted to continue using your existing hardware setup for this. Those two numbers in the corner, can be used to chart disk performance in a pinch. With the assurance the test is reading the entire 1TB of data from the disk. https://s10.postimg.org/x1zpk616x/hdtune_error_scan.gif I've recorded such an HDTune run with a screen recorder, then applied OCR to the screen dump, but it was far from "fun". The OCR of course, mixes up 0 and O with the usual comedic results. The log generated required significant manual correction. Sounds like hard work (-: ******* https://www.hgst.com/sites/default/f..._datasheet.pdf Your transfer curves are flat. That's unrealistic. HDDs have a declining curve. Yes, I thought so. Your access time is 2.8ms. That's unrealistic. One of your scans shows 225MB/sec sustained, when the data sheet allows at most 125MB/sec. That's unrealistic. Maybe your machine has a Robson cache (20GB Flash that goes with certain chipsets). Maybe you've loaded a third-party caching software that doesn't listen to "Do-Not-Cache" requests. The cache seems unlikely: I got the machine second-hand over a year ago. I can't see a manufacturing date on it. It only has 3G of RAM, so a 20G cache seems unlikely! But as it stands now, your scans lack a certain amount of realism. The drive isn't as fast as an SSD. But it's too fast for an HDD in certain ways. Paul Any thoughts on the suspicious similarity between where the gouges are and where I have data? (I think it's likely to be at the beginning of the partitions - I've done a few defrags, and only been using the new drive for a couple of weeks anyway.) And why they only show up over a certain "block size" (not there at 16 KB or less, there at 32 KB or more)? For the moment, all seems to be working fine - I've not been aware of sudden slowdowns. And it is a brand new drive, which I'd thought _unlikely_ to have about 10-15% bad sectors. Which I bought for capacity only; the cache is nice to have, but I didn't go looking for it, and I specifically went for a "5400" rather than a 7200, feeling that would probably run cooler and be more reliable. -- J. P. Gilliver. UMRA: 1960/1985 MB++G()AL-IS-Ch++(p)Ar@T+H+Sh0!:`)DNAf The average US shareholding lasts 22 seconds. Nobody knows who invented the fire hydrant: the patent records were destroyed in a fire. Sandcastles kill more people than sharks. Your brain uses less power than the light in your fridge. The Statue of Liberty wears size 879 shoes. - John Lloyd, QI supremo (RT, 2014/9/27-10/3) |
#10
|
|||
|
|||
Can HDTune go slower where there is data, if something's setwrong?
J. P. Gilliver (John) wrote:
Any thoughts on the suspicious similarity between where the gouges are and where I have data? (I think it's likely to be at the beginning of the partitions - I've done a few defrags, and only been using the new drive for a couple of weeks anyway.) And why they only show up over a certain "block size" (not there at 16 KB or less, there at 32 KB or more)? For the moment, all seems to be working fine - I've not been aware of sudden slowdowns. And it is a brand new drive, which I'd thought _unlikely_ to have about 10-15% bad sectors. Which I bought for capacity only; the cache is nice to have, but I didn't go looking for it, and I specifically went for a "5400" rather than a 7200, feeling that would probably run cooler and be more reliable. It looks like some cache in your setup, is defeating the "do-not-cache" features that HDTune will use. The authors of these benchmarks go to a great deal of trouble to defeat caches, and get raw media characteristics in a scan. These schemes are easily defeated by new hardware, requiring the author of the software to "re-jig" how they do stuff, to get the declining curve back. If you simply refuse to remove the drive from the laptop, that's OK. Get yourself the *trial* version of the current HDTune, which should be able to give a more realistic result. As the current HDTune is maintained by its author. Uninstall 2.55 and give this a try. http://www.hdtune.com/download.html "HD Tune Pro 5.70 4 August 2017 hdtunepro_570_trial.exe 2187 KB Licensing information: after installation you can try out the program for 15 days. If you want to you use the program after this period you have to purchase a serial number. " I think your "gouge" pattern is real. But when the other parameters of the scan look more realistic, who knows, maybe the pattern will be as realistic as possible too. The thing is, no regular computer can cache the entire 1TB drive. Only an Epyc with $40000 of DRAM in it, could come close (there are 128GB RDIMMs now). The benchmark curve reads a limited amount of data, making it easier to cache. The error scan reads the entire disk, but, it doesn't have a graph. Let's hope the trial version sees the cache problem and does something about it. Paul |
#11
|
|||
|
|||
Can HDTune go slower where there is data, if something's set wrong?
