?

Log in

No account? Create an account
kernel book - brad's life — LiveJournal [entries|archive|friends|userinfo]
Brad Fitzpatrick

[ website | bradfitz.com ]
[ userinfo | livejournal userinfo ]
[ archive | journal archive ]

kernel book [Nov. 5th, 2003|12:47 am]
Brad Fitzpatrick
I've been reading Robert Love's new "Linux Kernel Development" book which came this morning.

So far I've read the first four chapters: Intro, Processes, Scheduling, and System Calls. It's fun and addictive. I always have a fear with technical books they'll be too technical or too slow, but this book is right at my level so far.

I didn't go into work today but I got a lot done here at home without interruptions.

The folks at umem.com make the only NVRAM PCI card with open source drivers in the linux tree, but their website just gives an ASP error all day today: "Active Server Pages error 'ASP 0115' Unexpected error /iisHelp/common/500-100.asp A trappable error (C0000005) occurred in an external object. The script cannot continue running." Sad. I wanted to place an order too, but couldn't find what I needed. And nobody returned my emails yet. *sigh*

I found out Theodore Tso (Linux hacker) is on LJ as tytso (with a perm account!) so I politely bugged him about some ext3/jbd stuff. He replied right away with good info about jbd's state. I couldn't find the semantics of fsync on ext3... different things I found were contradictory about whether the sync was just to the journal, or to disk. Ted confirmed they were just to the journal, explicitly for the reason that people could put the journal on NVRAM.

Started benchmarking different I/O patterns with bonnie++ on different filesystems and journal locations, but ran into problems with available computers (oddly). I didn't want to mess with my stable server (changing my raid array), my laptop didn't have enough space (only root partition and too small swap space to temporarily use as filesystem), and my desktop has no disks. I'll have to benchmark stuff at work.

So far, though, it looks like ext3 with data=journal and the journal on NVRAM should kick ass. Even using a spinning journal I'm seeing random writes quite a bit faster (which makes sense, since the first time they're written they're sequential, then later they're batched) ... look forward to seeing it on memory.

Cement people are coming to jack hammer away some stupid steps on the side of my house and repour better ones tomorrow, and that's at like 8:30, so I should sleep and get out of here early.
LinkReply

Comments:
[User Picture]From: annenahaymus
2003-11-05 03:20 am (UTC)
Do you specialize in one language or are you versed in many?
(Reply) (Thread)
[User Picture]From: brad
2003-11-05 08:33 am (UTC)
programming language, or .... ?
(Reply) (Parent) (Thread)
[User Picture]From: taral
2003-11-05 08:29 am (UTC)
I really hope fsync isn't just journal when user data isn't being journalled...
(Reply) (Thread)
[User Picture]From: brad
2003-11-05 08:31 am (UTC)
ext3 has 3 modes for how to journal data. data=journal is the full mode.
(Reply) (Parent) (Thread)
[User Picture]From: taral
2003-11-05 10:43 am (UTC)
Yeah, but does fsync still sync the user data if the mode isn't data=journal?
(Reply) (Parent) (Thread)
[User Picture]From: brad
2003-11-05 12:35 pm (UTC)
Of course.
(Reply) (Parent) (Thread)
From: evan
2003-11-05 11:59 am (UTC)
whoa, tytso bought his own perm account? he deserves like eight of his own...
(Reply) (Thread)