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 |
#76
|
|||
|
|||
SSDs/HDDs, memory ... (was: Have hardware prices gone crazy during Covid?)
On Tue, 30 Jun 2020 12:02:41 +0100, "J. P. Gilliver (John)"
wrote: On Tue, 30 Jun 2020 at 09:12:20, Chris wrote: [] The only downside to SSDs is the per GB cost. I think the failure mode is another downside. (Sure, it isn't if you have a robust and frequent backup strategy.) [And I'm talking statistically/probabilistically; yes, I know, HDDs _can_ fail in those ways too (other than going read-only).] Whether you're talking about failure _mode_ or failure _likelihood_, I think SSDs win out over HDDs. As mentioned above, HDDs still win on cost per GB, but that gap is closing, especially for smaller amounts of storage. |
Ads |
#77
|
|||
|
|||
Have hardware prices gone crazy during Covid?
On 30/06/2020 16:44, Stan Brown wrote:
On Mon, 29 Jun 2020 00:18:33 +0200, Carlos E.R. wrote: On 28/06/2020 21.52, tesla sTinker wrote: bill gates is the corona virus.Â* He made it on purpose.\ Kill this subthread. Please don't respond to trolls -- it just encourages them to keep posting. Hard as it may be, the most effective strategy, over time, is to make no comment of any kind. Well, yes and no, it depends on the nature of the trolling. There is a particular problem with fake news, and while most people may be unaffected by it and just ignore it, and that's fine for them, I think there's now so much of it that it becomes almost accepted as 'fact', 'pseudo-fact' if you like, just by its ubiquity and that no-one can be arsed to contradict it. Hence my debunking of it in this case. I don't suppose for one moment I will persuade an empty-headed repeater of fake news like the one at fault here, but at least anyone else who unthinking comes along and steps in his **** has some facts to wipe their boots on. |
#78
|
|||
|
|||
SSDs/HDDs, memory ... (was: Have hardware prices gone crazyduring Covid?)
On 30/06/2020 15.42, nospam wrote:
you're also ignoring hard drive seek time, which is a significant overhead for smaller files as well as fragmented larger files. Absolutely. seek time on an ssd is effectively zero. that alone is a benefit, even without the faster i/o. Indeed it is. This is what makes the change to SSD brilliant in use, not only boot. This is the issue with my database example. The point being that in the vast majority of things that people do, an SSD offers no advantage. You're not going to is a difference if MS Word is saving your DOC to disk once every 5 minutes, spending 1 ms to do it instead of 2 ms. that is simply false. replacing a hard drive with an ssd will result in a substantial overall improvement in performance. the difference might be a little less noticeable with ms word since the limiting factor is the user typing, but that's not the only app people use. Nor is typing the only thing done with Word. Try to shape a brochure or magazine article, any complex document. -- Cheers, Carlos. |
#79
|
|||
|
|||
SSDs/HDDs, memory ... (was: Have hardware pricesgone crazy during Covid?)
J. P. Gilliver (John) wrote:
On Tue, 30 Jun 2020 at 09:12:20, Chris wrote: [] The only downside to SSDs is the per GB cost. I think the failure mode is another downside. You should always work in the basis that hardware will fail catastrophically. Don't depend on being able to rescue stuff at first signs of failure. At those first signs you should be in a position to just junk it. |
#80
|
|||
|
|||
SSDs/HDDs, memory ... (was: Have hardware pricesgone crazy during Covid?)