In message , Paul
writes: J. P. Gilliver (John) wrote: Any thoughts on the suspicious similarity between where the gouges are and where I have data? (I think it's likely to be at the [] It looks like some cache in your setup, is defeating the "do-not-cache" features that HDTune will use. Seems plausible. The authors of these benchmarks go to a great deal of trouble to defeat caches, and get raw media characteristics in a scan. These schemes are easily defeated by new hardware, requiring the author of the software to "re-jig" how they do stuff, to get the declining curve back. If you simply refuse to remove the drive from the Well, it's not so much _refuse_ (though I agree it'd be inconvenient), just that the only way I'd have to test it would then be via a USB2 interface. laptop, that's OK. Get yourself the *trial* version of the current HDTune, which should be able to give a more realistic result. As the current HDTune is maintained by its author. Uninstall 2.55 and give this a try. http://www.hdtune.com/download.html "HD Tune Pro 5.70 4 August 2017 hdtunepro_570_trial.exe 2187 KB [] OK, got it. Though not too encouraged by the other utility that appeared in the same program group as "HD Tune Pro" - "HD Tune Pro Drive Status" - which tells me Capacity: 1000 gB 1,000,202,273,280 bytes 1,953,520,065 sectors Available Space: 76.0 gB (8%) , which sounds like a wraparound error ... but still, carrying on: I think your "gouge" pattern is real. But when the other parameters of the scan look more realistic, who knows, maybe the pattern will be as realistic as possible too. The thing is, no regular computer can cache the entire 1TB drive. Only an Epyc with $40000 of DRAM in it, could come close (there are 128GB RDIMMs Certainly not in this little old laptop! now). The benchmark curve reads a limited amount of data, making it easier to cache. The error scan reads the entire disk, but, it doesn't have a graph. Let's hope the trial version sees the cache problem and does something about it. Paul Benchmark curve, using default settings (speed 2, 64K blocks): http://255soft.uk/temp/2018-3-9%201-...%2064%20KB.png No real difference - the flat line is still higher than you'd expect, and the gouge is still there. Doing the error scan now - it's done 7200 MB in nearly 5 minutes, so I'll let it run and upload when finished. Not showing any Damaged 381MB squares yet (and 8001 MB is beyond the initial glitches on the Benchmark curve). Do you _really_ think a new drive would have 10-15% bad sectors? (Even have that many "swap" ones, for that matter?) Now got up to "12 gB", no red squares yet. -- J. P. Gilliver. UMRA: 1960/1985 MB++G()AL-IS-Ch++(p)Ar@T+H+Sh0!:`)DNAf Beatrix Potter was a bunny boiler. - Patricia Routledge, on "Today" 2016-1-26 |
#12
|
|||
|
|||
Can HDTune go slower where there is data, if something's setwrong?
J. P. Gilliver (John) wrote:
Benchmark curve, using default settings (speed 2, 64K blocks): http://255soft.uk/temp/2018-3-9%201-...%2064%20KB.png No real difference - the flat line is still higher than you'd expect, and the gouge is still there. Doing the error scan now - it's done 7200 MB in nearly 5 minutes, so I'll let it run and upload when finished. Not showing any Damaged 381MB squares yet (and 8001 MB is beyond the initial glitches on the Benchmark curve). Do you _really_ think a new drive would have 10-15% bad sectors? (Even have that many "swap" ones, for that matter?) Now got up to "12 gB", no red squares yet. I take it you've reviewed the "Health" tab in HDTune, and verified the drive is new that way ? How many power on hours does it have ? Is the stated number of power on hours consistent with how long you've been using it ? Did you check Programs and Features for any third party cache programs ? Intel used to make a software cache program, and Samsung may have tried their hand at it too. I don't keep tabs on stuff like that, so can't suggest any programs names. I've had *one* drive here, enter into the "gouge" pattern, right where the OS was located. And the drive had become so slow, it was doing maybe 20MB/sec or so and I could *notice* it. Terrible. That's why I ran HDTune that day, because the drive was suspiciously slow. And the gouge pretty well aligned perfectly with the OS. And the reallocated raw data field was still "zero". My drive might have had 1000 hours on it at that point. I have another drive here with 39000 hours on it, and it doesn't have a mark on it. The benchmark looks as good as the day I bought it. Very strange. I've had other drives, with the same model number as "Mr.Gouge" where the reallocated field went non-zero, and that's the first thing I knew about health not being what it should be. I retired those from OS service. but, as scratch drives, I've put many additional hours on those drives (more than one drive showed reallocations), and they still haven't failed! If you cannot improve the quality of the scan, you can: 1) Increase frequency of backups. 2) Simply watch for performance issues by "feel". If the drive doesn't feel right some day, pull it and retire it. I probably have five or six drives in the "retired" pool, and even the drive that let out a "squeak" one day, haven't caused any more problems. In the old days, drives had an anti-static "spring" which contacted the shaft of the spindle, and that's where a squeak would come from. Modern drives are not set up that way. And to me, if a modern drive squeaks, you'd better do (1) immediately. My rule of thumb is "the more noise a drive makes, the sooner it will break". The head lock on my Barracuda 32550N which jammed and ground the heads into the platter, it was plenty noisy before the failure. Every day, when the heads unlocked, it sounded like the "free game" relay on a pinball machine. Well, one day, instead of unlocking, a sound like a giant watch spring was emitted by the drive, and that was the heads being pushed face-first into the platter. I don't know of too many heads with sufficient flying height to "fly over" a channel cut in the platter :-) Paul |
#13
|
|||
|
|||
Can HDTune go slower where there is data, if something's set wrong?
