pkgsrc-Users archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: Windows port?




Here's one more data point for this thread and then I'll shut up...

I timed the pkgsrc bootstrap under Cygwin at 71 minutes, 37 seconds, minus my response times.

I'd be interested in seeing some data showing WSL performing better than Cygwin, as I've seen a couple claims to this effect.

From my perspective, the choice between them would come down to factors other than performance, as they both look pretty abysmal for system-call intensive tasks.

Cheers,

    JB

On 05/27/18 12:58, Jason Bacon wrote:

I reran the bookstrap under "time".  Subtract a little from the totals to account for my response times to answer questions posed by auto-pkgsrc-setup.  Also note that auto-pkgsrc-setup times include creating a boostrap kit with both the source tree and installation directories.

WSL: ~104 minutes total, 16 minutes just to unpack.

FreeBSD running under HyperV: 21:47 total

Unpacking was quite slow under HyperV as well, but GNU configured scripts whizzed along at about native speed, on the order of a dozen checks / sec vs  ~1 / sec under WSL.  Compiles were much faster in the VM than in WSL as well.

I would hypothesize that WSL has a fork() bottleneck similar to Cygwin.

I suspect your much bigger RAM pool might explain some of the better performance on your end.

On 05/27/18 09:34, Chavdar Ivanov wrote:
It is an Intel Core i7-6700HQ@2.6GHz, 16GB memory, SanDisk SD7SN6S-512GB SSD(SATA), running build 17134.81 (v. 1803, a.k.a. the April update). Perhaps my expectations were not very high, I didn't measure the time taken precisely enough, but it seemed the extraction took most of the time. I have since tried to run some benchmarks; couldn't wait for filebench to complete (maybe I've selected the wrong configuration, but I suspect it was worse than that), bonnie++ ran though, with a result which seemed to me the reverse to what is in that discussion - it completed much faster when the work directory was /mnt/c/Temp than when it was just /tmp. The resulting figures  were not conclusive - some of the metrics favoured /tmp, some - /mnt/c/Temp, e.g. random file deletion was much faster when crossing the /mnt boundary.

The discussion in that thread was quite illuminating with respect to the problems Microsoft developers have with WSL, and, to be honest, it would be harsh to expect much more. It, after all, was thought of as a means of keeping developers stay with Windows while using a number of open source technologies, not for actual deployment in the wild.

Chavdar

On Sun, 27 May 2018 at 15:02 Jason Bacon <outpaddling%yahoo.com@localhost <mailto:outpaddling%yahoo.com@localhost>> wrote:


    What are your hardware specs?  I suspect you have both a faster
    machine
    and more patience than I do.

    I agree that WSL and Cygwin are fast enough for many purposes.
    wouldn't want to build gcc under WSL on the machine I have, though.

    For comparison, I ran auto-pkgsrc-setup on 3 identical machines
    running
    different Unix OSs.  These are old PowerEdge 1920 servers, 4-core
    Xeon
    5160 3.0GHz, 16G RAM.

    CentOS MD-RAID mirror      6:16
    FreeBSD Root-on-ZFS          5:57
    NetBSD PERC RAID              12:56 (I'd like to know what the
    bottleneck was here, I'd expect NetBSD to be on par with the others)

    zile build on NetBSD            7:10
    ....





Home | Main Index | Thread Index | Old Index