Mayayana wrote:
"Chris" wrote | It's important for *any* file I/O. The massive gain is the lack of latency | with all disk operations. Given that all applications touch a lot of files | in their general operation every micro/milli-second gain adds up. It is | genuinely transformative. | Yes, that's true. That's what Carlos was talking about with his on-disk database work. But most people are not doing massive file access. Most desktop applications are 100s of megabytes in size and read a great many files upon launching. I open a file in a text editor or graphic editor. Most of what's happening is in RAM until I save to disk. Overly simplistic. I load a webpage. It' save cache, etc, to disk. Even on older disks, the speed of loading a file is in the range of maybe 1 ms per MB. (I've tested it before and don't remember exactly, but it's extremely fast.) Browsers read a lot of library file, check caches, cookies and the downloads folder for new or existing files. They also load plugins and other addons. The point being that in the vast majority of things that people do, an SSD offers no advantage. You're not going to is a difference if MS Word is saving your DOC to disk once every 5 minutes, spending 1 ms to do it instead of 2 ms. Word has always been awful when saving files. It seems to check every drive responds before it even considers saving your existing open file in the place it has always been. I remember Word 6 would make the floppy drive seek every bloody time you hit save. Despite there being no disk it and you wanting to save to C: Anything that makes saving files better with word is *always* noticeable. | On my desktop I've both an SSD on C: and an HDD on D:. The difference in | application performance if it's installed on C or D is like night and day. | Then you must be using I/O-intensive software, or there's something wrong with your system. It depends on what you mean by "performance". I'm used to my responsive MacBook Pro with its very fast SSD so any kind of lag is very noticeable. |
#81
|
|||
|
|||
SSDs/HDDs, memory ... (was: Have hardware pricesgone crazy during Covid?)
Mayayana wrote:
"Chris" wrote | It's important for *any* file I/O. The massive gain is the lack of latency | with all disk operations. Given that all applications touch a lot of files | in their general operation every micro/milli-second gain adds up. It is | genuinely transformative. | Yes, that's true. That's what Carlos was talking about with his on-disk database work . Poorly performing databases are always due to poor design of the database or poorly constructed queries. It should not be I/O bound. That's the point of them, otherwise you're better off with text files. |
#82
|
|||
|
|||
SSDs/HDDs, memory ...
Chris wrote:
Mayayana wrote: "Chris" wrote | It's important for *any* file I/O. The massive gain is the lack of latency | with all disk operations. Given that all applications touch a lot of files | in their general operation every micro/milli-second gain adds up. It is | genuinely transformative. | Yes, that's true. That's what Carlos was talking about with his on-disk database work . Poorly performing databases are always due to poor design of the database or poorly constructed queries. It should not be I/O bound. That's the point of them, otherwise you're better off with text files. I'll have a bench soon, and we can discuss it. ******* OK, this benchmark involves a million files on RAMDisks, using two different file systems. The directory "0" contains a tree of folders, with 16 files per folder at the bottom. If all million files were placed in one folder, NTFS would "sink like a rock". I used the tree approach to give it a fighting chance. This filesystem is TMPFS which is a RAMDisk-based thing in Linux. This filesystem has a bit uneven performance, and can slow down a bit as it "ages". It's not perfect, but it does turn in some nice benchmarks. Since it completes the benchmark in ten seconds, that's roughly 100,000 files a second. time hashdeep -c md5 -j 12 -r /tmp/0 output.txt real 0m9.936s === 10 seconds user 0m19.495s sys 0m25.902s I also have NTFS on RAMDisk of the same size, in Windows. DataRAM RAMDisk. The TMPFS can do around 7.5GB/sec, while the Windows RAMdisk does around 4GB/sec. F:\timeit hashdeep64 -c MD5 -j 12 -r 0 output.txt Version Number: Windows NT 6.2 (Build 9200) Exit Time: 6:15 pm, Tuesday, June 30 2020 Elapsed Time: 0:01:49.568 === 110 seconds Process Time: 0:04:56.078 System Calls: 40544444 Context Switches: 12817643 Page Faults: 2772716 Bytes Read: 3098178548 Bytes Written: 70984398 Bytes Other: 665675206 NTFS is eleven times slower. It's holding back my Windows result. And this is a relatively good result for NTFS. I've seen worse. It makes some of my experiments... take hours. Paul |
#83
|
|||
|
|||
SSDs/HDDs, memory ... (was: Have hardware prices gone crazy during Covid?)
"Carlos E.R." wrote
| The point being that in the vast majority of things | that people do, an SSD offers no advantage. You're not | going to is a difference if MS Word is saving your DOC | to disk once every 5 minutes, spending 1 ms to do it | instead of 2 ms. | | You do if you handle a 20 page document with photos or figures on every | page. Or with active links to calc sheet portions. I can hear the disk | working as I go around. | But you're not waiting for disk writes, right? I think SSDs are great, personally. I just think it's important to distinguish between theoretical and practical advantages. |
#84
|
|||
|
|||
SSDs/HDDs, memory ...