In message , Paul
writes: J. P. Gilliver (John) wrote: Benchmark curve, using default settings (speed 2, 64K blocks): http://255soft.uk/temp/2018-3-9%201-...%2064%20KB.png No real difference - the flat line is still higher than you'd expect, and the gouge is still there. Doing the error scan now - it's done 7200 MB in nearly 5 minutes, so I'll let it run and upload when finished. Not showing any Damaged 381MB squares yet (and 8001 MB is beyond the initial glitches on the Benchmark curve). Do you _really_ think a new drive would have 10-15% bad sectors? (Even have that many "swap" ones, for that matter?) Now got up to "12 gB", no red squares yet. Completed - http://255soft.uk/temp/2018-3-9%205-58.png I take it you've reviewed the "Health" tab in HDTune, and verified the drive is new that way ? How many power on hours does it have ? Is the stated number of power on hours consistent with how long you've been using it ? Now showing 248 hours. Yes, that's about right: I bought it a week ago last Sunday, and fitted it within a couple of days, and haven't turned it off much since then. And it was in a sealed bag. Did you check Programs and Features for any third party cache programs ? Intel used to make a software cache program, and Samsung may have tried their hand at it too. I don't keep tabs on stuff like that, so can't suggest any programs names. This one is Toshiba. I can't see anything in Programs and Features that looks like a cacher. I've had *one* drive here, enter into the "gouge" pattern, right where the OS was located. And the drive had become so slow, it was doing maybe 20MB/sec or so and I could *notice* it. Terrible. That's why I ran HDTune that day, because the drive was suspiciously slow. And the gouge pretty well aligned perfectly with the OS. And the reallocated raw data field was still "zero". Oh dear. But at least this one is not giving me any cause for concern other than the HDTune graphs: if it wasn't for those, I'd not be worried at all. My drive might have had 1000 hours on it at that point. I have another drive here with 39000 hours on it, and it doesn't have a mark on it. The benchmark looks as good as the day I bought it. Very strange. I've had other drives, with the same model number as "Mr.Gouge" where the reallocated field went non-zero, and that's the first thing I knew about health not being what it should be. I retired those from OS service. but, as scratch drives, I've put many additional hours on those drives (more than one drive showed reallocations), and they still haven't failed! If you cannot improve the quality of the scan, you can: 1) Increase frequency of backups. 2) Simply watch for performance issues by "feel". If the drive doesn't feel right some day, pull it and retire it. I'll probably do both of those. I have three year guarantee specifically written on the receipt, so I'm OK there, apart from the inconvenience. I probably have five or six drives in the "retired" pool, and even the drive that let out a "squeak" one day, haven't caused any more problems. In the old days, drives had an anti-static "spring" which contacted the shaft of the spindle, and that's where a squeak would come from. Modern drives are not set up that way. And to me, if a modern drive squeaks, you'd better do (1) immediately. My rule of thumb is "the more noise a drive makes, the sooner it will break". The head lock on my Barracuda Indeed - I really don't expect _any_ noise from modern drives. 32550N which jammed and ground the heads into the platter, it was plenty noisy before the failure. Every day, when the heads unlocked, it sounded like the "free game" relay on a pinball machine. Well, one day, instead of unlocking, a sound like a giant watch spring was emitted by the drive, and that was the heads being pushed face-first into the platter. I don't know of too many heads with sufficient flying height to "fly over" a channel cut in the platter :-) So far, I've only had one that I _think_ was a contact fault - I think due to overheating (caused by the PC, not the drive); I couldn't hear it rotating even with my ear to it. After trying all the internet suggestions, I gave up and opened it (fortunately I had access to a positive-pressure clean air facility at work); sure enough the heads were at a position on the surface rather than parked. I applied slight rotational torque to the hub (it took the same size driver as the case screws), and I could feel it unstick, though I could _see_ no damage; after I put the case back on, I was able to get about 97-98% of the data off it, though definitely considered that one not even a junk drive as it had been opened. Since then I've been very aware of temperature, hence the recent concern when Corsair Link seemed to have turned off my fan! Paul John -- J. P. Gilliver. UMRA: 1960/1985 MB++G()AL-IS-Ch++(p)Ar@T+H+Sh0!:`)DNAf "I've got this shocking pain right behind the eyes." "Have you considered amputation?" - Vila & Avon |
#14
|
|||
|
|||
Can HDTune go slower where there is data, if something's setwrong?
