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
|
|||
|
|||
performance when copying large files
This has been discussed quite a lot before, but it
continues to interest me. I create a full image backup of my system once a week and an incremental backup each of the next six days. Backups (Acronis) go on a spare HD partition and I then copy them to an external (WD) Passport 2GB drive. Copying the full backup image takes over an hour. What I've noticed, usually, but not every time (!?) is that the transfer gradually slows down. See: https://imgur.com/hESFJq1 While the copy proceeds there is no other activity beyond the usual Windows background stuff running - no apps, browsers, etc. What might account for this? |
Ads |
#2
|
|||
|
|||
performance when copying large files
Jason wrote:
This has been discussed quite a lot before, but it continues to interest me. I create a full image backup of my system once a week and an incremental backup each of the next six days. Backups (Acronis) go on a spare HD partition and I then copy them to an external (WD) Passport 2GB drive. Copying the full backup image takes over an hour. What I've noticed, usually, but not every time (!?) is that the transfer gradually slows down. See: https://imgur.com/hESFJq1 While the copy proceeds there is no other activity beyond the usual Windows background stuff running - no apps, browsers, etc. What might account for this? I've seen this before, and one of the symptoms seems to be the system is leaking paged pool. And a garbage collector tries to scrape enough paged pool together to keep the system running. Modern Windows OSes allow a lot more of system memory to be used as Paged Pool, but this still won't prevent such behaviors from happening (eventually). The strange part, is what is leaking pool, and what combination of drivers does this ? I've never been able to definitely trace this down. I think I tried running PoolMon once, and didn't manage to reproduce the problem. The problem can become so bad, that the system will suffer "Delayed Write Failure" in the Event Log (that's if the Event Log can even be updated). You could probably simulate this situation, by using the evaluation version of the NeatVideo plugin, which used to attempt to limit customer usage of the evaluation version, by leaking pool on purpose. And you'll see the system become slower and slower. If you're not careful, you won't return to the system at the right moment in time, and your opportunity to run Task Manager and kill it, will be lost. Then, you have to use the power button to recover. The symptoms I've seen a 1) Report of increased page pool occupancy. (Easier to see in the WinXP Task Manager) 2) Declining write rate. Until eventually writes begin to time out, and an Event Viewer log entry is created. It might take 12 hours for it to get this bad. 3) Increasing CPU activity, as the OS attempts to garbage collect released pool memory. Even if I knew which application was doing it, that would be a start. But the thing is, drivers left behind in a system, can continue to be a cause of the issue, without the program in Ring3 running at the time. Drivers can cause problems "just by being there" when they're filter drivers and all the data flows through them anyway. Acronis and Macrium install services. Acronis is the poster boy "for adding stuff", because they love to dress up their products with an excess of services added (they will add more than one). I never get the feeling everything Acronis adds, is actually there for a reason. Another potential source of conflict, could be virtual machine hosting software. It's a wild possibility. Paul |
#3
|
|||
|
|||
performance when copying large files
|
#4
|
|||
|
|||
performance when copying large files
|
#5
|
|||
|
|||
performance when copying large files
Jason wrote:
In article - september.org, says... Thanks I should add - to make it weirder - that I don't always see this behavior. Sometimes, the transfer proceeds at a high rate to the end. Other times, it exhibits the gradual slowdown and then, suddently, jumps back up to what I believe is the max rate and stays there... ? That's not a match for the one I've seen. Task manager has the "resmon" button in it, which displays more information about resource usage. But even with that, I doubt there will be sufficient info to say "Aha!". I don't know of a really good way to debug stuff like this. All I could say, is it seemed to have a correlation with the state of Pool Memory usage on mine. Paul |
#6
|
|||
|
|||
performance when copying large files
On 5/20/2018 7:58 PM, Jason wrote:
