Major performance hit with 1.10.845.3



  • Originally I posted this in another thread but then I realized that this probably deserves its own thread.

    I upgraded from Vivaldi version 1.10.838.7 to 1.10.845.3 recently and now it's incredibly slow. Vivaldi was quite CPU-hungry before but responsive nevertheless, but now it's really slow. Some pages take twice as long to load with the new version (just an estimate; I didn't measure this, but I can if it would help). On Windows 10 I'm on the same version and I didn't notice a regression there, although this machine is more powerful so I'm not sure if this is a Linux thing only.

    This is probably unrelated to the new issue, but I want to mention it anyway: what I also noticed, I can play videos from Youtube with QMPlay2 with much less CPU use than with Vivaldi, so the high CPU use is definitely not a problem with my laptop per se.



  • @Pezo Hi. It might help for potential "community troubleshooting" of your problem if you include your relevant specs, eg, cpu, ram, distro, DE...

    FYI, my V-SS [current version] continues to be lovely to use for me, same as SS's have been for a very long time [for me]. Maybe i'm just lucky... or maybe your issue is local to your installation rather than a generic V-SS-Linux fault.

    ..............................................................................................
    Tower & Lappy = Maui Linux 17.03 x64 Plasma 5.9.3.
    Vivaldi 1.10.845.3 (Official Build) dev (64-bit)

    ..............................................................................................



  • @Steffie Yeah you're right, I'm running Arch Linux (updated at least every few days) on a Dell XPS 13 9360.

    Some specs: Intel Core i5-7200U CPU @ 2.50GHz (dual core), 8GB RAM, XFCE Desktop.
    I haven't considered that package updates from Arch may have caused this, but since the older version runs fine that's unlikely.



  • @Steffie I've the same issue with vivaldi performance since 1.9.811.13-1 and posted on this forum with details on packages in:
    [https://forum.vivaldi.net/topic/16884/poor-performance-from-vivaldi-snapshot-1-9-811-13-1-to-1-10-845-3-1].

    It's worth noting that all the performance issues go away simply by reverting to vivaldi-snapshot 1.9.804.3-1, without any other changes to the system. But if you think the hardware specs are relevant:
    Intel Xeon 3690 (6 cores @ 3.47Ghz)
    24G of RAM
    ASUS GTX 1070 graphics card

    debian testing distribution up to date as of May 15, using these packages:
    chromium-codecs-ffmpeg-extra 58.0.3029.96-0ubuntu1.1352
    libavcodec-ffmpeg-extra56:i386 7:2.8.5-1+b1
    linux-image-4.7.0-1-686-pae 4.7.6-1
    nvidia-driver 375.39-1
    XFCE desktop



  • Generic reply to you both: i assume you've already tried all the "standard" troubleshooting stuff, such as:

    1. Toggling h/w accel from its current state. vivaldi://chrome/settings/search#hardware
    2. Testing with a clean profile
    3. Uninstalling all extensions [works hand in glove with #2].

    ?


  • Moderator

    So I've tested 1.10.845.3 10.10.850.1 and 1.10.850.1-1 (including email on the 850 versions - oops, did I give something away? - and, no, email is not about to be released) on an old Intel Core2 Quad, 2.8 GHz, Intel onboard graphics, 8GB DDR3 RAM, dual-boot machine. Tested it in Win10 Creators Update 64-bit and completely updated Linux Mint 18.1 Serena/Cinnamon 64-bit.

    If anything, performance is at least as good and both CPU +-10% per core) and RAM usage (+-46%) are less than the last time I tested that machine on these platforms. (with a lot of tabs, it looks like RAM may be a little bit more, but CPU still less.

    I don't have Linux on my main tower here.



