The SPEC 2006 CPU benchmark was just released. As an update of the now 6 year old SPEC 2000 benchmark, it represents a new first target and testing point for new CPU’s, new systems, and new compilers.
And I’m very happy to say Sjeng is a part of it.
I have to admit it, this person on the FLAC mailing list is better at ranting than me.
He gets extra bonus points for the consultant remark, for being particularly rude and for the comparison to autotools. Then there’s an additonal x2 multiplier for actually providing some attempt at a solution to the problem.
And yes, FLAC’s API is really that bad. Note the particularly stupid replies from the people who disagreed with him.
I’m pleased to announce that libFooID is now publicly available. As far as I know, it is the highest performance audio fingerprinting tool available under a free license.
Foosic.org has been updated with support for libFooID. Also available is AutoTag, a simple tagger which makes use of both of the above services.
http://www.theinquirer.net/?article=31543
I have to agree. In fact, I’m sure even Linus agrees. Wasn’t he telling everyone to use a vendors kernel if they wanted stability, and stressing the fact that there doesn’t exist such a thing as a stable kernel anymore starting with 2.6.x?
Here’s a, hrm, “funny”, story about dealing with production Linux servers:
Linux needed an update from 2.6.15 to newer to protect against a critical vulnerability. This broke the ethernet drivers. On a remote server. And it’s not some weird card either, it’s an Intel Pro 1000 which is a quite common server card. The bug didn’t come out of nowhere: it is known, and it’s not getting fixed. Makes you feel very confident about your server OS.
One RHEL3 server has an annoying bug in Apache that causes total hangs every now and then. Known bug, Red Hat isn’t fixing it. Manually upgrading Apache is possible, but then every security update must also be added manually. There are of course other distributions than RHEL, but in my experience they’re even worse for production servers.
The solution of course, is to use a serious operating system. FreeBSD is a good option. It will be interesting to see how many years it takes Linux to come up with something which approaches the ease of administration of portupgrade/portaudit/mergemaster.
The first problem wouldn’t have happened with FreeBSD. A security update wouldn’t have needed a new version of the kernel with 1000 buggy changes, just a make update; make buildkernel; make installkernel; reboot, which gets *only* the security fix. The fact that there was a vulenarbility in the first place would have been alerted via the mailinglist or (in case of a port) via portaudit. No need to track news sites manually. And this is all part of the base install and present by default.
The second one would be resolved simply by portupgrade. If there’s a security issue, again, portaudit would have raised a warning automagically.
From an administration point of view, FreeBSD is so far ahead of Linux it just makes me sad. Linux does have good points, such as the speed with which it runs MySQL. Not that that does you any good though, without a network card to talk to the world…And don’t try to make backups either, because the VM subsystem is so broken that your webserver will go down with them.
mysql> SELECT COUNT(*) FROM songs WHERE album IN ('S&M (Disc 1)');
+----------+
| COUNT(*) |
+----------+
| 12 |
+----------+
1 row in set (0.00 sec)
mysql> SELECT COUNT(*) FROM songs WHERE album IN ('S&M (Disc 1)', 'MySQL ate My query');
+----------+
| COUNT(*) |
+----------+
| 0 |
+----------+
1 row in set (0.03 sec)
Seems to be caused by the same bug as yesterdays.
The last remark I made in the previous entry was “Score another round for MySQL”.
Unfortunately, today MySQL scored a rather sad own-goal, which rather diminishes it’s good showing of yesterday.
http://bugs.mysql.com/bug.php?id=19080
Of course, realizing it was not me who was at fault took a while, and it doesn’t help my confidence that foosic.org is very dependant on exactly those indexes that seem to be buggered. Oh boy.
I imported the entire foosic database into both MySQL 5.0.19 and PostgreSQL 8.1.3, and ran the Top Album and Top Songs queries on them.
PostgreSQL: Top Album in 15 seconds, Top Songs in 11 seconds.
MySQL: Top Album in 8 seconds, Top Songs in 7 seconds.
Score another round for MySQL.
On the SQLite wiki you can find a series of benchmarks of various DB engines. I took the test, axed everything but MySQL-ISAM and PostgreSQL nosync, and ran it on FreeBSD 6.1/libthr/4BSD, together with the fix for the PostgreSQL config you can find in the discussion.
Two runs, one with the given-but-fixed PostgreSQL settings, one with more parameters entirely defaulted, but cache increased and fsync off.
fixed SQLite-wiki settings
Defaults + higher buffers
PostgreSQL is fast on non-indexed queries, in some join situations, and with deletes. Unfortunately, my applications are really about inserting, selecting and updating indexed tables…so MySQL really looks like the winner here.
I’ve been considering for some time to move to FreeBSD based servers, but there were 2 things holding me back a bit. The first was good Java support, the second was MySQL performance. FreeBSD recently announced official Java support, so the first one is completely out of the way. The second has reportedly improved, but I’m not the man to believe someone elses claims about something like this, so I ran my own benchmarks.
Ran under VMWare Workstation 5.5.1, on an Athlon64 X2 4400 host, using 1 CPU. Super-smack 1.3 running update-select, average of 3 runs. Results in queries per second.
linuxthreads being faster than NPTL was surprising, although it might just be noise. FreeBSD 6.1 is clearly a lot faster than 5.5, libthr and the 4BSD scheduler seems like the best combination. And Linux is still quite a bit faster, although for me personally the difference is now “small” enough that it wouldn’t influence the server choice any more. Note that “FreeBSD 7.0″ in the table is really just CURRENT from the time I wrote this.
Matched Donots to The Roots
with album Rock Against Bush, Volume 2 matching Rock Against Bush (Disc Vol. 2)
Matched Doop to Tool
with album Ænima matching Aenima
Matched Tartaros to Tori Amos
with album Tales Of A Librarian matching Tales from a Librarian
Matched H-Man to Zwan
with album Mary, Star Of The Sea matching Mary Star Of The Sea
Matched The Darnells to The Darkness
with album One Way Ticket To Hell & Back matching One Way Ticket To Hell ...And Back
Matched Break to Cream
with album Royal Albert Hall - London May 2005 (Disc 1) matching Royal Albert Hall - London May 2005 (Disc 1)
Matched Ween to Queen
with album Live At Wembley (Disc 1) matching Live At Wembley '86 (CD1)
Matched SWAP to Zwan
with album Mary, Star Of The Sea matching Mary Star Of The Sea
Matched The Shades to Slade
with album Get Yer Boots On: The Best of Slade matching Get Yer Boots On - The Best of
Matched The Silos to Tiles
with album Window Dressing (CD1) matching Window Dressing
Matched SLAPSHOT to Slipknot
with album Vol. 3: ( The Subliminal Verses ) matching Vol. 3 The Sumbliminial Verses
Matched The Bats to The JAMs
with album 1987 What the fuck is going on? matching 1987 (What The Fuck Is Going On?)
Matched The Ronettes to The Soviettes
with album Rock Against Bush (Disc Vol. 1) matching Rock Against Bush (Disc 1)
and a final cutie:
Matched Under The Sun to Wintersun
with album Wintersun matching Under The Sun