This has been discussed quite a lot before, but it continues to interest me. I create a full image backup of my system once a week and an incremental backup each of the next six days. Backups (Acronis) go on a spare HD partition and I then copy them to an external (WD) Passport 2GB drive. Copying the full backup image takes over an hour. What I've noticed, usually, but not every time (!?) is that the transfer gradually slows down. See: https://imgur.com/hESFJq1 While the copy proceeds there is no other activity beyond the usual Windows background stuff running - no apps, browsers, etc. What might account for this? I've seen odd slowdowns using file manager. Same copy using totalcommander doesn't have that problem. How fast is your external drive? I'm assuming you meant 2TB?? When I copy to external media, it starts fast, but then drops to a much lower speed somewhere around the 1GB transferred point. Pausing for a while restores the speed. I'm theorizing that the fast speed is how fast the internal drive can write to a buffer. The slow speed is the speed that the external drive can accept the data. This is exceptionally obnoxious with 256GB USB thumb drives. |
#7
|
|||
|
|||
performance when copying large files
Jason wrote:
This has been discussed quite a lot before, but it continues to interest me. I create a full image backup of my system once a week and an incremental backup each of the next six days. Backups (Acronis) go on a spare HD partition and I then copy them to an external (WD) Passport 2GB drive. Copying the full backup image takes over an hour. What I've noticed, usually, but not every time (!?) is that the transfer gradually slows down. See: https://imgur.com/hESFJq1 While the copy proceeds there is no other activity beyond the usual Windows background stuff running - no apps, browsers, etc. What might account for this? You know, rather than cook up some corny explanation, we'd better be careful here that this isn't "ordinary" sharing of a resource. An Intel chipset (PCH), has a set-of-six ports off one 60MB/sec USB2 controller and a set-of-eight ports off a second 60MB/sec controller. The bandwidth is shared inside a group. That's an example of some sharing going on. Your Imgur picture looks to be going faster than that, so that's not it. I don't know how, or whether, USB3 does sharing. While we're all used to seeing ~400MB/sec transfers from a single USB3 device, what would happen if your config had some sort of sharing going on ? Have a look at what ports are being used in this scenario of yours, so we don't end up postulating space aliens for nothing :-) Even some HDTune pictures of the source device and the destination device, might provide you with some hints of how the stuff works. Paul |
#8
|
|||
|
|||
performance when copying large files
Paul wrote:
Modern Windows OSes allow a lot more of system memory to be used as Paged Pool, but this still won't prevent such behaviors from happening (eventually). The strange part, is what is leaking pool, and what combination of drivers does this ? I've never been able to definitely trace this down. I think I tried running PoolMon once, and didn't manage to reproduce the problem. I did manage to identify leaky drivers a few times with poolmon, but that was back in the days of NT4 and W2K, not needed it for many years. |
#9
|
|||
|
|||
performance when copying large files
|
#10
|
|||
|
|||
performance when copying large files
In article , ham789
@netzero.net says... On 5/20/2018 7:58 PM, Jason wrote: This has been discussed quite a lot before, but it continues to interest me. I create a full image backup of my system once a week and an incremental backup each of the next six days. Backups (Acronis) go on a spare HD partition and I then copy them to an external (WD) Passport 2GB drive. Copying the full backup image takes over an hour. What I've noticed, usually, but not every time (!?) is that the transfer gradually slows down. See: https://imgur.com/hESFJq1 While the copy proceeds there is no other activity beyond the usual Windows background stuff running - no apps, browsers, etc. What might account for this? I've seen odd slowdowns using file manager. Same copy using totalcommander doesn't have that problem. How fast is your external drive? I'm assuming you meant 2TB?? When I copy to external media, it starts fast, but then drops to a much lower speed somewhere around the 1GB transferred point. Pausing for a while restores the speed. I'm theorizing that the fast speed is how fast the internal drive can write to a buffer. The slow speed is the speed that the external drive can accept the data. This is exceptionally obnoxious with 256GB USB thumb drives. Opps. Yes, TB not GB. It's a USB3 drive. When things are clipping along the copy proceeds at 100MB/s +/- a little. I see the slowdown start very early into the transfer. The graph shown is of a 450+ GB transfer. I haven't tried pausing it and re-starting. I'll see if that has an effect. |
#12
|
|||
|
|||
performance when copying large files
pjp wrote:
