Improve so it works on HDD disks

Bug #338822 reported by Scott James Remnant (Canonical) on 2009-03-06
12
This bug affects 1 person
Affects Status Importance Assigned to Milestone
sreadahead (Ubuntu)
Wishlist
Scott James Remnant (Canonical)

Bug Description

Binary package hint: sreadahead

There's nothing about the sreadahead software that means it couldn't be also used for HDD disks, for example an option to sort and use the pack file differently.

If we do this, we can then replace readahead-list with sreadahead

Changed in sreadahead:
assignee: nobody → scott
Changed in sreadahead (Ubuntu):
status: New → In Progress

Testing has shown that due to the better pack generation, sreadhead is actually just as good as readahead-list *without* being optimised - HDD optimisation is still desirable though to get even more performance out of it

Changed in sreadahead (Ubuntu):
importance: Undecided → Wishlist
Hernando Torque (htorque) wrote :

I cannot confirm your observation. I've 'bootcharted' two systems and with both sreadahead was slower than not using preloading at all - readahead-list was faster [I've stopped before GDM loading for pack file generation and used that list with readahead].

Sys 1 (AMD Opteron 144 @ 2.65GHz, 2GB RAM, 7200rpm disk):
- pack file: 733 files with 73.456.165 bytes
* no (s)readahead: 18.0s
* readahead: 17.4s
* sreadahead: 18.4s
* sreadahead started with --no-fork: 19.3s

Sys 2 (Intel Core Duo T2400 @ 1.83GHz, 2GB RAM, 5400rpm disk):
- pack file: 829 files with 57.408.165 bytes
* no (s)readahead: 23.1s
* readahead: 21.9s
* sreadahead: 24.6s
* sreadahead started with --no-fork: 24.3s

The gap is not big but it's there.

---

2.6.31-9-generic (Core Duo system used -7-generic)

readahead:
  Installed: 1:0.20050517.0220-1ubuntu5

sreadahead:
  Installed: 1.0-2

After you installed sreadahead, did you reboot twice and measure the second
time? I believe sreadahead needs to profile the boot the first time around.

On 09/01/2009 11:22 AM, Hernando Torque wrote:
> I cannot confirm your observation. I've 'bootcharted' two systems and
> with both sreadahead was slower than not using preloading at all -
> readahead-list was faster [I've stopped before GDM loading for pack file
> generation and used that list with readahead].
>
> Sys 1 (AMD Opteron 144 @ 2.65GHz, 2GB RAM, 7200rpm disk):
> - pack file: 733 files with 73.456.165 bytes
> * no (s)readahead: 18.0s
> * readahead: 17.4s
> * sreadahead: 18.4s
> * sreadahead started with --no-fork: 19.3s
>
> Sys 2 (Intel Core Duo T2400 @ 1.83GHz, 2GB RAM, 5400rpm disk):
> - pack file: 829 files with 57.408.165 bytes
> * no (s)readahead: 23.1s
> * readahead: 21.9s
> * sreadahead: 24.6s
> * sreadahead started with --no-fork: 24.3s
>
> The gap is not big but it's there.
>
> ---
>
> 2.6.31-9-generic (Core Duo system used -7-generic)
>
> readahead:
> Installed: 1:0.20050517.0220-1ubuntu5
>
> sreadahead:
> Installed: 1.0-2
>

Hernando Torque (htorque) wrote :

Yes. First boot was to a login prompt w/o GDM loading (so the pack file would only contain GRUB -> GDM files). Then I went to /var/lib/sreadahead and waited for the pack file to show up. Next boot I ran with bootchart enabled.

Hernando Torque (htorque) wrote :

I repeated the test with the AMD system but this time took bootcharts of a full boot to desktop:

- pack file: 1531 files with 79.994.830 bytes
* readahead: 37s
* sreadahead --no-fork: 41s

Just for the record: I'm extracting the file list for readahead from the pack file, I'm not just using the pack file with readahead. So those file lists should be the same.

Hernando Torque (htorque) wrote :
Hernando Torque (htorque) wrote :

That's not a fair test.

Please compare readahead-list for your system with the file list that *readahead* generated, against sreadahead for your system with the file list that *sreadahead* generated.

Part of the reason for switching is that sreadahead generates better pack file lists, such that sreadahead results in a generally as fast or faster boot than readahead-list, even though sreadahead isn't optimised for HDD yet.

Hernando Torque (htorque) wrote :

Ok, this time with untouched file lists (again on the AMD system).

readahead: 36s
- /etc/readahead/boot: 2399 files with 204.983.516 bytes (stopped readahead's profiling after the desktop was loaded)

sreadahead --no-fork: 42s
- /var/lib/sreadahead/pack: 1530 files with 81.780.819 bytes

Hernando Torque (htorque) wrote :
Hernando Torque (htorque) wrote :

On Wed, 2009-09-02 at 21:53 +0000, Hernando Torque wrote:

> readahead: 36s
> - /etc/readahead/boot: 2399 files with 204.983.516 bytes (stopped readahead's profiling after the desktop was loaded)
>
Again, please stop screwing around with the readahead lists.

If you're not going to compare the readahead package *as it is shipped
with Ubuntu* against the sreadahead package *as it is shipped with
Ubuntu*, you're not helping.

 status invalid

Scott
--
Scott James Remnant
<email address hidden>

Changed in sreadahead (Ubuntu):
status: In Progress → Invalid
Hernando Torque (htorque) wrote :

> If you're not going to compare the readahead package *as it is shipped
> with Ubuntu* against the sreadahead package *as it is shipped with
> Ubuntu*, you're not helping.

I'm sorry, but that's not what you've asked for in #8. :-)

Using the 'boot' and 'desktop' file shipped with readahead I indeed get equal times for readahead, sreadahead and sreadahead --no-fork (usable desktop after ~24s on the Core Duo system with a 7200rpm disk).

Please note that I didn't want to make sreadahead look bad, I just couldn't confirm your observation. Thanks for clearing things up!

To post a comment you must log in.
This report contains Public information  Edit
Everyone can see this information.

Other bug subscribers