Development mailing list

Syndicate content
Archive of posts for haiku-development at FreeLists
Updated: 8 min 17 sec ago

[haiku-development] Re: hpkgs and compression (Julian Harnath)

Mon, 2014-03-24 16:45
Jonathan Schleifer js-haiku-development@xxxxxxxxxxx schrieb: I haven't tested it, but I remember that even not-too-old-run-haiku hardware only achieved below 1 MB/s for Deflate with zlib. I guess we don't need to debate whether a few 100 KB/s is fast enough, right? :) Because that would be as fast as just streaming everything from the internet (well, streaming from the internet would even be faster with today's connections…). Another option would be switching the hpkg compression algorithm to LZ4, which is optimized for performance and many times faster than ...
Categories: Development

[haiku-development] ProcessController scheduler mode naming (Humdinger)

Mon, 2014-03-24 16:45
Hi there, ProcessController offers two scheduler modes Low latency and Power saving. Anyone against renaming those to Performance mode and Power saving mode? While low latency doesn't necessarily equals higher performance, I think in the PC setting, it's easier to understand by the end user. Regards, Humdinger ...
Categories: Development

[haiku-development] Re: hpkgs and compression (Jonathan Schleifer)

Mon, 2014-03-24 12:45
Am 11.03.2014 um 10:35 schrieb Axel Dörfler axeld@xxxxxxxxxxxxxxxx: Have you actually tested the performance penalty on old hardware? It doesn't matter much if you cannot read from packages 300 MB/s on modern hardware either; application data usually isn't large to begin with, and is afterwards cached in RAM. I haven't tested it, but I remember that even not-too-old-run-haiku hardware only achieved below 1 MB/s for Deflate with zlib. I guess we don't need to debate whether a few 100 KB/s is fast enough, right? :) Because that would be ...
Categories: Development

[haiku-development] Re: hpkgs and compression (Axel Dörfler)

Sun, 2014-03-23 23:45
On March 10, 2014 at 6:35 PM Jonathan Schleifer js-haiku-development@xxxxxxxxxxx wrote: Any opinions on this? This would give us the best of both worlds: * We would have small packages to transfer and save a *lot* of bandwidth. * There would be no extra Decompressing package step. * We would not have a performance penalty. Have you actually tested the performance penalty on old hardware? It doesn't matter much if you cannot read from packages 300 MB/s on modern ...
Categories: Development

[haiku-development] Re: hpkgs and compression (chase rayfield)

Sun, 2014-03-23 23:45
I'm just going to throw this idea out there... but Haiku servers really could use some relief especially during public releases. For hpkg files on the server no holds bared maximum compression (with the exception decompression ram requirements should be reasonable) would help a lot I imagine. But its also unacceptable to use hpkgs compressed that way in a running system. So, recompress on installation. Pick a format that has high compression/decompression like LZ4 or Snappy. Hpkgs could be recompressed with LZ4 HC since they are read only. ...
Categories: Development

[haiku-development] Re: hpkgs and compression (David Given)

Sun, 2014-03-23 03:45
On 3/10/14, 5:35 PM, Jonathan Schleifer wrote: [...] * An old processor gives a lower throughput in MB/s when using Deflate than just reading the file form the hard disk. Is there a block cache in between the hpkg virtual file system and the user? If so, then this will amelioriate a lot of the performance problems. (If not, there should be...) In fact, given the kind access patterns that this kind of application ...
Categories: Development

[haiku-development] Re: hpkgs and compression (Fredrik Holmqvist)

Sun, 2014-03-23 03:45
2014-03-10 19:21 GMT+01:00 Jonathan Schleifer js-haiku-development@xxxxxxxxxxx: Sure, but does aiming for good performance need we need to effectively make Haiku unusable on old machines? It would be nice if all machines that could run BeOS R5 could run Haiku just as well as R5 ;). Not if you pick a good solution, but the old hardware doesn't have that much of a future. So it is a dead end to let them take top priority. ...
Categories: Development

[haiku-development] Re: hpkgs and compression (Jonathan Schleifer)

Sun, 2014-03-23 03:45
Am 10.03.2014 um 19:04 schrieb Fredrik Holmqvist fredrik.holmqvist@xxxxxxxxx: I think it is worth evaluating. Also I'd aim for the best solution for modern architecture, not for older ones. Sure, but does aiming for good performance need we need to effectively make Haiku unusable on old machines? It would be nice if all machines that could run BeOS R5 could run Haiku just as well as R5 ;). If checksums are needed I'd also take a look at using the SSE4.2 CRC ...
Categories: Development

[haiku-development] Re: hpkgs and compression (Fredrik Holmqvist)

