Btrfs Allocator Hints(lwn.net) |
Btrfs Allocator Hints(lwn.net) |
- https://github.com/kakra/linux/pull/36
- https://wiki.tnonline.net/w/Btrfs/Allocator_Hints
What do you think?
It seems these patches possibly fix that.
I used cryptsetup with BTRFS because I encrypt all of my stuff. One day, the system froze and after reboot the partition was unrecoverably gone (the whole story[1]). Not a real problem because I had a recent backup, but somehow I lost trust in BTRFS that day. Anyone experienced something like that?
Since then I switched to ZFS (on the same hardware) and never had problems - while it was a real pain to setup until I finished my script [2], which still is kind of a collection of dirty hacks :-)
1: https://forum.cgsecurity.org/phpBB3/viewtopic.php?t=13013
These same people are the only ones in the world suggesting btrfs is "basically" stable. I'll never touch this project again with a ten foot pole, afaic it's run by children. I'll trust adults with my data.
That said, I avoid it like the plague on servers, to get acceptable performance (or avoid fragmentation) with VMs or databases you need to disable COW which disables many of it's features, so it's better just to roll with XFS (and get pseudo-snapshots anyway).
https://wiki.tnonline.net/w/Blog/SQLite_Performance_on_Btrfs
https://btrfs.readthedocs.io/en/latest/Status.html
edit: just checked, it says the same thing in man pages — not for production use, testing/development only.
I was reluctant to use BTRFS on my Linux laptop but for the last 3 years I have been using it with 4K cryptsetup with no issues.
it looks like you didn't use raid, so any FS could fail in case of disk corruption.
While it could have been a bit flip that destroyed the whole encryption layer, BTRFS debugging revealed that there was some traces of BTRFS headers after mounting cryptsetup and some of the data on the decrypted partition was there...
This probably means the encryption layer was fine. The BTRFS part just could not be repaired or restored. The only explanation I have for this that something resulted in a dirty write, which destroyed the whole partition table, the backup partition table and since I used subvolumes and could not restore anything, most of the data.
Well, maybe it was my fault but since I'm using the exact same system with the same hardware right now (same NVMe SSD), I really doubt that.
And how would RAID help in that situation?
Now, there is a cambrian explosion going on. Ext4, xfs, btrfs,bcachefs, zfs. They each have their pros and cons, and it takes a while before you find out you run into an expensive limit. E.g. Ext3/4 is good, until it ran out of inodes. ZFS is good, but has only 1 password for full disk encryption and I want to store a second one with IT. According to the jungle drums, btrfs eats your data once in a while. Bcachefs stupidly tries to get itself rejected from Linux, not good for long term stability. I'm on XFS now, but let's see how that ends.
(pretty much the 3 filesystems I think about ATM are ext4 as a standard boot drive, zfs for large, long-lived data storage, and FAT/exFAT for interoperability with windows. It'd have to be a pretty niche use-case for me to consider another option. BcacheFS sounds really interesting but only to experiment with right now)
I only really trust ZFS on Linux, but it's such a bother it can't be upstreamed and isn't fully integrated with the native Linux caching, as the native file systems are. Ext is fine too but it's missing features like checksumming and compression, and has limitations as you mentioned.
This feature is for performance, not the case you described.
I will definitely try this.
Does it really matter in the modern world where a vanilla two-socket rack unit has a terabyte of DRAM? Everything at scale happens in RAM these days. Everything. Replicating across datacenters gets you all the reliability you need, with none of the fussing about storage latency and block device I/O strategy.
Sun's ZFS7420 had a terabyte of RAM per controller, and these work in tandem, and after a certain pressure, the thing can't keep up even though it also uses specialized SSDs to reduce HDD array access during requests, and these were blazingly fast boxes for their time.
When you drive a couple thousand physical nodes with a some-petabytes sized volumes, no amount of RAM can save you. This is why Lustre divides metadata servers and volumes from file ones. You can keep very small files in metadata area (a-la Apple's 0-sized, data-in-resource-fork implementation), but for bigger data, you need to have good filesystems. There are no workarounds from this.
If you want to go faster, take a look at Weka and GPUDirect. Again, when you are pumping tons of data to your GPUs to keep them training/inferring, no amount of RAM can keep that data (or sustain the throughput) during that chaotic access for you.
When we talked about performance, we used to say GB/sec. Now a single SSD provides that IOPS and throughput provided by storage clusters. Instead, we talk about TB/sec in some cases. You can casually connect terabit Ethernet (or Infiniband if you prefer that) to a server with a couple of cables.
And that's just at the largest scale. I'm pretty sure banks still insist that the data is written to (multiple) disks (aka "stable storage") before completing a transaction.
Considering that multiple ZFS developers get paid to make ZFS work well on petabyte-sized disk arrays with SSD caching, and one of them often reports on progress in this area in his podcasts (2.5admins.com and bsdnow if you're interested) .. then yes?
Your point raid 5/6 not being tested heavily by actual users is entirely on point, those enterprise heavy users are only running RAID 10 like configurations.
If you want RAID 5/6, just use ZFS as they have solved all of these issues. I don't know if its due to sheer luck or maybe the fact is that Sun at its time was actually running RAID 5/6 in production (hard drives were not as cheap back then as they are now)?
Regardless, development isn't going to stop, we may just have to switch to shipping as a DKMS module. And considering the issues we've had with getting bugfixes out that might have been the better way all along.
