BTRFS performance in 2020

[GreyGeek]

Before I was forced to switch to EXT4 I was using ResierFS. It was a killer filesystem!

[oshunluvr]

Me Too! We must have parallel file system history, lol

[GreyGeek]

I appreciate this kind of reply that reviews how some procedure is done. I've picked up a lot over the years reading them. In that spirit..

:flinch:

I prefer to always mount this via /etc/fstab:

LABEL=foo     /mnt/foo      btrfs   defaults,noatime              0   0

or to enable a user mount:

LABEL=foo     /mnt/foo     btrfs   defaults,noatime,noauto,user   0   0

(With the proviso that snapshots aren't backups for some purposes.) I prefer using sudo btrfs, and I use the file /etc/sudoers.d/btrfs with this line:

john     ALL=(ALL) NOPASSWD:/bin/btrfs*

so that I don't have to type my password for btrfs commands.

I sudo -i to root so that I don't have to prepend "sudo " in front of all of my subsequent commands. If fear of making a mistake while in root is a concern, it is no less a concern if one types "sudo " in front of a careless command submitted while in the home account.

I, too, have @ and @home mount stanzas in /etc/fstab but they only execute during boot.

When I do
mount /dev/disk/by-uuid/ce2b5741-c01e-4b3d-b6ba-401ad7f7fcdf /mnt
while in root, my system, as you know, is still running and it can continue to do things and let things run while I continue with the snapshotting, etc... But, since it rarely takes more than 30-40 seconds to do snapshot stuff because the commands are already in history and easy to edit a c&p, and I don't need to be concerned about background stuff.

My fstab contains

UUID=ce2b5741-c01e-4b3d-b6ba-401ad7f7fcdf /               btrfs   defaults,subvol=@ 0       1
# /home was on /dev/sda1 during installation
UUID=ce2b5741-c01e-4b3d-b6ba-401ad7f7fcdf /home           btrfs   defaults,subvol=@home 0       2

BUT, those are only executed during boot up and I certainly would not execute them again while running. I could easily add the /mnt option in fstab but there is no reason or purpose to do that unless I want to work with snapshots.


With my <ROOT_FS> mounted to /mnt while in root, I show the following folders, files and subvolumes:
/mnt/@
/mnt/@home

After the initial installation I added a standard folder:
/mnt/snapshots
into which hold the snapshots I create. Neither @, @home or ../snapshots are accessible to the user during normal operations without first mounting the <ROOT_FS> to /mnt (or whatever).

[mr_raider]

Well looks like I spoke too soon:

https://www.phoronix.com/scan.php?pa...esktop-Approve

[oshunluvr]

Well looks like I spoke too soon:

https://www.phoronix.com/scan.php?pa...esktop-Approve

Well, that interesting. Hopefully they will implement it properly. I was checking out some random distro a year or more ago and it had BTRFS as an install option, BUT DIDN'T USE SUBVOLUMES!

Totally defeats the purpose of BTRFS IMO. I even got on the distro IRC channel and asked about it and they all got insulted and told me to basically shove off. Obviously I deleted that VM about a second after I closed the IRC window.

[jlittle]

The benchmarks were run with "default" options, so no -o noatime, and I wonder if that had an impact.

I'd been wondering about the poor performance of apt. I'd like to do a Kubuntu install to ext4 to time the updates, but that will have to wait for another SSD for a fair comparison. Also, the comment about btrfs being vulnerable to drives lying about having written data when outages occur might explain my btrfs failure a year or so ago.

Well looks like I spoke too soon:

In what way may you have spoken too soon?

https://www.phoronix.com/scan.php?pa...esktop-Approve

The comments were an interesting read. (If only to reaquaint myself with how easily such discussions can degenerate into flaming, with a bit of trolling. I couldn't work out whether some of the trolls were being meta as an implied snipe at the earlier trolls; vi vs nano? seriously?) btrfs had some strong support.

[GreyGeek]

Interestingly, Oracle is putting BTRFS into is Unbreakable Enterprise Kernel (UEK 6) release using the 5.4 kernel series because of its compatibility with BTRFS.

