ARMv6 (was: Re: Firefox Android: Improving Performance)
Henri Sivonen
> The other ones are Flash and "I want
> this on my device". The latter is a "good" problem to have (except for
> the fact that there are many ARMv6 devices out there).
Are there active plans to support and supply builds for ARMv6? It
seems that lower end Android phones are catching up with the RAM
requirements of Firefox but are staying on ARMv6 for now.
--
Henri Sivonen
hsiv...@iki.fi
http://hsivonen.iki.fi/
Doug Turner
Doug
> _______________________________________________
> dev-planning mailing list
> dev-pl...@lists.mozilla.org
> https://lists.mozilla.org/listinfo/dev-planning
Henri Sivonen
> Not to discourage you from attempting to get something up and running again, but we have been actively ignoring ARMv6. I'd hate to focus any energy on lower end processors until we are incredible on faster ones.
Are there any plans for a Firefox Home for ARMv6 then? If the
massmarket phones are unsupported for sync, it creates an incentive
for users of massmarket phones to switch to another desktop browser
whose sync counterpart runs on massmarket phones.
Doug Turner
Gervase Markham
> I don't really know anything about Firefox Home. Not even sure if
> this is the right mailing list for stuff like that…
Firefox Home does have a mailing list, but is currently non-public :-(
Stuart Parmenter tells me that his team will soon be announcing future
plans in this area, and development will move into the public eye.
Gerv
Robert Kaiser
> Are there any plans for a Firefox Home for ARMv6 then?
IIRC, the plan for Firefox Home is to be developed into a web app that
can run on a really huge set of devices and browsers - the first demo I
saw of that was showing it even on Symbian phones.
Robert Kaiser
--
Note that any statements of mine - no matter how passionate - are never
meant to be offensive but very often as food for thought or possible
arguments that we as a community should think about. And most of the
time, I even appreciate irony and fun! :)
Wes Garland
> Henri Sivonen schrieb:
>
>> Are there any plans for a Firefox Home for ARMv6 then?
>>
>
> IIRC, the plan for Firefox Home is to be developed into a web app
Okay, I know this probably isn't the place for this. But as a happy Firefox
Home user, this alarmed me. For about 16 nanoseconds. Then I realized how
awesome it would be.
Then I realized that with a carefully-crafted Proxy Auto Config file, some
custom proxy servers and a bunch of Firefox-sync wizardry, you'd be able to
share cookies between Firefox and your remote browsers. (Use the PAC to
proxy for a single to hit to new domain based on freshness instead of URL;
proxy server whips in a fake cookie header in both directions with data
scooped from Firefox)
*hmm*
(Not saying I've thought through the security issues, just thought it was an
interesting idea to throw out there)
Wes
--
Wesley W. Garland
Director, Product Development
PageMail, Inc.
+1 613 542 2787 x 102
Dietrich Ayala
clear on the rationale for focusing exclusively on high-end devices.
On Sun, Aug 28, 2011 at 11:57 PM, Doug Turner <do...@mozilla.com> wrote:
> Not to discourage you from attempting to get something up and running again, but we have been actively ignoring ARMv6. I'd hate to focus any energy on lower end processors until we are incredible on faster ones.
>
> Doug
>
> On Aug 28, 2011, at 11:53 PM, Henri Sivonen wrote:
>
>> On Fri, Aug 26, 2011 at 10:19 PM, Thomas Arend <tarend...@gmail.com> wrote:
>>> The other ones are Flash and "I want
>>> this on my device". The latter is a "good" problem to have (except for
>>> the fact that there are many ARMv6 devices out there).
>>
>> Are there active plans to support and supply builds for ARMv6? It
>> seems that lower end Android phones are catching up with the RAM
>> requirements of Firefox but are staying on ARMv6 for now.
>>
Martijn
> Why would you hate to focus energy on lower end processors? I'm not
> clear on the rationale for focusing exclusively on high-end devices.
I assume it takes a lot of time and effort to fix everything that
would be related to ARMv6 and lower-end devices and by the time we
would have something that is working well enough, virtually nobody
would have those devices anymore.
Regards,
Martijn
--
Martijn Wargers - Help Mozilla!
http://quality.mozilla.org/
http://wiki.mozilla.org/Mozilla_QA_Community
irc://irc.mozilla.org/qa - /nick mw22
Dietrich Ayala
Is the problem that ARMv6 compat requires a massive retooling? Or just
that the reasonable perf targets are so far off that it's not worth
attempting? Both? Other?