I think it had nothing to do with the encryption layer... the FS layer was the problem.
anecdotes could be exchanged in both directions: I run heavy data processing with max possible throughput on top of btrfs raid for 10 years already, and never had any data loss. I am absolutely certain if you expect data integrity while relying on single disk: it is your fault.
But you know who's fault is it? It's btrfs's one. Other filesystems don't lose entire volumes that easily.
Over time, I've abused ext4 (and ext3) in all sorts of ways: override random sector, mount twice (via loop so kernel's double-mount detector did not work), use bad SATA hardware which introduced bit errors.. There was some data loss, and sometimes I had to manually sort though tens of thousands of files in "lost+found" but I did not lose the entire filesystem.
I only saw the "entire partition loss" only happened to me when we tried btrfs. It was a part of ceph cluster so no actual data was lost.. but as you may guess we did not use btrfs ever again.
there are scenarious where btrfs is currently can't be replaced: high performance + data compression.
we are in disagreement on this. If partition table entry corrupted, you can't mount without some low level surgery.
> And how would RAID help in that situation?
depending on raid, your data will be duplicated on another disk, and will survive in case of one/few disks corruption.
There are a lot of RAIDs and configurations. Some of them may do what you describe, but most don't.
You aren't doing that with ZFS or btrfs, though. Datacenter-scale storage solutions (c.f. Lustre, which you mention) have long since abandoned traditional filesystem techniques like the one in the linked article. And they rely almost exclusively on RAM behavior for their performance characteristics, not the underlying storage (which usually ends up being something analogous to a pickled transaction log, it's not the format you're expected to manage per-operation)
ZFS can, and is actually designed to, handle that kind of workloads, though. At full configuration, ZFS7420 is a 84U configuration. Every disk box has its own set of "log" SSDs and 10 additional HDDs. Plus it was one of the rare systems which supported Infiniband access natively, and was able to saturate all of its Infiniband links under immense load.
Lustre's performance is not RAM bound when driving that kind of loads, this is why MDT arrays are smaller and generally full-flash while OSTs can be selected from a mix of technologies. As I said, when driving that number of clients from a relatively small number of servers, it's not possible to keep all the metadata and query it from the RAM. Yes, Lustre recommends high RAM and core count for servers driving OSTs, but it's for file content throughput when many clients are requesting files, and we're discussing file metadata access primarily.
But... everything does fit in RAM at scale. I mean, Cloudflare basically runs a billion dollar business who's product is essentially "We store the internet in RAM in every city". The whole tech world is aflutter right now over a technology base that amounts to "We put the whole of human experience into GPU RAM so we can train our new overlords". It's RAM. Everything is RAM.
I'm not saying there is "no" home for excessively tuned genius-tier filesystem-over-persistent-storage code. I'm just saying that it's not a very big home, that the market has mostly passed the technology over, and that frankly patches like the linked article seem like a waste of effort to me vs. going to Amazon and buying more RAM.
source: https://blog.cloudflare.com/why-we-started-putting-unpopular...
> Our storage layer, which serves millions of cache hits per second globally, is powered by high IOPS NVMe SSDs.
Moreover, you again give examples from the end product. Finished sites, compacted JS files, compressed videos, compiled models...
There's much more than that. The model is in RAM, but you need to rake tons of data over that GPU. Sometimes terabytes of data. You have raw images to process, raw video to color-grade, unfiltered scientific data to sift through. These files are huge.
A well processed JPG from my camera is around 5MB, but RAW version I process is 25MB per frame, and it's a 24MP image, puny for today's standards. Your run of the mill 2K video takes a couple of GBs after final render at movie length. RAWs take 10s of terabytes, at minimum. Unfiltered scientific data again comes in terabytes to petabytes range depending on your project and instruments you work on, and multiple such groups pull their own big datasets to process real-time.
In my world, nothing fits in RAM except the runtime data, and that's your application plus some intermediate data structures. The rest is read from small to gigantic files and written in files of unknown sizes, by multiple groups, simultaneously. These systems experience the real meaning of "saturation", and they would really swear at us at some cases.
Sorry, but you can't solve this problem by buying more RAM, because these workloads can't be carried to clouds. They need to be local, transparent and fast. IOW, you need disk systems which feel like RAM. Again, look what Weka (https://www.weka.io/) does. It's one of the most visible companies which make systems behave like a huge RAM, but with multiple machines and tons of cutting edge SSDs, because what they process doesn't fit in RAM.
Lastly, oh, there's a law which I forget its name every time, which tells you if you cache 10 most used files, you can serve up to 90% of your requests from that cache, if your request pattern is static. In cases I cite, there's no "popular" file. Everybody wants their own popular files which makes access "truly random".
Sadly, there are people (and distributions) which recommend btrfs for general-purpose root filesystem, even for the cases where reliability matters much more than performance. I think that part is a mistake,
Hell even the compression algorithm that ZFS has uses/has access to (LZ4) is faster than what btrfs uses and with enough IO that matters.
and your claim is that you tested it against btrfs on the same workload? Maybe you could post some specific numbers from running command from this thread? https://www.reddit.com/r/zfs/comments/1i3yjpt/very_poor_perf...
> Hell even the compression algorithm that ZFS has uses/has access to (LZ4) is faster than what btrfs uses and with enough IO that matters.
lz4 compression rate was 2x vs 7x for zstd on my data (bunch of numbers), so I didn't see point of uzing lz4 compression at all because benefits are not large enough.