From Dapper To Lucid, Four Years Of Ubuntu Benchmarks

[phoronix]

Phoronix: From Dapper To Lucid, Four Years Of Ubuntu Benchmarks

Last week we shared that we were benchmarking Ubuntu's current and past LTS releases and began by running graphics benchmarks looking at how the proprietary drivers from the past compare to open-source drivers from the present, but now we have our assortment of system benchmarks to publish from the Long-Term Support releases of Ubuntu 6.06.1, Ubuntu 8.04.4, and an Ubuntu 10.04 development snapshot. In this article, we are looking at how Ubuntu's performance has evolved over the past four years.

http://www.phoronix.com/vr.php?view=14770

[talvik]

How am I supposed to choose Ubuntu for servers with such big regressions in server apps(Apache and PostgreSQL)?
The four year old release beat the other releases by a really large margin! A lot of devs here at work use Ubuntu, but I think Ubuntu is just to bleeding edge for the production enviroment.

[kraftman]

How am I supposed to choose Ubuntu for servers with such big regressions in server apps(Apache and PostgreSQL)?
The four year old release beat the other releases by a really large margin! A lot of devs here at work use Ubuntu, but I think Ubuntu is just to bleeding edge for the production enviroment.

Use different file system or mount options. Article explains why there are such differences... Btw. you should probably care about data integrity?

Other benchmarks at Phoronix have shown similar performance drops in Apache when using EXT4. Of course, we are just looking at the performance numbers, but the data stored on the EXT4 file-system should be more reliable and offer other benefits not shown within benchmark results.

[talvik]

Use different file system or mount options. Article explains why there are such differences... Btw. you should probably care about data integrity?

There is still a regression in postgresql from 6.06 to 8.04 in the same filesystem. I think ext3 is quite reliable, hasn't it proven itself all these years in most linux servers?

[MaestroMaus]

There is still a regression in postgresql from 6.06 to 8.04 in the same filesystem. I think ext3 is quite reliable, hasn't it proven itself all these years in most linux servers?

EXT 3 is alright but isn't futureproof (because of the sizelimit). Also, I imagine that if someone's making a new server right now he cares more about de reliability then about the speed (in which case EXT4 is a better choice).

[Shining Arcanine]

EXT 3 is alright but isn't futureproof (because of the sizelimit). Also, I imagine that if someone's making a new server right now he cares more about de reliability then about the speed (in which case EXT4 is a better choice).

This is about filesystem integrity and not about reliability. A kernel patch can only increase reliability if it can keep disks from failing, which is not possible, unless you consider the pseudo-disk that you get when you run RAID, but since this is not about RAID, such details are irrelevant.

Anyway, filesystem integrity is not very important if it renders the server incapable of performing its job, and while whatever caused the regressions should help with integrity, it is a poor substitute for backups, which should be done on a regular basis regardless of what mount options are used.

[atmartens]

It would be doing 10.04 with ext3, and with ext4 mounted differently. Since these benchmarks are pretty much automated, I can't imagine it'd be too difficult. This will keep some of the variables constant and improve the picture.

[blackshard]

Can't really understand if there's evolution or devolution.

Wasn't ext4 famous for its speed (Boot time?) and infamous for some little instability problems?

BTW such a regression in apache is a failure: any sane website admin will never accept an order of magnitude regression, mostly because this is a LTS release.

[kraftman]

There is still a regression in postgresql from 6.06 to 8.04 in the same filesystem.

It's possible default mount options changed between releases. Afaik they were also modified quite long ago (maybe between 6.06 and 8.04?).

I think ext3 is quite reliable, hasn't it proven itself all these years in most linux servers?

Yes, Ext3 should be reliable and I don't know why Ext4 wants to be, so "super" reliable as some results may suggest. I'd like Ubuntu to be set to maximum performance (it has even debugging enabled...) and only Ubuntu Server Edition to use some safer options.

[kraftman]

BTW such a regression in apache is a failure: any sane website admin will never accept an order of magnitude regression, mostly because this is a LTS release.

Phoronix Apache benchmark doesn't benchmark real world Apache performance, because visitors aren't connecting from the server where the site they're visiting is hosted Phoronix Apache benchmark tests something else, but I don't know what ;>