On 6/29/2020 4:19 PM, Carlos E.R. wrote:
On 29/06/2020 21.30, Yousuf Khan wrote: On 6/29/2020 7:08 AM, Carlos E.R. wrote: Well, the attribute/permission set is different, which affects both directions. Maybe more issues? Legal? I'm sure one of the older public-domain filesystems, such as Ext3fs can be modified to include Microsoft attributes instead? I mean the source code is completely available for free. Yes, but then your own code based on it would also have to be available for free to competitors, and is not in the M$ DNA ;-p Yeah, but if it's just to add the information like the permissions and ownership information, then what kind of competitive information is left in that now? At this point in time, NTFS has been thoroughly reverse-engineered already, and most external OS's can handle NTFS as a side-gig these days, so they already know most of the attributes inside it. Besides that was the old Microsoft, this is the new Linux-friendly Microsoft. :-) Maybe they would need a team to extract the full detailed specs, and another team that never ever had a look at the code, create another code set from scratch. It seems like a total waste of time, when the source code is already available for them for free. All they'd have to do is reveal the hierarchy for their permission and ownership attributes. I don't see where there is any competitive advantage left for hiding that information anymore. Yousuf Khan |
#85
|
|||
|
|||
SSDs/HDDs, memory ...
On 6/30/2020 5:27 AM, Chris wrote:
Some years back I was hearing rumours of Microsoft developing its MS-SQL database software as its new filesystem. I guess that didn't pan out? It was called WinFS https://en.m.wikipedia.org/wiki/WinFS Yeah, that's the one. Yousuf Khan |
#86
|
|||
|
|||
Have hardware prices gone crazy during Covid?
On 6/30/2020 12:54 PM, Java Jive wrote:
Well, yes and no, it depends on the nature of the trolling.Â* There is a particular problem with fake news, and while most people may be unaffected by it and just ignore it, and that's fine for them, I think there's now so much of it that it becomes almost accepted as 'fact', 'pseudo-fact' if you like, just by its ubiquity and that no-one can be arsed to contradict it.Â* Hence my debunking of it in this case.Â* I don't suppose for one moment I will persuade an empty-headed repeater of fake news like the one at fault here, but at least anyone else who unthinking comes along and steps in his **** has some facts to wipe their boots on. Yep, you got to do a little to combat it, but don't go overboard in trying to convince anyone, once they start showing signs that they can't be convinced, then start ignoring them. Yousuf Khan |
#87
|
|||
|
|||
SSDs/HDDs, memory ...
On 01/07/2020 11.41, Yousuf Khan wrote:
On 6/29/2020 4:19 PM, Carlos E.R. wrote: On 29/06/2020 21.30, Yousuf Khan wrote: On 6/29/2020 7:08 AM, Carlos E.R. wrote: Well, the attribute/permission set is different, which affects both directions. Maybe more issues? Legal? I'm sure one of the older public-domain filesystems, such as Ext3fs can be modified to include Microsoft attributes instead? I mean the source code is completely available for free. Yes, but then your own code based on it would also have to be available for free to competitors, and is not in the M$ DNA ;-p Yeah, but if it's just to add the information like the permissions and ownership information, then what kind of competitive information is left in that now? At this point in time, NTFS has been thoroughly reverse-engineered already, and most external OS's can handle NTFS as a side-gig these days, so they already know most of the attributes inside it. Besides that was the old Microsoft, this is the new Linux-friendly Microsoft. :-) Maybe they would need a team to extract the full detailed specs, and another team that never ever had a look at the code, create another code set from scratch. It seems like a total waste of time, when the source code is already available for them for free. All they'd have to do is reveal the hierarchy for their permission and ownership attributes. I don't see where there is any competitive advantage left for hiding that information anymore. It is available for free, if you comply with the requirements of the license of the code you use. It is not that simple. Possibly, all licenses of M$ libraries that use the library containing the imported code would have to be opened as well. It has been called a viral license. -- Cheers, Carlos. |
#88
|
|||
|
|||
SSDs/HDDs, memory ... (was: Have hardware prices gone crazyduring Covid?)