For me, BTRF, like KDE, is an nonnegotiable requirement for any distro I choose to use. If Ubuntu depreciates or removes BTRFS I will move to a distro that supports both.

[GreyGeek]

Me Too! We must have parallel file system history, lol

Seems so!

In the spring of 2002 I was using SuSE 6.4, which featured ReiserFS. I had it on a second box in my cubical at work, running PostgreSQL and I was also using it on my SuSE powered workstation, as a dual boot with Windows at the time. The Dept of Revenue had a Win98 desktop running WildCat BBS system with 2 phones connected in roll-over. Tax consultants could post their client's tax returns using the system and then dial in later to see the status of those returns. The problem was that Win98 was left on 24/7 during the tax season had a tendency to crash. During business hours in was no problem for someone in IT to walk to the server room and reboot the Win98 to get the BBS running again. But on weekends it meant that someone from IT had to drive to work and reboot the Win98. Sometimes twice or three times or more on a weekend.

I had previously used my SuSE 6.4 server to read the data off of a defunct Kodak 1GB Unix HD and the head of Revenue and IT came by my cubical to see if I could do something about the BBS. I asked for a desktop with two 1GB HDs and got them. (I learned later that the IT guy gave them to me because he didn't consider them reliable and was sending them to surplus for sale.) I wrote three Python scripts to mimic the BBS and made the system turnkey by making my looping menu script the login executable. Files that used to be in /etc/skel helped with setting that up. Settings in the /etc/inittab script made the two phones roll-over. While I was demonstrating it to the honchos one asked what would happen if the power failed. I reached over and pulled the power cord out of the socket. "Like this?". I plugged it back in and the computer booted (power switches were mechanical, not software, back then) and it displayed the BBS menu. ReiserFS allowed that. The Revenue dept had several power interruptions prior to that because of squirrels and other problems and I had learned that ReiserFS recovered gracefully from ABENDS. That Python application ran for 18 months without a single failure of either the BBS or the SuSE system underneath.

[mr_raider]

The benchmarks were run with "default" options, so no -o noatime, and I wonder if that had an impact.

I'd been wondering about the poor performance of apt. I'd like to do a Kubuntu install to ext4 to time the updates, but that will have to wait for another SSD for a fair comparison. Also, the comment about btrfs being vulnerable to drives lying about having written data when outages occur might explain my btrfs failure a year or so ago.

In what way may you have spoken too soon?

The comments were an interesting read. (If only to reaquaint myself with how easily such discussions can degenerate into flaming, with a bit of trolling. I couldn't work out whether some of the trolls were being meta as an implied snipe at the earlier trolls; vi vs nano? seriously?) btrfs had some strong support.

Well Fedora just approved something RHEL dumped!

[oshunluvr]

Seems so!

In the spring of 2002 I was using SuSE 6.4, which featured ReiserFS. I had it on a second box in my cubical at work, running PostgreSQL and I was also using it on my SuSE powered workstation, as a dual boot with Windows at the time. The Dept of Revenue had a Win98 desktop running WildCat BBS system with 2 phones connected in roll-over. Tax consultants could post their client's tax returns using the system and then dial in later to see the status of those returns. The problem was that Win98 was left on 24/7 during the tax season had a tendency to crash. During business hours in was no problem for someone in IT to walk to the server room and reboot the Win98 to get the BBS running again. But on weekends it meant that someone from IT had to drive to work and reboot the Win98. Sometimes twice or three times or more on a weekend.

