[Yum-devel] yum on an olpc machine (slooooooooooow)

[Panu Matilainen]

On Sat, 16 Dec 2006, seth vidal wrote:

> On Sat, 2006-12-16 at 17:35 +0200, Panu Matilainen wrote:
>>>
>>> -bash-3.1# time yum --disablerepo='*' --enablerepo='extras' list
>>> available >> /dev/null
>>              ^^^^^^^^^^^^
>>
>>> looks like it is less important on the olpc machine. given the numbers.
>>> It shaves a little less than a second. Keep in mind in that run of 1m38s
>>> it is only parsing primary.xml.gz for extras.
>>
>> That's because you're redirecting output of both to /dev/null. Time it
>> without that redirection like it was in my examples.
>
> Do you seriously think it's going to shave off massive amounts of time
> to suddenly make the processing time USABLE if I don't redirect it
> to /dev/null?

No. I was only suggesting that on *normal operation* where you probably
don't redirect to /dev/null it might be way worse.

> So I did the tests w/o /dev/null. It shaved off 7 seconds out of 4m35s
> for the makecache run.
>
> so it dropped to 4m28s
>
> This feels like deckchair shuffling.

A "normal" modern system wastes significant percentage (1.5s of 9.5s is no 
small amount) of time just drawing progress bars. I assumed it might be on 
the same order on the OLPC (I don't have one so how am I supposed to 
know), but obviously that's not the case, it's bogged down enough by 
something else, disk access or whatever.

> So let's step away from the deckchair shuffling and do a run of 'yum
> list available' w/o any special options or anything. This is a command
> that I think someone would commonly run:
>
> This system has 8 repos enabled:
>
> core 2242 pkgs
> olpc_devel_kernel_repo 15 pkg
> updates 1098 pkgs
> olpc_content 1 pkg
> olpc_development 36 pkgs
> extras 4984 pkgs
> olpc-csound 1 pkg
> olpc-etoys 15 pkgs
>
>
> here is the results:
>
> real    3m46.294s
> user    2m35.380s
> sys     0m11.020s
>
>
> now, I'll run it again w/ the info cached so it doesn't have to do the
> xml parsing:
>
> real    1m13.360s
> user    1m6.730s
> sys     0m2.030s
>
>
> It's still a lot of time at 1m13s so there's room for improvement in
> areas outside of the metadata parsing, too.
>
> But think about this in terms of user experience. Sit still for 2m30s
> and do nothing, except count the time. It'll drive you bonkers, quickly.

Sure, that hurts.

 	- Panu -