Sat, 2014-03-22 23:45
2014-03-10 18:35 GMT+01:00 Jonathan Schleifer js-haiku-development@xxxxxxxxxxx: * Store the data uncompressed in the hpkg itself and block-align it. * Compress the whole hpkg on the repository server (e.g. using xz or bzip2). This also means that similar files can be compressed much better. * Decompress the hpkg on the fly when the package is downloaded. This will reduce the time to install a package to half the time in many cases! While you download, the CPU is idle now, so it could decompress the data already received instead of doing nothing. As only about half the data needs to be transfered, this is a big speedup in package installation! ...
Categories: Development

[haiku-development] hpkgs and compression (Jonathan Schleifer)

Sat, 2014-03-22 23:45
Hi! As we all know, hpkgs are compressed using Deflate - a compression algorithm quite old by todays standard and not very efficient by todays standard. This causes many problems: * It is not possible to further compress an hpkg file, as compressing the output of Deflate with e.g. LZMA does not gain much. Compression works best if you pick the most efficient algorithm and let it work on the uncompressed data. The entropy of an compressed file is already increased and compression works ...
Categories: Development

[haiku-development] Re: BUrl in the kernel? (Jérôme Duval)

Sat, 2014-03-22 21:45
2014-03-09 20:25 GMT+01:00 Ingo Weinhold ingo_weinhold@xxxxxx: On 03/09/2014 01:31 PM, Bruno Albuquerque wrote: I need to parse a URL in the kernel (more specifically, in a file system add-on). BUrl does everything I need but it is normally part of the Support Kit. Is there any trick to use it that does not involve actually copying its code (or the relevant parts of its code) to the add-on itself? ...
Categories: Development

[haiku-development] Re: BUrl in the kernel? (Ingo Weinhold)

Sat, 2014-03-22 21:45
On 03/09/2014 01:31 PM, Bruno Albuquerque wrote: I need to parse a URL in the kernel (more specifically, in a file system add-on). BUrl does everything I need but it is normally part of the Support Kit. Is there any trick to use it that does not involve actually copying its code (or the relevant parts of its code) to the add-on itself? ...
Categories: Development

[haiku-development] BUrl in the kernel? (Bruno Albuquerque)

Sat, 2014-03-22 19:45
I need to parse a URL in the kernel (more specifically, in a file system add-on). BUrl does everything I need but it is normally part of the Support Kit. Is there any trick to use it that does not involve actually copying its code (or the relevant parts of its code) to the add-on itself? Thanks in advance. ...
Categories: Development

[haiku-development] Re: Profiler question (Ingo Weinhold)

Sat, 2014-03-22 19:45
On February 27, 2014 12:17:48 AM CET, Julian Harnath julian.harnath@xxxxxxxxxxxxxx wrote: In the Profile all teams mode (-a switch), it only loads the symbol information from the kernel and all newly loaded images -- but not from the images of already running teams. This sounds like a regression in the system profiler kernel code. It is supposed to send respective events for the images that are already loaded when ...
Categories: Development

[haiku-development] Re: Profiler question (looncraz)

Sat, 2014-03-22 19:45
On 2/26/2014 15:17, Julian Harnath wrote: Should I commit that change or is there a reason behind the current behaviour? -- So long, jua ...
Categories: Development

[haiku-development] Profiler question (Julian Harnath)

Sat, 2014-03-22 17:45
Hello, while looking into a bug, I used our profiler for the first time, which is a really really neat tool I have to say. However, I had to make a little change to it and I wonder if the original behaviour is intentional: In the Profile all teams mode (-a switch), it only loads the symbol information from the kernel and all newly loaded images -- but not from the images of already running teams. Since I wanted to profile a thread ...
Categories: Development

[haiku-development] Re: GSoC (Urias McCullough)

Sat, 2014-03-22 17:45
Thanks for the reply. I can test my interests in this project. How should I start for this project? Any particular bug solving or directly working on the idea? We have some helpful information on our Students page: - Urias ...
Categories: Development

[haiku-development] Re: GSoC (Sean Collins)

Sat, 2014-03-22 07:45
Karan Poddar wrote: Hi Sean, Thanks for the reply. I can test my interests in this project. How should I start for this project? Any particular bug solving or directly working on the idea? ...
Categories: Development

[haiku-development] Re: GSoC (Karan Poddar)

Sat, 2014-03-22 07:45
Hi Sean, Thanks for the reply. I can test my interests in this project. How should I start for this project? Any particular bug solving or directly working on the idea? *Thank,Karan Poddar*200901059, IIIT-H +91-9030735037 ...
Categories: Development

[haiku-development] Re: GSoC (Sean Collins)

Thu, 2014-03-20 23:45
Karan Poddar wrote: Hi, , India. I wanted to take part in GSoC 2014 and was searching for projects which interests me. For the past year, I have been using Qt in my B. Tech project. I have worked on CUDA during my internship at CERN. ...
Categories: Development