  • @Steffie I've just tested vivaldi-snapshot 1.10.845.3-1 on my computer, cleaning out the entire .config/vivaldi-snapshot directory with the exception of the three files below, so it looked like this before running:

        .config => ls -lR vivaldi-snapshot/
        vivaldi-snapshot/:
        total 4
        4 drwxr-xr-x 2 jack jack 4096 May 21 17:35 Default/
    
        vivaldi-snapshot/Default:
        total 1056
        688 -rw------- 1 j123 j123 701202 May 21 17:33 Bookmarks
        132 -rw------- 1 j123 j123 135168 May 21 17:33 Login Data
        104 -rw------- 1 j123 j123 105279 May 21 17:35 Preferences
    

    The performance remains absolutely pathetic: I'm entering this reply on the latest vivaldi snapshot and I can type an entire word before it shows up on the text box. I tried a video and the performance is poor, as described in my original post https://forum.vivaldi.net/topic/16884/poor-performance-from-vivaldi-snapshot-1-9-811-13-1-to-1-10-845-3-1

    [It does not make sense to upgrade if it means losing all my saved passwords, bookmarks and preferences, which is why I kept those 3 files.]

    Below is the result of top, which shows very low CPU utilization from this version of vivaldi (notice I've also raised vivaldi's CPU scheduling priority by setting its nice value to -5, from 0):

    top - 17:54:58 up 8 days,  2:47,  1 user,  load average: 0.28, 0.95, 0.97
    Tasks: 315 total,   1 running, 313 sleeping,   0 stopped,   1 zombie
    %Cpu(s):  1.0 us,  0.4 sy,  0.0 ni, 91.5 id,  7.2 wa,  0.0 hi,  0.0 si,  0.0 st
    KiB Mem : 24915488 total, 15143832 free,  2344352 used,  7427304 buff/cache
    KiB Swap: 31145300 total, 31145220 free,       80 used. 20096176 avail Mem 
    
      PID USER      PR  NI    VIRT    RES    SHR S  %CPU %MEM     TIME+ COMMAND                                                                                     
    17402 j123      15  -5  561728 182300  91904 S   1.7  0.7   0:25.08 vivaldi-bin                                                                                 
     6829 j123      20   0  123404  37912  26736 S   1.1  0.2   1:51.22 xfce4-terminal                                                                              
    17309 j123      15  -5 1371660 169004 106652 S   1.1  0.7   1:04.63 vivaldi-bin                                                                                 
    17617 j123      20   0    9896   3832   3116 R   1.1  0.0   0:04.86 top                                                                                         
    17984 j123      15  -5  900280 412492 298448 S   1.1  1.7   0:34.46 vivaldi-bin                                                                                 
    15723 root      20   0       0      0      0 S   0.6  0.0   0:04.49 kworker/u32:1                                                                               
    17447 j123      15  -5  580264 247288 142884 S   0.6  1.0   1:11.93 vivaldi-bin
    

  • Moderator

    @JSJB Have you tried changing hardware acceleration settings or anything? I can't get it to be "pathetic" on any of the 9 machines/OS'es here.



  • @Ayespy said in Major performance hit with 1.10.845.3:

    @JSJB Have you tried changing hardware acceleration settings or anything? I can't get it to be "pathetic" on any of the 9 machines/OS'es here.

    I haven't changed any nvidia drivers and am not sure why this is even an issue because:

    1. I can run flightgear (flight simulator) perfectly with the current settings
    2. glxgears shows just over 23,000 fps performance with the current settings
    3. vivaldi-snapshot 1.9.804.3-1 works beautifully with the current settings

    I'm writing this reply with the 1.9.803.3-1 version, contrast the CPU consumption from top with the 1.10 version posted above:

    top - 18:10:41 up 8 days,  3:02,  1 user,  load average: 0.58, 1.02, 1.01
    Tasks: 313 total,   1 running, 311 sleeping,   0 stopped,   1 zombie
    %Cpu(s):  8.7 us,  1.5 sy,  0.0 ni, 76.2 id, 13.5 wa,  0.0 hi,  0.1 si,  0.0 st
    KiB Mem : 24915488 total, 14533400 free,  2873488 used,  7508600 buff/cache
    KiB Swap: 31145300 total, 31145220 free,       80 used. 19831256 avail Mem 
    