As people have brought up in this and other threads, new ARMv6 devices
are still shipping at this time, and doing very well in some areas of
the world.
Mark Finkle
for ARMv6. I don't know what state those patches are in currently. I
don't see any harm in getting Mike's work reviewed and landed if it
allows us to run on some ARMv6 laptops, netbooks, devices or emulators.
The bigger chunk of work will probably be getting to an acceptable level
of performance. Something we are not even at for "high-end" devices.
IIRC, we would need to disable JIT for ARMv6 as well.
However, if doing so increases our reach and gets more people involved
in contributing to the project, it would be worth it.
Matt Brubeck
> Is the problem that ARMv6 compat requires a massive retooling? Or just
> that the reasonable perf targets are so far off that it's not worth
> attempting? Both? Other?
Mostly the latter. We can (and previously did) produce builds of Fennec
for ARMv6 processors. At the moment it will not actually run on ARMv6
because of crashing bugs, but that's just because no one is building and
testing it there. If we worked on it, we could fix the bugs and keep it
running.
But our concern is that it would not be good enough. It would take even
more disk space than current builds (because ARMv6 doesn't include the
more-compact Thumb-2 instruction set). This is a big concern on low-end
devices which typically have very little storage. Even the newest ARMv6
device I could find - the Samsung Galaxy Y, which was announced this
month and isn't even shipping yet - has only 160MB of onboard storage,
and is unlikely to ship with a large SD card.
Startup is noticeably worse on lower-end devices because they have
slower CPUs and storage. Fennec still takes almost 10 seconds to start
on the N900, an older ARMv7 device that is still faster than current
ARMv6 Android phones. And some of our options for reducing footprint
make startup speed even worse.
Nearly all ARMv6 phones in the wild have 128 to 256 MB of RAM. The ones
with less than 256 MB have never been able to run Fennec at all. The
ones with 256 MB can run Fennec, but it crashes frequently and forces
other apps to quit. It also gets evicted from memory more often, so the
user has to restart the browser frequently - making slow startup even worse.
A very large percentage of ARMv6 devices are stuck on Android 2.1 or
1.6. We can't support 1.6 at all because it doesn't have sufficient
native code support. We are hoping to drop Android 2.1 support soon
because it is becoming a maintenance burden.
In a 6- or 12-month time frame with some dedicated engineering
resources, I'm sure we could fix bugs, reduce our footprint, and improve
our startup to the point where the browser is more or less usable on a
subset of low-end devices. But it might be only a small subset, for the
reasons above, making the reward smaller than you might have estimated.
And at that point, when we've disabled enough features to get
acceptable footprint and performance, what do we have to offer? After
we make it usable, how much more work would we need to make it
competitive? Why would these users download and install Firefox when
their devices already have a decent, basic (but fast) browser built in?
Personally I think that a Firefox Home app or other Firefox Sync
integration would be a more practical way to help our users who have
low-end mobile devices and gain some influence in that part of the
market, even though it doesn't give us the same leverage to shape the
web as a full browser that we control.
A year ago we ran on only a handful of high-end devices. Now, dozens of
mid-range devices are coming out all the time that are just as powerful
as last year's high end, and we are ready to support them all
immediately. Next year we'll start seeing low-end devices that we can
easily support. ARMv6 won't die any time soon -- it'll get cheaper
still, and migrate to even lower-end markets currently served by feature
phones -- but I think newer hardware is where the long-term growth is,
and our engineering efforts there will pay larger dividends for a longer
time. We have plenty of work left to do just to get good performance
even on modern ARMv7 phones.
Dietrich Ayala
Mike Hommey
> At one time, Mike Hommey was keeping a set of patches to allow
> building for ARMv6. I don't know what state those patches are in
> currently. I don't see any harm in getting Mike's work reviewed and
> landed if it allows us to run on some ARMv6 laptops, netbooks,
> devices or emulators.
>
> The bigger chunk of work will probably be getting to an acceptable
> level of performance. Something we are not even at for "high-end"
> devices. IIRC, we would need to disable JIT for ARMv6 as well.
>
> However, if doing so increases our reach and gets more people
> involved in contributing to the project, it would be worth it.
I have patches for ARMv4T, not ARMv6. And iirc all of them landed on m-c
a little while ago. And to support ARMv4T, I need to disable methodjit.