I had previously used my SuSE 6.4 server to read the data off of a defunct Kodak 1GB Unix HD and the head of Revenue and IT came by my cubical to see if I could do something about the BBS. I asked for a desktop with two 1GB HDs and got them. (I learned later that the IT guy gave them to me because he didn't consider them reliable and was sending them to surplus for sale.) I wrote three Python scripts to mimic the BBS and made the system turnkey by making my looping menu script the login executable. Files that used to be in /etc/skel helped with setting that up. Settings in the /etc/inittab script made the two phones roll-over. While I was demonstrating it to the honchos one asked what would happen if the power failed. I reached over and pulled the power cord out of the socket. "Like this?". I plugged it back in and the computer booted (power switches were mechanical, not software, back then) and it displayed the BBS menu. ReiserFS allowed that. The Revenue dept had several power interruptions prior to that because of squirrels and other problems and I had learned that ReiserFS recovered gracefully from ABENDS. That Python application ran for 18 months without a single failure of either the BBS or the SuSE system underneath.

My experience wasn't a professional one, but had more to do with performance. ReiserFS was so much faster and less likely to have problems (as you pointed out) I quickly became a huge fan - this is back when it was new. I think I first heard of it 2002-03. It was the first journeled FS I had heard of and it was B-tree like BTRFS (coincidence?) so I dove right into it. It was shocking when Hans went to prison and Reiser4 never got a foot-hold, so I moved on.

[GreyGeek]

... It was shocking when Hans went to prison and Reiser4 never got a foot-hold, so I moved on.

That's why I called RiserFS a "killer" app.

It amazed me that someone so smart could be so dumb.

When I was operating my consulting business, which included criminal forensics (taking advantage of my Physics, Math and Biology courses and teachings). A lawyer contracted with me to investigate the murder his client was charged with in hopes of finding information which could point to another person. This was a year after the shooting. The defendant was a graduate of UNL and had his own business selling and servicing IBM PC Jrs. A very smart guy. You can tell where I am going with this.

The victims, his ex girl friend and her new lover, were shot several times. He in the hallway and she while hiding under the bed. He died at the scene. She died almost a week later in the hospital after verbally identifying our genius as the shooter. (Hospital record printouts showed that she was given incompatible blood but was cross out by pencil and the correct blood type printed in. She very well could have died of a major system wide blood clotting.) Anyway, I had all the photos of the victims and their wounds. Her photos included hospital wrappings, which made the entry point of the abdominal wound difficult to determine. The wrappings made it appear that the shot entered from front right side of her abdomen, almost in from the side. When I showed him that photo he corrected me and said that the entry would was in the back, near the right side, about four inches away from where the entry wound "appeared" to be. At that moment I realized that only the shooter would know that bit of information because NONE of the photos made that obvious.

I pointed this out to his attorney and submitted my bill. The attorney refused to pay. Two years later the genius's parents hired that lawyer to sue the state for wrongful imprisonment and the lawyer asked for my report. I told him that I would be happy to give it to him as soon as I receive a cashier's check for my unpaid bill. He paid it. They lost.

His elaborate plan included telling everyone he was taking a vacation to England and would be gone for a month. He left and drove to Des Moine, where he paid for gas for his van with a credit card, making sure he was noticed by the cashier. He then drove back to Lincoln, arriving before sunrise, committed the murders, and drove up I29, stopping at a car park 20 miles east of I29, instead of at one of the rest stops on I29. His van had several gas cans in the back and he drove into Canada and across most of Canada to re-enter the US at a Maine border. There he planned to sell the van and fly to England. It wasn't the car dealer's first rodeo. He noticed the cans and camping gear and assumed the guy was a runner, so he called his friend the sheriff.

His estate was later sued for $800,000 and in his defense, acting as his own lawyer , he said the Dr that operated on her didn't know the entry wound from the exit wound. The Dr said she had died after 5 days of suffering because her blood was "low on Oxygen" and she ripped out her tubing during a state of confusion. That explanation also fits blood clotting. He's been serving two life terms since his second trial in 1995.

[oshunluvr]

Interesting story. I think it must be almost impossible to get away with murder in those circumstances - spurned lover and all. Too much rush to get the job done and not enough "space" between victim and killer.

I looked up Hans - denied parole this year, next up is 2023. He was so dumb (or disconnected from reality) he had books about committing murder in the car he killed her in. I've got an ex-wife I for sure don't care much for but I can't imagine killing someone for something as dumb as ending a marriage.