Welcome!

Please read this important information about MX sources lists.
News
  • MX Linux on social media: here
  • Mepis support still here
Current releases
  • MX-16.1 release info here
  • antiX-17-b1-full release info here
  • antiX-16.2 release info here
New users
  • Please read this first, and don't forget to add system and hardware information to posts!
  • Read Forum Rules

MX-16 issue: Super low performance and CPU governor / frequency issues on DELL 3150 (Silvermont / Bay Trail-M)

Message
Author
User avatar
MX-16_fan
Forum Regular
Forum Regular
Posts: 456
Joined: Mon Feb 13, 2017 12:09 pm

MX-16 issue: Super low performance and CPU governor / frequency issues on DELL 3150 (Silvermont / Bay Trail-M)

#1 Postby MX-16_fan » Fri May 19, 2017 11:10 am

@all:

As reported on other occasions, I'm currently testing an MX-16 system installed on a 128 GB USB in combination with a DELL Latitude 3150 notebook (Latitude 11 3150-4517), aiming at creating a system on USB stick which behaves just like any standard installation, the only difference being that I can carry it around and use it on other machines as well.

Since I never totally understood the "persistence" concept (in spite of heaving read the manual and having watched dolphin_oracle's video several times :happy:), I just formatted the USB stick with btrfs and installed MX-16 on it the classic way.

While I can say that this works o.k. in terms of function, it is catastrophic as far as speed is regarded. The system very often seems to have the hiccups, meaning that e.g. you open a new Firefox tab, and then the system kind of freezes, and you have to wait five seconds or more until the tab is created or loaded. In the meantime, wmhdplop sometimes indicates there's extremely high activity on the USB stick.

My first thought was that this stems from the fact that the USB stick, which is a Silicon Power "Jewel J80", is not extremely fast (it has a high capacity while being pretty small). Nevertheless, IMHO it should be perfectly sufficient - I read some customer feedback stating that it gets you a writing speed of 16 to 22MB/s and a reading speed up to 1.5GB/s. Plus, the machine does have a USB 3.0 or 3.1 port, and the processor isn't all too slow, so I would guess it should run better than that.

While investigating this for quite some time, I found a number of hints on the internet that this could be an issue related to the platform, which is Silvermont / Bay Trail-M. I found this here in Wikipedia: https://en.wikipedia.org/wiki/Silvermont#Erratum and this article here: http://www.theregister.co.uk/2017/02/06 ... ulty_chip/ about bugs in some processors. However, I'm not quite sure whether this relates exactly to the chip I have. I asked lshw-gtk for the CPU's SKU but it doesn't know it.

As you can see here: https://en.wikipedia.org/wiki/List_of_I ... Pentiums_2, the CPU's maximum standard speed is 2.16 Ghz, plus it features 2.66 GHz in CPU turbo mode (burst mode).

Initially wmhdplop told me that the governor used was "powersave", so I changed that to "performance" using the instructions given in this section of the Debian wiki: https://wiki.debian.org/HowTo/CpuFreque ... figuration. (I thought I'd done that before - maybe some update wiped that out).

However, this didn't lead to any remarkable increase in speed but to some strange side-effect:

* The "performance" configuration got lost after installing and/or calling lshw-gtk. According to wmhdplop, it fell back to "powersave".

* Before that, wmhdplop showed me weird different frequencies for the four different CPUs, something like "2.16 Ghz" on CPU0, while "2.17 Ghz" on CPU 3.

According to wmhdplop, the driver is intel_pstate. I have the MX-16 microcode correction package installed. I have switched off C-States in the UEFI.

Any idea what I can do about this?

I noticed that the Arch Linux guys have special optimizations for Silvermont (https://wiki.archlinux.org/index.php/re ... d_packages) and BayTrail (https://aur.archlinux.org/packages/linu ... 0-headers/), but I couldn't find anything the like in any of MX-16's repos.

AFAIK the notebook has been designed aiming at the university market, so I guess there's quite a number of them around.


Greetings, and a nice weekend to all of you :happy: , Joe

User avatar
entropyfoe
Forum Regular
Forum Regular
Posts: 595
Joined: Thu Apr 19, 2007 11:42 am

Re: MX-16 issue: Super low performance and CPU governor / frequency issues on DELL 3150 (Silvermont / Bay Trail-M)

#2 Postby entropyfoe » Fri May 19, 2017 11:42 am

Those write speeds sound pretty bad, and the read speed sounds unbeliveably good, comparable to Samsung 960 nVM PCI 3.0 x 4 speeds !

