From mboxrd@z Thu Jan 1 00:00:00 1970 From: David Masover Subject: Re: Bug report Date: Wed, 13 Apr 2005 23:45:37 -0500 Message-ID: <425DF571.3000004@slaphack.com> References: <5a59ce530504110707e0a4791@mail.gmail.com> <1113233427.4420.221.camel@tribesman.namesys.com> <5a59ce53050412065715fc6929@mail.gmail.com> <1113396590.5168.446.camel@tribesman.namesys.com> Mime-Version: 1.0 Content-Transfer-Encoding: 7bit Return-path: list-help: list-unsubscribe: list-post: Errors-To: flx@namesys.com In-Reply-To: <1113396590.5168.446.camel@tribesman.namesys.com> List-Id: Content-Type: text/plain; charset="us-ascii" To: Vladimir Saveliev Cc: studdugie , reiserfs-mailing-list -----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Vladimir Saveliev wrote: [...] > reiser4 has bitmap blocks spread uniformly over whole device. Those > blocks are to to be read on mount. > > There is mount option "-o dont_load_bitmap" which speedups mount but may > make performance worse. Is that really "don't load" or "don't preload"? In other words, does this just make the FS slow for awhile, then as fast as normal as these things get cached in RAM? Or does this make it so that the bitmap never gets created in RAM, and must always be read from disk? An analogy might be Software Suspend. A suspend can be really quick -- just swap out everything, then write the kernel, process, and general state info out to disk. When this boots up, it's really quick to get back into the OS, but it's insanely slow to actually get a desktop, because of all the seeks as it frantically tries to load everything that's being told to reload a window -- and it's loading everything at once, which means a thrashing of seeks. A suspend can also be much longer -- write out the caches, too, basically dump at least half of RAM out to disk, and probably the whole thing. But when you resume this, it doesn't take much longer to get into the OS, and you get your desktop back quicker, because everything's already been cached with a minimum of seeks. (Coolness: This isn't just theoretical anymore. I've actually gotten suspend2 and reiser4 working on one machine, and reiser4 + unionfs + gmailfs + kqemu + nvidia's gl drivers working on another. I think it's definitely stable enough to go into a 2.6 tree, even marked as "experimental".) If this is how reiser4 works, I think dont_load_bitmap should be the default, because if you're going to do a bunch of seeks anyway, why not do them as needed -- a millisecond here, a millisecond there -- instead of all at once while the user's waiting for the thing to boot? -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.1 (GNU/Linux) Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org iQIVAwUBQl31cXgHNmZLgCUhAQK0nA//Uym9KEpYivb4OsJNXuYCQ3u9GP/K9GNa KriyufqAo+Pth5rwBD8cgYWQwgiqZY4P8C7RsXwzKRz8ZUjt1KJ83DEekkDNPVRC zhF1P7WWI/rlFLQEg6OWtU/tuCG36VuPZK77faxxbkrcNZoPY4k9hBL/aoKi8L9A DagItRqDerevjhfky4OjM83cIrug4hQApxVjnpopT5LY2y4Md+ygqwtgVYHWEg6l wwqBgtstpgrGnyId6BB9m/asI1evCuuD2NO3kzOpRuX8xgF/POnXDtVCMpUymCmb r6IcWLlv6Z6fAk/mJ+QqpQZIflDu1Wmr+bzJ4vGznN2lkeiZU3dAdyWWVC28tofi oJPrdfwydcYqfFr6jSGDXwqeQUrh5AYHGyPYgRH65fy4cBs3KxvY5A8135xtL0+t CGLXKQCn6zt/Hw8xbcAusrIbi+pxE6M5BY2LspJ1SjzF/xdYs+w+foaz2/Ktnkgo vQdQ6b0nB4RApuTx5ajJwUbuVzNDn2kTpqgSq8aZhzinIl4ILv+N1DQ+1Cnw6UON lMv2KmVSoLEtVQSm5n6EpE6fTfh364ECNuL75cPYnuaPOH/73zQ6cOyOCukIz0oM SiTfPPAxz+6MnAM9Qre32ksUFSZ0yp5o/FvW+pbNL0V1MLLur0uh3mj8+HXFVdtJ vTpk7eGklig= =T6Pu -----END PGP SIGNATURE-----