I know David Baron was trying to build for ARMv6 a couple weeks ago and
had some problems with it, I don't know how further he went.
Mike
Henri Sivonen
> But our concern is that it would not be good enough. It would take even
> more disk space than current builds (because ARMv6 doesn't include the
> more-compact Thumb-2 instruction set). This is a big concern on low-end
> devices which typically have very little storage. Even the newest ARMv6
> device I could find - the Samsung Galaxy Y, which was announced this month
> and isn't even shipping yet - has only 160MB of onboard storage, and is
> unlikely to ship with a large SD card.
It doesn't really need to *ship* with a large SD card as long as it
supports a large SD card. Or maybe that depends on what kind of user
you're thinking about: users trying out stuff without planning or
users who really want Firefox specifically (but who don't want it
enough to pay for an ARMv7 device or don't want Firefox enough to
sacrifice the Wildfire / Galaxy Mini form factor). That latter kind of
user could easily buy a large-enough SD card.
> Nearly all ARMv6 phones in the wild have 128 to 256 MB of RAM. The ones
> with less than 256 MB have never been able to run Fennec at all. The ones
> with 256 MB can run Fennec, but it crashes frequently and forces other apps
> to quit. It also gets evicted from memory more often, so the user has to
> restart the browser frequently - making slow startup even worse.
Has e10s increased memory usage significantly? Does Android use more
RAM for the system than Maemo? Until July 2010, Fennec ran on N800
which has 128 MB of RAM (and ARMv6). (And sure, Opera Mobile runs much
much better on the device, but Fennec still worked.)
What caught my attention about how phones match with Firefox systems
requirements is that HTC Wildfire S has 512 MB of RAM but ARMv6.
> A very large percentage of ARMv6 devices are stuck on Android 2.1 or 1.6.
> We can't support 1.6 at all because it doesn't have sufficient native code
> support. We are hoping to drop Android 2.1 support soon because it is
> becoming a maintenance burden.
Does Mozilla have any kind of user-facing promise or plan about
retaining OS support?
Maemo 4 got dropped without any discussion or warning whatsoever one
day Fennec just got autoupdated to a build that didn't run anymore.
Maemo 5 is getting dropped while the corresponding device is still in
principle being part of an active product line and being sold (or more
to the point, it's been only about a year since the product was still
being bought). On the Android side support started with Android 2.0,
but 2.0 has already gotten dropped silently (or at least I never saw
any announcement on the main Mozilla blog telling Android 2.0 users
what was coming). Now 2.1 is getting dropped. Meanwhile, the OS
upgrade story on the devices isn't that great but the devices aren't
that old, either.
I fully realize that from the developer point of view, it doesn't make
sense to put effort into platforms with few users now and no future,
but from the user point of view, it really sucks to drop OS support
when the user has gotten in the habit of a browsing experience that
includes Sync. I'm worried that dropping platforms the way they have
been and are being dropped is going to create user unhappiness. (OTOH,
I don't currently have a handheld device capable of running the latest
Firefox, because I'm not convinced that the current crop of Android
devices is going to get OS updates to 4.0 and beyond and I'm not
convinced that Firefox is past the point of dropping support for old
OS versions early.)
> Why would these users download and install
> Firefox when their devices already have a decent, basic (but fast) browser
> built in?
To get the full Sync experience, including passwords and syncing the
phone state back to desktop.
> Personally I think that a Firefox Home app or other Firefox Sync integration
> would be a more practical way to help our users who have low-end mobile
> devices and gain some influence in that part of the market, even though it
> doesn't give us the same leverage to shape the web as a full browser that we
> control.
Realistically, I think we can't really "gain influence" within parts
of the market that we don't ship Gecko to.
> A year ago we ran on only a handful of high-end devices. Now, dozens of
> mid-range devices are coming out all the time that are just as powerful as
> last year's high end, and we are ready to support them all immediately.
> Next year we'll start seeing low-end devices that we can easily support.
> ARMv6 won't die any time soon -- it'll get cheaper still, and migrate to
> even lower-end markets currently served by feature phones -- but I think
> newer hardware is where the long-term growth is, and our engineering efforts
> there will pay larger dividends for a longer time. We have plenty of work
> left to do just to get good performance even on modern ARMv7 phones.