gets you a writing speed of 16 to 22MB/s and a reading speed up to 1.5GB/s


That write speed is like 5X slower than a 7200 RPM hard drive, so might be a problem.
Asus Prime 370X-Pro
AMD Ryzen 1600X (6 cores @ 3.6 GHz)
16 Gig DDR4 3200 (G Skill)
Nvidia -MSI GeForce GT 710 fanless
Samsung 960 NVMe SSD nvme0n1 P1,P2, and P3=MX-16, P4=antiX-17 beta 2
2TB WD sda1= data
on-board ethernet & sound

User avatar
dolphin_oracle
Forum Veteran
Forum Veteran
Posts: 6350
Joined: Sun Dec 16, 2007 1:17 pm

Re: MX-16 issue: Super low performance and CPU governor / frequency issues on DELL 3150 (Silvermont / Bay Trail-M)

#3 Postby dolphin_oracle » Fri May 19, 2017 12:13 pm

MX-16_fan wrote:@all:

While I can say that this works o.k. in terms of function, it is catastrophic as far as speed is regarded. The system very often seems to have the hiccups, meaning that e.g. you open a new Firefox tab, and then the system kind of freezes, and you have to wait five seconds or more until the tab is created or loaded. In the meantime, wmhdplop sometimes indicates there's extremely high activity on the USB stick.


could seriously be related to cacheing or something like that. as soon as any app writes to the usb, its going to slow to a crawl for a few seconds. I run into this all the time on my ram-strapped persistence setups because I use static persistence, which doesn't do the writes in ram like the regular persistence setups. that loading the persist files in ram is what makes the liveusb so speedy in comparison to a "installed" stick on a usb.

firefox does a lot of writes to disk.

also, if anything moves to swap, and swap is on the usb, you get a double hit.

those speeds you researched almost certainly are usb 3.0 speeds already, given that usb 2.0 maxes out at 480 mpbs or something like that.

powersave vs. performance won't do anything for write speeds either.
http://www.youtube.com/runwiththedolphin
eeepc 904ha antix 15 32 bit (streaming media player)
acer c720p chromebook - chromeos / MX15-64 (custom snapshot) live USB.
lenovo s21e - antiX-16 MX16(live-usb)

User avatar
MX-16_fan
Forum Regular
Forum Regular
Posts: 456
Joined: Mon Feb 13, 2017 12:09 pm

Re: MX-16 issue: Super low performance and CPU governor / frequency issues on DELL 3150 (Silvermont / Bay Trail-M)

#4 Postby MX-16_fan » Fri May 19, 2017 12:28 pm

@entropyfoe:

entropyfoe wrote:Those write speeds sound pretty bad, and the read speed sounds unbeliveably good, comparable to Samsung 960 nVM PCI 3.0 x 4 speeds !

gets you a writing speed of 16 to 22MB/s and a reading speed up to 1.5GB/s


That write speed is like 5X slower than a 7200 RPM hard drive, so might be a problem.


You're perfectly right, I have to correct that. Just checked it out myself using gnome-disk-utility. It tells me that the stick has an average reading speed of 213,7 MB/s and an average access time of 1,01 ms. I couldn't test the writing speed, however - gnome-disk-utility warns me that I should make a backup before performing any write test, and I'm not quite sure about how I would backup the stick as a whole.

In comments I found on Amazon etc. people report up to 30 MB/s write spead while not even using the advantages of btrfs (don't know if that makes any difference when it comes to benchmarks.

While I agree with you that that is not remarkable fast, this is pretty much what HDDs used to deliver six or seven years ago, so that's why I'd think that the main bottleneck is somewhere else.

I have to add that there's some additional strange behaviour that I forgot to report about. I rebooted the system before I did the test, and this time it was a lot faster (even though still far from normal) than before. I notice this every now and then on fresh boot. There's no obvious logic in it. (This time I didn't have the power plug attached but that was the only thing that was different.) IMHO this also indicates that MX-16 has some general issues with this Intel hardware platform.


Greetings, Joe

User avatar
MX-16_fan
Forum Regular
Forum Regular
Posts: 456
Joined: Mon Feb 13, 2017 12:09 pm

Re: MX-16 issue: Super low performance and CPU governor / frequency issues on DELL 3150 (Silvermont / Bay Trail-M)

#5 Postby MX-16_fan » Fri May 19, 2017 12:34 pm

@dolphin_oracle:

dolphin_oracle wrote:(...)
also, if anything moves to swap, and swap is on the usb, you get a double hit.

those speeds you researched almost certainly are usb 3.0 speeds already, given that usb 2.0 maxes out at 480 mpbs or something like that.


Fun fact is, I don't even have any swap partition in use (at least that's what xfce4-systemload-plugin tells me). When I installed the system, I chose "... or existing" for the swap partition, instead of exactly choosing it. Seems that the installer didn't find it all by itself. I can file that as a separate issue if you want, or maybe you want to put this into the bug tracker :happy:.


