eee 901 SSD nilfs2 fedora (work in progress)
* fedora 8
* kde 3.5
* Compile WiFi driver howto
* Compile nilfs2 howto/dowload from www.nilfs.org howto
Benefits:
* WiFi self evident (Compile howto - builtin for FC9, FC10)
* nilfs2 highest SSD write performance. Not available for windows.
 - Thanks go to Nippon Telephone and Telegraph of Japan for nilfs
 - Considerable performance enhancement over Xandros/Unionfs
* Replacement SOE for record low price. Just plugin existing mouse and panel displays (dual screen operation)
* Fan is the only mechanical device left (some people cut the wire to kill it).
Config:
 - Boot partition still ext2 (onboard 4GB). No kernel mods.
 - most writeable(hot) directories moved to nilfs 16GB SSD (16GB SSD is replacable, onboard 4GB flash soldered in for life)
 - 16GB SDHC inserted into slot. Using nilfs2 for highest performance, almost as fast as internal SSD. 36GB total.
 - nilfs-cleanerd.conf retuned for SSD (still playing here).
Warning:
 - Ensure you have a copy of the nilfs kernel module source code in /root.
 - After a kernel upgrade I found that all nilfs filesystems were no longer accessible (obviously because nilfs is not built in).
 - Ensure you can compile and load the nilfs .ko with all nilfs filesystems unmounted ('cause that's where you'll be after a yum update - reboot scenario). I was just lucky!
VirtualBox WinXP emulation on Fedora/nilfs2 faster that eee 901 SSD Winodws. When in full screen mode the two are identical (including USB device access - my Nokia N95 8GB thinks it's talking to real windows).
[screen shots here]
Nilfs testing on our enterprise servers is underway. Looking to replace 15K SAS drives in Dell 2950 with SSD. So far so good. Latest news:
 - Problems with nilfs_cleanerd: unable to reclaim space quickly enough.
 - Cleanerd problem solved. The host now has 2 SSDs raid stripped to double capacity and speed. The raiding uses the device mapper with /dev/mapper/ssd connected directly to the xen domu machine. The Domu virtual is running nilfs on the virtual device. While there is only 97Gb or so on the SSD(256Gb) it is regularly at 80% "full". When there is a lull in activity the used space drops back towards 50% or less as cleanerd does its thing.
 - I'm now thinking that cleanerd needs a smarter cleaning algorithm to find the "dirtiest" blocks first and the "cleanest" last to speedup reclaimation. This, obviously, requires an index of blocks ordered on "dirtyness".
 - Complete success as the app has been running for a couple of weeks. The write speed is about 7 Mb/Sec continuous. The SSDs should live for at least 3 years and will probably die in 5 years.
 - Keep the DB log files on physical disks though. The write rate for DB logs (IBM DB2 and MySQL Innodb) is very very high.
 - Had a mysterious hardware issue where one of the SSDs "unplugged" itself and then "replugged" 15 seconds later. Investigating was inconclusive as it looked like a human did this. Video from the computer room showed no one was there. Never happened again. Nilfs, DB2 and Innodb all recovered fully. The apps didn't see anything other than a delay and a restart. Amazing really. All that logging stuff actually works.
- OCZ 128G SSD drive failure. Returned for warranty a couple of weeks ago. Shipped back to the US. Taking a *LONG* time. Looks like NILFS might be an SSD killer. And/Or maybe the lifetime numbers from OCZ are somewhat optimistic?
* fedora 8
* kde 3.5
* Compile WiFi driver howto
* Compile nilfs2 howto/dowload from www.nilfs.org howto
Benefits:
* WiFi self evident (Compile howto - builtin for FC9, FC10)
* nilfs2 highest SSD write performance. Not available for windows.
 - Thanks go to Nippon Telephone and Telegraph of Japan for nilfs
 - Considerable performance enhancement over Xandros/Unionfs
* Replacement SOE for record low price. Just plugin existing mouse and panel displays (dual screen operation)
* Fan is the only mechanical device left (some people cut the wire to kill it).
Config:
 - Boot partition still ext2 (onboard 4GB). No kernel mods.
 - most writeable(hot) directories moved to nilfs 16GB SSD (16GB SSD is replacable, onboard 4GB flash soldered in for life)
 - 16GB SDHC inserted into slot. Using nilfs2 for highest performance, almost as fast as internal SSD. 36GB total.
 - nilfs-cleanerd.conf retuned for SSD (still playing here).
Warning:
 - Ensure you have a copy of the nilfs kernel module source code in /root.
 - After a kernel upgrade I found that all nilfs filesystems were no longer accessible (obviously because nilfs is not built in).
 - Ensure you can compile and load the nilfs .ko with all nilfs filesystems unmounted ('cause that's where you'll be after a yum update - reboot scenario). I was just lucky!
VirtualBox WinXP emulation on Fedora/nilfs2 faster that eee 901 SSD Winodws. When in full screen mode the two are identical (including USB device access - my Nokia N95 8GB thinks it's talking to real windows).
[screen shots here]
Nilfs testing on our enterprise servers is underway. Looking to replace 15K SAS drives in Dell 2950 with SSD. So far so good. Latest news:
 - Problems with nilfs_cleanerd: unable to reclaim space quickly enough.
 - Cleanerd problem solved. The host now has 2 SSDs raid stripped to double capacity and speed. The raiding uses the device mapper with /dev/mapper/ssd connected directly to the xen domu machine. The Domu virtual is running nilfs on the virtual device. While there is only 97Gb or so on the SSD(256Gb) it is regularly at 80% "full". When there is a lull in activity the used space drops back towards 50% or less as cleanerd does its thing.
 - I'm now thinking that cleanerd needs a smarter cleaning algorithm to find the "dirtiest" blocks first and the "cleanest" last to speedup reclaimation. This, obviously, requires an index of blocks ordered on "dirtyness".
 - Complete success as the app has been running for a couple of weeks. The write speed is about 7 Mb/Sec continuous. The SSDs should live for at least 3 years and will probably die in 5 years.
 - Keep the DB log files on physical disks though. The write rate for DB logs (IBM DB2 and MySQL Innodb) is very very high.
 - Had a mysterious hardware issue where one of the SSDs "unplugged" itself and then "replugged" 15 seconds later. Investigating was inconclusive as it looked like a human did this. Video from the computer room showed no one was there. Never happened again. Nilfs, DB2 and Innodb all recovered fully. The apps didn't see anything other than a delay and a restart. Amazing really. All that logging stuff actually works.
- OCZ 128G SSD drive failure. Returned for warranty a couple of weeks ago. Shipped back to the US. Taking a *LONG* time. Looks like NILFS might be an SSD killer. And/Or maybe the lifetime numbers from OCZ are somewhat optimistic?