      PID USER      PR  NI    VIRT    RES    SHR S  **%CPU** %MEM     TIME+ COMMAND                                                                                     
    20849 j123      15  -5  929812 653160  33344 S  **28.7**  2.6   1:21.33 vivaldi-bin                                                                                 
    20543 j123      15  -5  640040 239928  78424 S  **21.1**  1.0   1:10.97 vivaldi-bin                                                                                 
    20443 j123      15  -5 1082684 182180 111672 S  **11.5**  0.7   0:47.28 vivaldi-bin                                                                                 
     6824 j123      20   0 1132864 134936  84516 S   2.9  0.5  10:54.00 spotify                                                                                     
     6665 j123      12  -8 2192228  14008  11520 S   1.9  0.1 266:06.51 pulseaudio                                                                                  
     7450 j123      20   0  509152 199580 105596 S   1.9  0.8   2:49.96 spotify                                                                                     
    20583 j123      15  -5  554272 207756 139428 S   1.4  0.8   0:24.73 vivaldi-bin                                                                                 
     6829 j123      20   0  122372  37456  26736 S   1.0  0.2   1:54.28 xfce4-terminal
    

    The 1.9.804.3 version of vivaldi-snapshot uses up to 10 times the CPU as the newer one; clearly this has to be related to its better performance.


  • Moderator

    @JSJB Sometimes it's not OS+Hardware or Broswser+Hardware or Browser+OS, but Browser+OS+Hardware. It is, I fear, a fallacy to say that because this works, that should work.

    The very fact that the browser seems perhaps unable to fully access the hardware on your OS is, to me, a clear indication of Browser+OS+Hardware. This is going to be quite difficult for the developers to reproduce, as they don't have you machine on a bench - and a wide variety of Linux users on .deb-based systems are having no trouble.



  • @Ayespy said in Major performance hit with 1.10.845.3:

    @JSJB Sometimes it's not OS+Hardware or Broswser+Hardware or Browser+OS, but Browser+OS+Hardware. It is, I fear, a fallacy to say that because this works, that should work.

    seems too vague, can you be more specific (I'm an experienced networking software developer in both unix/linux and embedded)?

    The very fact that the browser seems perhaps unable to fully access the hardware on your OS is, to me,

    What evidence do you have that the browser is not accessing the hardware on my OS?

    a clear indication of Browser+OS+Hardware. This is going to be quite difficult for the developers to reproduce,

    Should be pretty easy as I'm assuming that, as usual, there's an OS abstraction layer and an OS specific layer, and further, that the linux OS specific layer doesn't vary by linux distribution. A diff on the version control system between 1.9.804.3-1 and 1.9.818.25-1 (the first version where I saw this issue) should show the code changes between the versions and make it easier to see which could have been performance-affecting. I've done similar comparisons dozens of times.

    as they don't have you machine > on a bench - and a wide variety of Linux users on .deb-based systems are having no trouble.

    Can you post the utilization in top from some other linux systems to compare?



  • @Steffie said in Major performance hit with 1.10.845.3:

    Toggling h/w accel from its current state. vivaldi://chrome/settings/search#hardware

    You still should try this, not merely opine that it won't help or should not be necessary.

    I am not doubting the sincerity of your problem, only that it is absolutely not generic across all or even most Linux V users, as best i can judge from these fora.


  • Moderator

    @JSJB I surmise from the fact that it is performing poorly and under-utilizing your resources, that it cannot, for some reason, use them. If it could, it would and it would perform better I think.



  • @Steffie said in Major performance hit with 1.10.845.3:

    @Steffie said in Major performance hit with 1.10.845.3:

    Toggling h/w accel from its current state. vivaldi://chrome/settings/search#hardware

    The setting has checked "Use hardware acceleration when available" already and shows poor performance.