Greetings, Joe

User avatar
Jerry3904
Forum Veteran
Forum Veteran
Posts: 19215
Joined: Wed Jul 19, 2006 6:13 am

Re: MX-16 issue: Super low performance and CPU governor / frequency issues on DELL 3150 (Silvermont / Bay Trail-M)

#6 Postby Jerry3904 » Fri May 19, 2017 12:42 pm

Does btrfs and its Copy On Write feature play a role here?
Production: 4.7.0-0.bpo.1-amd64, MX-15 RC1, AMD FX-4130 Quad-Core, GeForce GT 630/PCIe/SSE2, 8 GB, Kingston SSD 120 GB and WesternDigital 1TB
Testing: AAO 722: 3.16-0-4-686-pae. MX-15, AMD C-60 APU, 4 GB

skidoo
Forum Regular
Forum Regular
Posts: 573
Joined: Tue Sep 22, 2015 6:56 pm

Re: MX-16 issue: Super low performance and CPU governor / frequency issues on DELL 3150 (Silvermont / Bay Trail-M)

#7 Postby skidoo » Fri May 19, 2017 12:46 pm

if anything moves to swap, and swap is on the usb, you get a double hit.
other factors that most would-be micromanaging tweakers neglect to consider when they post:
-- messing with zram
-- setting swappiness to 90 (OMG, asinoro posted that as a "tip")
-- writing to savefile residing on a FAT32 -formatted LiveUSB

Joe, in another topic we've already done "the cpufreq dance" and I reported that I found ZERO benefit from (installing extras and) tweaking the defaults. On that subject, I have nothing further to add here except to point out "c'mon... cpufreq cain't hardly be the limiting factor related to slow pendrive write operations"

firefox does a lot of writes to disk.
Yep. Lookit here
skidoo wrote: http://forum.mxlinux.org/viewtopic.php?f=108&t=40822&p=396827&hilit=cache+firefox

to avoid the continual disk i/o, all you need to do is change a firefox preference, instructing the browser to write to tmpfs

prefkey:
browser.cache.disk.parent_directory
value:
/tmp/happyfox

note: No need to manually mkdir whatever subdir under /tmp. As needed, firefox will create it.

skidoo
Forum Regular
Forum Regular
Posts: 573
Joined: Tue Sep 22, 2015 6:56 pm

Re: MX-16 issue: Super low performance and CPU governor / frequency issues on DELL 3150 (Silvermont / Bay Trail-M)

#8 Postby skidoo » Fri May 19, 2017 12:48 pm

Jerry3904 wrote:Does btrfs and its Copy On Write feature play a role here?
good point. Seeker might should A/B test, rule out whether that is a limiting factor.

User avatar
MX-16_fan
Forum Regular
Forum Regular
Posts: 456
Joined: Mon Feb 13, 2017 12:09 pm

Re: MX-16 issue: Super low performance and CPU governor / frequency issues on DELL 3150 (Silvermont / Bay Trail-M)

#9 Postby MX-16_fan » Fri May 19, 2017 12:52 pm

Jerry3904 wrote:Does btrfs and its Copy On Write feature play a role here?


I am extremely sorry, Jerry, I was wrong here again. Just checked it out. I had made use of btrfs on my first attempt but then switched to XFS because I thought that this here: viewtopic.php?f=97&t=41690 had something to do with my choice of btrfs.

I went for something different then ext4 because I had great trouble concerning data corruption on different media with ext4 (on a different machine, and not with MX-16).


Sorry again, greetings, Joe

User avatar
Stevo
Forum Veteran
Forum Veteran
Posts: 13412
Age: 59
Joined: Fri Dec 15, 2006 8:07 pm

Re: MX-16 issue: Super low performance and CPU governor / frequency issues on DELL 3150 (Silvermont / Bay Trail-M)

#10 Postby Stevo » Fri May 19, 2017 1:14 pm

I'm sure that the CPU power management doesn't have anything to do with it, but with p-state, all you get is powersave and performance, and there doesn't really seem to be any difference in performance or CPU temperature to my casual eye (i5-6200u CPU).

You can turn p-state off with a kernel boot cheat, "intel_pstate=disable", and then the kernel will fall back to cpufreq-- you can again have the fine-grained control over its frequencies. But again, I don't think that's the issue.

There's also a setting somewhere in Firefox/Chrome where you can adjust how often it backs up its state--the current (15 second?) setting ends up writing many GB to the drive over time, and that worries a lot of SSD owners.


Return to “Software / Configuration”

Who is online

Users browsing this forum: No registered users and 3 guests