Part of the problem is that it's really hard to manage user
expectations when you can't tell if Firefox for Android is going to
run by looking at the specs that manufacturers and retailers give
about Android devices or when it might not even occur to a would-be
user that Firefox for Android might not run. I've been researching the
Android device landscape to help a family member make an informed
device purchase and it seems to me that if a person has heard that
there's such a thing as "Firefox for Android" and wants to run it but
isn't reading Mozilla's wikis and development discussions, it's very
easy to be disappointed by buying an Android device that turns out not
to run Firefox for Android. (Though maybe one can be disappointed with
an ARMv7 device, too, if the perf isn't good enough.)
While I realize that giving users something that's slow is going to
disappoint users and supporting CPUs, OSs or OS versions unfocuses
development effort, I'm also worried that Mozilla is creating user
disappointment when the "for Android" product turns out not to run on
a newly-purchased Android device or, worse, when support gets dropped
while a device is still within its reasonable life time.
Robert Kaiser
> Has e10s increased memory usage significantly? Does Android use more
> RAM for the system than Maemo? Until July 2010, Fennec ran on N800
> which has 128 MB of RAM (and ARMv6).
I'm pretty sure that yes is the answer for both questions here. And on
N900 I hear that the 256MB of RAM are the largest problem, but apart
from a long launch time, Fennec runs reasonably there, esp. since the
9.0a1 Nightly builds. On N800, it was almost unusable speed-wise while
the stock browser, which also was Gecko-based, worked pretty nicely.
Irina Sandu
I calculated the % of Android devices on ARMv6 in the US, Western
Europe (UK, FR, DE, IT, SP) and Japan. These countries account for a
little over half of the Androids in the world.
Looking at the situation in June this year, 34% of all Android devices
in these markets run on ARMv6 CPUs. In December 2010 the rate was at
38%. If we leave out Japan, which is a rather different market than
the rest and has an unusually low % of ARMv6, the numbers are at 38%
for US + Western Europe for June and at 41% in December 2010.
The conclusion for the mature markets seems to be that the proportion
of ARMv6 devices is slowly going down and even right now it is under
the 50% mark. I know that in the emerging markets ARMv6 devices are
likely to have a bigger proportion and while I don't have any data
there (but I am looking for it), I think we can expect that they will
follow the same pattern that we see now in the mature markets, albeit
with a lag.
Irina
> _______________________________________________
> dev-planning mailing list
> dev-pl...@lists.mozilla.org
> https://lists.mozilla.org/listinfo/dev-planning
Irina Sandu | Mozilla | +49 1520 56 58 596
Mark Finkle
> On Thu, Sep 1, 2011 at 6:21 AM, Matt Brubeck<mbru...@mozilla.com> wrote:
>> Nearly all ARMv6 phones in the wild have 128 to 256 MB of RAM. The ones
>> with less than 256 MB have never been able to run Fennec at all. The ones
>> with 256 MB can run Fennec, but it crashes frequently and forces other apps
>> to quit. It also gets evicted from memory more often, so the user has to
>> restart the browser frequently - making slow startup even worse.
>
> Has e10s increased memory usage significantly? Does Android use more
> RAM for the system than Maemo? Until July 2010, Fennec ran on N800
> which has 128 MB of RAM (and ARMv6). (And sure, Opera Mobile runs much
> much better on the device, but Fennec still worked.)
Fennec barely ran on the N800, mostly due to the lack of swap (it wasn't
on by default). We ran out of memory on the N800 a lot!
The N900 is a bit better and has swap as well. Many of our OOM bugs went
away on the N900.
> What caught my attention about how phones match with Firefox systems
> requirements is that HTC Wildfire S has 512 MB of RAM but ARMv6.
It's possible we could run without OOM'ing all the time on this phone.
>> A very large percentage of ARMv6 devices are stuck on Android 2.1 or 1.6.
>> We can't support 1.6 at all because it doesn't have sufficient native code
>> support. We are hoping to drop Android 2.1 support soon because it is
>> becoming a maintenance burden.
>
> Does Mozilla have any kind of user-facing promise or plan about
> retaining OS support?
Not that I know about.
> Maemo 4 got dropped without any discussion or warning whatsoever one
> day Fennec just got autoupdated to a build that didn't run anymore.