    More practically, I ran strace on a running instance of 1.10.845.3-1 and it seems there are lots of futex calls (fast user-space locking), which would seems a very plausible explanation why the new version of vivaldi is not consuming too much CPU but also not performing well.
    From the futex man page:

    FUTEX(7)                                                          Linux Programmer's Manual                                                         FUTEX(7)
    
    NAME
           futex - fast user-space locking
    
    SYNOPSIS
           #include <linux/futex.h>
    
    DESCRIPTION
           The Linux kernel provides futexes ("Fast user-space mutexes") as a building block for fast user-space locking and semaphores.  Futexes are very basic
           and lend themselves well for building higher-level locking abstractions such as mutexes, condition variables, read-write locks, barriers,  and  sema‐
           phores.
    
           **Most  programmers will in fact not be using futexes directly but will instead rely on system libraries built on them, such as the Native POSIX Thread
           Library (NPTL) (see pthreads(7))**.
    

    You still should try this, not merely opine that it won't help or should not be necessary.

    It didn't help and was always set.



  • @Ayespy said in Major performance hit with 1.10.845.3:

    @JSJB I surmise from the fact that it is performing poorly and under-utilizing your resources, that it cannot, for some reason, use them. If it could, it would and it would perform better I think.

    That's an incorrect assumption: it can also underperform because different threads spawned by the vivaldi process are paused waiting for a semaphore, for example (as the futex in the earlier post). Then, even though they have no problem accessing resources, different threads are locking each other out, causing them to wait excessively. That is typically shown by the combination of low CPU utilization and low performance and happens with cooperating threads.

    See the portion of the futex manual posted earlier: it seems to depend on the pthreads library used, which could explain why different versions of linux experience different performance with the same vivaldi-snapshot_1.10.845.3

    My versions of libpthread come from libc6_2.24-5. What version of pthreads are you using?


  • Moderator

    @JSJB I wouldn't know. I just update per the content of the Mint 18.1 repos.



  • @Ayespy can you post the results of 'apt-cache policy libc6' on your system? That will give the exact version of this library.

    Also, the vivaldi-snapshot script checks for these ffmpeg libraries, in this order:

    VIVALDI_FFMPEG_FOUND=NO
    checkffmpeg "/usr/lib/$DEBARCH/oxide-qt/libffmpeg.so" '/usr/share/doc/oxideqt-codecs-extra'
    checkffmpeg '/usr/lib/chromium-browser/libffmpeg.so' '/usr/share/doc/chromium-codecs-ffmpeg-extra'
    
    \# Check if the user has manually placed suitable libs.
    checkffmpeg "$HOME/.local/lib/vivaldi/libffmpeg.so"                   <<<<< same as /usr/lib/chromium-browser/libffmpeg.so , see below
    checkffmpeg "$HERE/libffmpeg.so"
    

    Can you check what versions of libffmpeg you're using? These are the possible ones on my system:
    vivaldi-snapshot: /opt/vivaldi-snapshot/lib/libffmpeg.so
    chromium-codecs-ffmpeg-extra: /usr/lib/chromium-browser/libffmpeg.so <<<<< used by vivaldi-snapshot and needed for netflix/amazon video
    oxideqt-codecs-extra:i386: /usr/lib/i386-linux-gnu/oxide-qt/libffmpeg.so


  • Moderator

    @JSJB - Momentarily. I'll have to log out of Win10 on that machine and boot back in to Mint. And I'll have to change the wireless dongle as .deb Linux has "updated" drivers to where my good one no longer works on it.


  • Moderator

    @JSJB Mine is 2.23-0ubuntu7


  • Moderator

    @JSJB In the Vivaldi folder, I only have libffmpeg.so. I also have usr/lib/chromium-browser/libffmpeg.so and usr/lib/x86_64-linux-gnu/oxide-qt/libffmpeg.so


Log in to reply
 

Looks like your connection to Vivaldi Forum was lost, please wait while we try to reconnect.