J. P. Gilliver (John) wrote:
Oh dear. But at least this one is not giving me any cause for concern other than the HDTune graphs: if it wasn't for those, I'd not be worried at all. OK, found another way to do it. You can add a script to perfmon, and have it record Disk Read bytes/sec for a selected disk or for the whole disk pool. https://www.veritas.com/support/en_US/article.000036737 Using that, that gives an instantaneous bandwidth recorder. Your "script" is stored in the "Counter logs" section. Now, you could run the error scan again, as a stimulus for your perfmon recording. You can set the Start and Stop on your perfmon script to "manual", and then stop it from the menu in perfmon. Then go to C:\perflogs (it insists they go there) to find your numbered log. I presume if a log file gets large enough, it starts a new log or something. https://s10.postimg.org/zbk5kfqhl/us...r_csv_text.gif Anyway, that solves the OCR problem I was having. Another way to generate a stimulus, is with dd. http://www.chrysocome.net/dd dd --list # dump the naming convention # Harddisk2 is my golden disk, # Partition 0 the entire disk dd if=\\?\Device\Harddisk2\Partition0 bs=1048576 count=476940 NUL That reads my golden drive and throws the data away in NUL (the Windows equivalent of /dev/null). When you don't specify of= or of=- , the output defaults to STDOUT, the "" redirects STDOUT to NUL. (I did it this way, in case dd.exe doesn't like of=NUL . I made it so the issue is handled at the shall level via "". ) I stopped the dd command with ctrl-C , then stopped the perfmon counter script from its menu, and this is the test log. C:\PerfLogs\perftest.log_000001.csv "03/09/2018 04:43:57.156","0","DIsk Read Monitoring" "03/09/2018 04:43:58.156","0","DIsk Read Monitoring" "03/09/2018 04:43:59.156","43093231.386613525","DIsk Read Monitoring" "03/09/2018 04:44:00.156","123731611.41122638","DIsk Read Monitoring" "03/09/2018 04:44:01.156","124779021.6261901","DIsk Read Monitoring" "03/09/2018 04:44:02.156","124779218.15853371","DIsk Read Monitoring" "03/09/2018 04:44:03.156","123730390.24815789","DIsk Read Monitoring" "03/09/2018 04:44:04.156","121633710.42662525","DIsk Read Monitoring" "03/09/2018 04:44:05.156","123730268.35547325","DIsk Read Monitoring" "03/09/2018 04:44:06.156","124778744.01596877","DIsk Read Monitoring" "03/09/2018 04:44:07.156","124779738.2623236","DIsk Read Monitoring" "03/09/2018 04:44:08.156","123730429.52062416","DIsk Read Monitoring" "03/09/2018 04:44:09.156","118491019.49676122","DIsk Read Monitoring" "03/09/2018 04:44:10.156","124779564.52027358","DIsk Read Monitoring" "03/09/2018 04:44:11.156","123731317.9742112","DIsk Read Monitoring" The title on the end, is the name I typed in for the script name. From one second to the next, there is some variation in the amount of data charged to that second. You should be able to plot those in Excel, as the data is CSV. If there is caching or read-ahead or any other distorting behavior, it's still going to be there. But at least now there will be no more OCR experiments :-) Paul |
#15
|
|||
|
|||
Can HDTune go slower where there is data, if something's set wrong?