On 01/07/2020 01.10, Mayayana wrote:
"Carlos E.R." wrote | The point being that in the vast majority of things | that people do, an SSD offers no advantage. You're not | going to is a difference if MS Word is saving your DOC | to disk once every 5 minutes, spending 1 ms to do it | instead of 2 ms. | | You do if you handle a 20 page document with photos or figures on every | page. Or with active links to calc sheet portions. I can hear the disk | working as I go around. | But you're not waiting for disk writes, right? Of course we are. I think SSDs are great, personally. I just think it's important to distinguish between theoretical and practical advantages. Yes, and I can measure the practical advantages in overall speed of the machines where I change to SSD. -- Cheers, Carlos. |
#89
|
|||
|
|||
SSDs/HDDs, memory ... (was: Have hardware prices gone crazyduring Covid?)
On 30/06/2020 23.47, Chris wrote:
Mayayana wrote: "Chris" wrote | It's important for *any* file I/O. The massive gain is the lack of latency | with all disk operations. Given that all applications touch a lot of files | in their general operation every micro/milli-second gain adds up. It is | genuinely transformative. | Yes, that's true. That's what Carlos was talking about with his on-disk database work . Poorly performing databases are always due to poor design of the database or poorly constructed queries. It should not be I/O bound. That's the point of them, otherwise you're better off with text files. Exactly. -- Cheers, Carlos. |
#90
|
|||
|
|||
SSDs/HDDs, memory ...
On 01/07/2020 00.38, Paul wrote:
I'll have a bench soon, and we can discuss it. ******* OK, this benchmark involves a million files on RAMDisks, using two different file systems. The directory "0" contains a tree of folders, with 16 files per folder at the bottom. If all million files were placed in one folder, NTFS would "sink like a rock". I used the tree approach to give it a fighting chance. This filesystem is TMPFS which is a RAMDisk-based thing in Linux. This filesystem has a bit uneven performance, and can slow down a bit as it "ages". It's not perfect, but it does turn in some nice benchmarks. Since it completes the benchmark in ten seconds, that's roughly 100,000 files a second. time hashdeep -c md5 -j 12 -r /tmp/0 output.txt realÂ*Â*Â* 0m9.936sÂ* === 10 seconds userÂ*Â*Â* 0m19.495s sysÂ*Â*Â* 0m25.902s I also have NTFS on RAMDisk of the same size, in Windows. DataRAM RAMDisk. The TMPFS can do around 7.5GB/sec, while the Windows RAMdisk does around 4GB/sec. F:\timeit hashdeep64 -c MD5 -j 12 -r 0 output.txt Version Number:Â*Â* Windows NT 6.2 (Build 9200) Exit Time:Â*Â*Â*Â*Â*Â*Â* 6:15 pm, Tuesday, June 30 2020 Elapsed Time:Â*Â*Â*Â* 0:01:49.568Â*Â* === 110 seconds Process Time:Â*Â*Â*Â* 0:04:56.078 System Calls:Â*Â*Â*Â* 40544444 Context Switches: 12817643 Page Faults:Â*Â*Â*Â*Â* 2772716 Bytes Read:Â*Â*Â*Â*Â*Â* 3098178548 Bytes Written:Â*Â*Â* 70984398 Bytes Other:Â*Â*Â*Â*Â* 665675206 NTFS is eleven times slower. It's holding back my Windows result. And this is a relatively good result for NTFS. I've seen worse. It makes some of my experiments... take hours. That's very interesting. I noticed, years ago, that accessing in Linux an NTFS disk (mounted via fuse, thus not kernel) was slow and CPU intensive. At that time my CPU was not very powerful: people with more powerful CPUs did not see the problem. Of course, a reverse engineered code running with fuse has to be slower, but your testing confirms that there are other issues. It heard that NTFS will now be implemented in the Linux kernel :-? That would speed things a bit. -- Cheers, Carlos. |
Thread Tools | |
Display Modes | Rate This Thread |
|
|