In article , says... In article , ham789 @netzero.net says... On 5/20/2018 7:58 PM, Jason wrote: This has been discussed quite a lot before, but it continues to interest me. I create a full image backup of my system once a week and an incremental backup each of the next six days. Backups (Acronis) go on a spare HD partition and I then copy them to an external (WD) Passport 2GB drive. Copying the full backup image takes over an hour. What I've noticed, usually, but not every time (!?) is that the transfer gradually slows down. See: https://imgur.com/hESFJq1 While the copy proceeds there is no other activity beyond the usual Windows background stuff running - no apps, browsers, etc. What might account for this? I've seen odd slowdowns using file manager. Same copy using totalcommander doesn't have that problem. How fast is your external drive? I'm assuming you meant 2TB?? When I copy to external media, it starts fast, but then drops to a much lower speed somewhere around the 1GB transferred point. Pausing for a while restores the speed. I'm theorizing that the fast speed is how fast the internal drive can write to a buffer. The slow speed is the speed that the external drive can accept the data. This is exceptionally obnoxious with 256GB USB thumb drives. Opps. Yes, TB not GB. It's a USB3 drive. When things are clipping along the copy proceeds at 100MB/s +/- a little. I see the slowdown start very early into the transfer. The graph shown is of a 450+ GB transfer. I haven't tried pausing it and re-starting. I'll see if that has an effect. Don't those things use some type of buffer? I'd assume once buffer is full and keeps being filled the actual writing would slow down. Small file and buffer doesn't get filled, large file and it's continually being filled. Modern hard drives have a 64MB or 128MB cache on the drive controller board. For a sustained transfer scenario, that buffer could fill in under a second (the buffer could fill if the input rate exceeded the platter rate, subject to the tagged buffer limits of AHCI). The System Write Cache can be larger. It seems to use a maximum of a certain percentage of the system RAM. Like maybe 8% or so. You can watch the available system RAM drop, if the source storage device is faster than the destination storage device. And the System Write Cache isn't that smooth either. It doesn't drain in as predictable a way as it should. But it still doesn't account for the transfer curve in the picture. You can use perfmon.msc and select some disk read and disk write counters, to get a "second opinion" on what the system is actually doing at the time. There is also an option in the snap-in, to log the counters to a text file, for later study (curve fitting or whatever). A hard drive is 2x as fast near the outer circumference, than near the hub, but I hope that's not what we're seeing in that transfer curve. You can use HDTune to convince yourself what the curve looks like when the entire drive is being used up. I have drives here, that start at 200MB/sec on the outer diameter, and finish up at 100MB/sec near the hub. And the perfmon.msc graph doesn't show enough minutes of the transfer, to show that characteristic in one view. Whereas the File Explorer transfer curve shows info for the entire period of the transfer (even if it exaggerates the beginning of the transfer, when it's filling the write cache or something). But these are all effects you've seen before, so none of them should be a real surprise. Windows is like "file transfer theater" when it comes to this stuff :-) The OPs picture looks a bit too flat and "straight line" to be a disk circumference issue. If the decline in speed showed some overall curve to it, just the hard drive characteristic might explain it. I would need to know: 1) Disk fill before transfer (assumes files are packed to the left on the partition, but they could really be anywhere in actual conditions). Like say "I had 100GB on the 500GB USB3 hard drive before I started". 2) Size of transfer - "I was adding a 250GB file to the 100GB of files already on the drive". That would make it easier to justify the decline in transfer rate being linear, if the sampling we're seeing in the File Transfer dialog is showing us just a small portion of the entire disk surface transfer speed curve. If you had a 500GB short-stroked drive, those can almost maintain their transfer rate over the entire surface. WDC shipped a few of those, when closing out the 500GB 3.5" business. They were then using 1TB drives, and only activating the first half of the circumference (the faster "outer donut"), rather than continuing to make 500GB drives. And when you do that on a drive, you get to keep the best part of the transfer curve. But WDC wasn't consistent, because two drives in the pile at the computer store, one was short stroked and the second one wasn't. And the serial numbers didn't betray anything. I had no prior warning that's what I was getting - this is not in the spec sheet. The one on the right here, shows the regular 2:1 ratio between the outer circumference and the inner circumference. The one on the left is a short-stroked drive 1TB in size, where only the outer 500GB is used. The the speed drops from about 175MB/sec to 155MB/sec over the available surface. http://s29.postimg.cc/8b7cj872v/wd500gb.gif Paul |
Thread Tools | |
Display Modes | Rate This Thread |
|
|