> Maemo 5 is getting dropped while the corresponding device is still in
> principle being part of an active product line and being sold (or more
> to the point, it's been only about a year since the product was still
> being bought). On the Android side support started with Android 2.0,
> but 2.0 has already gotten dropped silently (or at least I never saw
> any announcement on the main Mozilla blog telling Android 2.0 users
> what was coming). Now 2.1 is getting dropped. Meanwhile, the OS
> upgrade story on the devices isn't that great but the devices aren't
> that old, either.
Fennec still supports Android 2.0 (check out AndroidManifest.xml.in), we
just stopped talking about Android 2.0 cause it's so darn old and
insignificant.
Dropping Maemo 4 and 5 were discussed during their respective cycles.
Since neither was ever a tier 1 platform, we didn't feel the need to
over-communicate. The intention was always, and still is, that the
community could pick up the Maemo builds and keep builds coming.
That never happened for Maemo 4. I don't think it will happen for Maemo
5. Meego does have some community work being contributed.
Gen Kanai
CPU usage across major markets.
One interesting piece of data that came out of Japan (at least) is
that most Android devices shipping are ARMv7, but even with that more
powerful CPU, the manufacturers are installing many services such as
NFC and terrestrial digital TV tuners which take a lot of RAM (that
Firefox would be using.)
So, faster CPUs are not a panacea. As CPUs get faster, more services
are added, etc.
Gen
On Thu, Sep 1, 2011 at 9:46 AM, Dietrich Ayala <auto...@gmail.com> wrote:
> Yeah I haven't been able to find any details about those various assumptions.
>
> Is the problem that ARMv6 compat requires a massive retooling? Or just
> that the reasonable perf targets are so far off that it's not worth
> attempting? Both? Other?
>
> As people have brought up in this and other threads, new ARMv6 devices
> are still shipping at this time, and doing very well in some areas of
> the world.
>
> On Wed, Aug 31, 2011 at 4:10 PM, Martijn <martijn...@gmail.com> wrote:
>> On Thu, Sep 1, 2011 at 12:39 AM, Dietrich Ayala <auto...@gmail.com> wrote:
>>> Why would you hate to focus energy on lower end processors? I'm not
>>> clear on the rationale for focusing exclusively on high-end devices.
>>
>> I assume it takes a lot of time and effort to fix everything that
>> would be related to ARMv6 and lower-end devices and by the time we
>> would have something that is working well enough, virtually nobody
>> would have those devices anymore.
>>
>> Regards,
>> Martijn
>>
>>> On Sun, Aug 28, 2011 at 11:57 PM, Doug Turner <do...@mozilla.com> wrote:
>>>> Not to discourage you from attempting to get something up and running again, but we have been actively ignoring ARMv6. I'd hate to focus any energy on lower end processors until we are incredible on faster ones.
>>>>
>>>> Doug
>>>>
>>>> On Aug 28, 2011, at 11:53 PM, Henri Sivonen wrote:
>>>>
>>>>> On Fri, Aug 26, 2011 at 10:19 PM, Thomas Arend <tarend...@gmail.com> wrote:
>>>>>> The other ones are Flash and "I want
>>>>>> this on my device". The latter is a "good" problem to have (except for
>>>>>> the fact that there are many ARMv6 devices out there).
>>>>>
>>>>> Are there active plans to support and supply builds for ARMv6? It
>>>>> seems that lower end Android phones are catching up with the RAM
>>>>> requirements of Firefox but are staying on ARMv6 for now.
>>>>>
>>>>> --
>>>>> Henri Sivonen
>>>>> hsiv...@iki.fi
>>>>> http://hsivonen.iki.fi/
>>>>> _______________________________________________
>>>>> dev-planning mailing list
>>>>> dev-pl...@lists.mozilla.org
>>>>> https://lists.mozilla.org/listinfo/dev-planning
>>>>
>>>> _______________________________________________
>>>> dev-planning mailing list
>>>> dev-pl...@lists.mozilla.org
>>>> https://lists.mozilla.org/listinfo/dev-planning
>>>>
>>> _______________________________________________
>>> dev-planning mailing list
>>> dev-pl...@lists.mozilla.org
>>> https://lists.mozilla.org/listinfo/dev-planning
>>>
>>
>>
>>
>> --
>> Martijn Wargers - Help Mozilla!
>> http://quality.mozilla.org/
>> http://wiki.mozilla.org/Mozilla_QA_Community
>> irc://irc.mozilla.org/qa - /nick mw22
>>