In message , Paul
writes: J. P. Gilliver (John) wrote: Oh dear. But at least this one is not giving me any cause for concern other than the HDTune graphs: if it wasn't for those, I'd not be worried at all. OK, found another way to do it. I'm afraid my ability (or, perhaps, willingness) to take in the level of detail in your instructions isn't what it was even a few years ago. I'll have to stop asking questions here, as I don't want to put you and others to the amount of effort you're putting in, only for me to waste your effort by saying "that's too complicated for me". For the moment, unless (is anyone else still reading this thread?) a _simple_ drive health method comes to light (that actually tells me there _is_ something wrong with my new drive), I'll just perhaps do the two things you suggested in an earlier post: 1. do backups more often, 2. listen out for any noises. Sorry to have wasted your time. You can add a script to perfmon, and have it record Disk Read bytes/sec for a selected disk or for the whole disk pool. https://www.veritas.com/support/en_US/article.000036737 That looks a well-written article - certainly written by people who understand the subject very well - but is a bit more than I can take in. (Also, it's titled "How to use Performance Monitoring (Perfmon.msc) to investigate slow disk performance (DSSU)" - and I'm still not _convinced_ that I _do_ have slow disc performance, or in fact anything wrong with the drive at all.) Using that, that gives an instantaneous bandwidth recorder. Your "script" is stored in the "Counter logs" section. Now, you could run the error scan again, as a stimulus for your perfmon recording. You can set the Start and Stop on your perfmon script to "manual", and then stop it from the menu in perfmon. Then go to C:\perflogs (it insists they go there) to find your numbered log. I presume if a log file gets large enough, it starts a new log or something. https://s10.postimg.org/zbk5kfqhl/us...r_csv_text.gif Anyway, that solves the OCR problem I was having. Another way to generate a stimulus, is with dd. http://www.chrysocome.net/dd dd --list # dump the naming convention # Harddisk2 is my golden disk, # Partition 0 the entire disk dd if=\\?\Device\Harddisk2\Partition0 bs=1048576 count=476940 NUL That reads my golden drive and throws the data away in NUL (the Windows equivalent of /dev/null). When you don't specify of= or of=- , the output defaults to STDOUT, the "" redirects STDOUT to NUL. (I did it this way, in case dd.exe doesn't like of=NUL . I made it so the issue is handled at the shall level via "". ) I stopped the dd command with ctrl-C , then stopped the perfmon counter script from its menu, and this is the test log. C:\PerfLogs\perftest.log_000001.csv "03/09/2018 04:43:57.156","0","DIsk Read Monitoring" "03/09/2018 04:43:58.156","0","DIsk Read Monitoring" "03/09/2018 04:43:59.156","43093231.386613525","DIsk Read Monitoring" "03/09/2018 04:44:00.156","123731611.41122638","DIsk Read Monitoring" "03/09/2018 04:44:01.156","124779021.6261901","DIsk Read Monitoring" "03/09/2018 04:44:02.156","124779218.15853371","DIsk Read Monitoring" "03/09/2018 04:44:03.156","123730390.24815789","DIsk Read Monitoring" "03/09/2018 04:44:04.156","121633710.42662525","DIsk Read Monitoring" "03/09/2018 04:44:05.156","123730268.35547325","DIsk Read Monitoring" "03/09/2018 04:44:06.156","124778744.01596877","DIsk Read Monitoring" "03/09/2018 04:44:07.156","124779738.2623236","DIsk Read Monitoring" "03/09/2018 04:44:08.156","123730429.52062416","DIsk Read Monitoring" "03/09/2018 04:44:09.156","118491019.49676122","DIsk Read Monitoring" "03/09/2018 04:44:10.156","124779564.52027358","DIsk Read Monitoring" "03/09/2018 04:44:11.156","123731317.9742112","DIsk Read Monitoring" The title on the end, is the name I typed in for the script name. From one second to the next, there is some variation in the amount of data charged to that second. You should be able to plot those in Excel, as the data is CSV. If there is caching or read-ahead or any other distorting behavior, it's still going to be there. But at least now there will be no more OCR experiments :-) Paul John -- J. P. Gilliver. UMRA: 1960/1985 MB++G()AL-IS-Ch++(p)Ar@T+H+Sh0!:`)DNAf My movies rise below vulgarity. - Mel Brooks, quoted by Barry Norman in RT 2016/11/26-12/2 |
|
Thread Tools | |
Display Modes | Rate This Thread |
|
|