6.xx is 2.25 times slower than 5.59 !!!

[RayMark]

I am mostly downloading directly to 5400 rpm USB 2.0 drives (I use them instead of blank DVDs, basically).

When copying a large file to such a drive I get write speed about 200 Mbps
When downloading large files with Newbin 5.59 - the speed is about 90 Mbps - with 10 simultaneous download threads
When downloading large files with Newsin 6.10 RC4 - the speed is about 40 Mbps - with 10 simultaneous download threads, but also about 90 Mbps with 1 download thread.

The CPU and memory usage is pretty low with 6.10.
So the low download speed probably is caused by excessive operations with the disk - and clearly 6.10 cannot handle as many download threads as 5.59 can.
Such a speed difference is pretty significant.
A larger number of simultaneous threads would help to cope with network congestions, using just one might not ensure stable speed.

BTW, the 90 Mbps top speed may be a result of bandwidth limitations due to network congestions somewhere on the way as well, not by Newsbin - but unlikely, it is pretty stable.
Besides, 6.10 with 1 thread seems to be slightly faster than 5.59 with 10: maybe about 92 Mbps.
More testing is needed, perhaps also with faster sata drives, and with various numbers of threads with both versions.
And also with locating chunks on a different or on the same drive as downloads - if it is still possible in 6.10, I can't find it in options.

For my tests I gave Newsbin (up to) 300 Mbps bandwidth.
Clearly, a big overkill.
And the bottleneck might be Newsbin - otherwise I would see something closer to 200 Mbps. Obviously, to achieve the whole 200 Mbps would be unrealistic.

But the difference between 5.xx and 6.10 is very significant.
And I hope it can be eliminated.
As I understand, the changes in 6.xx are supposed to be related to the handling of headers, not to downloading.
There is no reason to have more extensive disk operations, is there?
I am speaking about the hard drive to which files are downloaded, the NBData with headers is on an entirely different fast sata drive - operations with headers should not slow down downloads to a different disk.

[mho]

If possible, set the download directory to an internal (faster) disk/SSD and use "AutoPAR Options->Override UnRAR Path" (or some other mechanism) to extract the files to the external drive. While newsbin is downloading, it is doing a lot of disk operations (seeks and writes) in batches, so a slow disk will have trouble keeping up. The unrar is much more sequential, so it will work better with a slower disk, especially if the .rar files are on another disk so you don't have trashing due to reading and writing at the same time from the same disk.

- mho

[RayMark]

I'm just surprised how significant the difference (regress) is between 5.x and 6.x when using 10 threads.
Even if the handling of headers is different, the downloading/decoding itself should be similar in intensity of disk operations.
Fortunately, my bandwidth is so huge, that I can afford to reduce the number of threads to 1-2 and the problem goes away.

As to unrarring, I do not use auto-unrar at all. Just download and unrar later myself.
Which might be stupid. I also prefer command line to mouse.

[itimpi]

I must admit I do not understand why you see this behavior? I get significantly faster throughput using v6 compared to v5, and I typically run about 16 download threads.

[Quade]

Yeah, for me 6 is easily twice as fast as 5.59.

You have many weird problems that I'm not seeing in addition to this. Internal crashes, DB problems, unable to map file problems (which shouldn't even be possible because the file is already open). I'm not sure where to take it when you report problems like this. I'm not ignoring it. I just have no way to reproduce it.

[RayMark]

I will have to retest with RC5.

Yesterday I was downloading with RC5, and I thought that I had reduced the number of threads to 2, but turns out, I reduced the number to 2 for the fill-in server only, the main one was still on 10.
And the speed was close to 100 Mbps.
When I wrote this days ago, I was getting 40 Mbps with 6.x (RC1, I think) and about 90 Mbps with 5.x

Among my weird problems there was not working search in the loaded group with RC2 and RC3, but working with RC1 and again working with RC5.

Perhaps the speed problem went away together with the search problem...