Lack of update information
Ralf Corsepius
rc040203 at freenet.de
Wed Jan 28 04:31:59 UTC 2009
Jesse Keating wrote:
> On Tue, 2009-01-27 at 03:11 +0100, Ralf Corsepius wrote:
>> Jesse Keating wrote:
>>> On Tue, 2009-01-27 at 01:22 +0100, Ralf Corsepius wrote:
>>>> The problem with this: Users don't notice the cases in which upgrades
>>>> "simply work" - They only notice the case, something already has gone wrong.
>>> Many users notice a plethora of updates for unknown reasons that chew up
>>> their bandwidth, cpu, and disk time while they're just trying to get
>>> work done. So much so that many opt out of applying updates at all,
>>> because there are so many of them, which hurts our abilities to deliver
>>> security updates.
>> These problems do not originate from the number of upgrades/update, they
>> originate from the _size_ of updates few packages introduce.
>
> Size is in direct correlation to number.
With all due respect, Mr. Keating, this is non-sense.
One update to openoffice/evolution/kde/gnome/kernel or some game's data
outweighs many other packages.
>>> and with new expensive fast hardware
>>> may not see a problem.
>> Well, I would label this an urban legend - Updating recent Fedoras using
>> yum has not been much of a performance problem, even on slower and older
>> HW (e.g. a 1GHz i686/512MB)
>>
>> Admitted, on an 64MB/166MHz i586 w/Fedora 10 yum updates tend to be
>> "really slow".
>
> Try doing updates on a netbook,
I own one - Typical "yum update" times are in the order of "very few"
minutes at average.
Example from a couple minutes ago (comprising kernel/kernel-devel,
PackageKit, selinux-policy and some rpmfusion updates - 20 packages in
total; Using local (fast) mirrors):
# cat /proc/cpuinfo
..
model name : Intel(R) Atom(TM) CPU N270 @ 1.60GHz
..
# time yum -y update
...
real 7m48.916s
user 4m50.155s
sys 0m57.314s
Not fast, but hardly a serious problem.
On an i586, the times requires for yum updates are much higher
(typically hours), times on old i686's are in the order of several 10
minutes.
> or any other system with relatively slow
> "disk" I/O, or "slow" CPU.
The cases I mention are real - I do own such machines.
On these, the real yum update related problems are bandwidth and
disk-space requirements (yum-caches).
PackageKit's GUI and parsing activities adds further (quite measurable)
requirements on CPU-power and RAM.
> Yum isn't really the problem here, it's the
> rpms themselves, constantly doing ldconfigs, cache updates, tonnes of
> hardlinking, etc...
From what I experience, SELinux policy updates are by far means the
most resource intense/most time consuming part of updating.
>> However, wrt. to bandwidth, I see another problem: Your strategy of
>> pushing updates in "big batches", instead of "small chunks".
>>
>> This makes a significant difference to users with limited bandwidth.
>> While they could easily poll "small chunks", e.g. once a day (blocking
>> their network for, e.g. 1 hour), the current strategy would block they
>> network for very much longer (e.g. 7 hours).
>
> If I were to push updates once a day, the mirrors would be in shambles,
I can't prove nor counter-prove this claim, but I doubt that this would
make much of a difference.
> constantly trying to pick up the new content and users would have a
> terrible experience. Pushing updates faster doesn't magically make the
> number of updates get smaller.
On the end-user's side it makes a substantial difference if whether one
of your "big pushes" blocks a machine/network for the whole day,
or whether a "yum update" finishes during a "work break".
>> Another aspect related to bandwidth: Revisiting
>> * the sizes of rpms (e.g. different payload compressors).
>> * differential updates/rpms.
>>
>> Seems to me as if these once "hot topics" have dropped off the Fedora
>> radar, in favor of "strangling/nagging the distro's contributors".
>>
>> Ralf
>>
>
> People are working on both of the above issues.
Really? I doubt your claim. At least I am not aware of any activities in
RH's rpm to change rpm's payload rsp. to perform measurements on the
impact of using modern compressors on rpm-payloads.
Ralf
More information about the fedora-devel-list
mailing list