Untitled
Other urls found in this thread:
lkml.org
lkml.org
twitter.com
Linus is wrong, direct IO is faster than keeping the cache consistent.
This is the man who made the kernel for your OS, Jow Forums.
Didn't he semi-retire?
this, reading a single byte from my spinning seagate disk is much faster than reading a single byte in RAM.
No, silly. Direct IO refers to DAX/zero copy IO. The price of keeping the cache consistent basically means that you have software in the data transfer path.
It was only for a few months.
Linus is back baby
>FUCK YOU DAVE
BASED and linuspilled
buts its much faster to read and write to the cache!!!111
Read the next posts. Linus' point is that the cache doesn't get involved in the cases where DIO *is* used and *is* faster, yet Dave is insisting that the cache is bad.
That said, the page cache is still far, far slower than direct IO, and the gap is just getting wider and wider as nvme SSDs get faster and faster. PCIe 4 SSDs are just going to make this even more obvious - it's getting to the point where the only reason for having a page cache is to support mmap() and cheap systems with spinning rust storage.
fuck you Dave thats not what I said
That's my beef with relying on the page cache - the page cache is rapidly becoming a legacy structure that only serves to slow modern IO subsystems down. More and more we are going to bypass the page cache and push/pull data via DMA directly into user buffers because that's the only way we have enough CPU power in the systems to keep the storage we have fully utilised.
>Linus' point is that the cache doesn't get involved in the cases where DIO *is* used and *is* faster
I think you should read it again, not only is he calling bullshit on the original quote, but he continues to say that caches are useful for NVMes and nvdimms that are capable of direct IO.
You should really read Dave's answer: lkml.org
>Project Morpheus
They must be good friends. Now that I think of it, the only people I've told "fuck you" are those I consider my friends.
This. The page cache makes sense for spinning disks and most SSDs, but with the emergence of blazing fast NVMes that are capable of DMA and even DMA in parallel, it's quickly becoming a bottleneck.
Uh oh, seems like that break was not enough, still breaking cocs
Which is why we should disable it on those device.
kek
That's what the DAX project is intended for, but it's currently grinding to a halt.
Dave is a fag but so is Linus.
Will Linus let him add it as an option for People Who Know What They're Doing?
At least they're alive.
Inb4 "vacation" again
Hopefully anita will find a moment to reeducate linus once she's done fixing cyberpunk 2077
I think you should actually read
Even so it's still the case that DIO is slower than caching even on NVMes. It's a fair point about NVDIMMs, but again, if someone's actually using NVDIMM for something particularly performance sensitive, they're also using DIO exclusively, and in the rare case that they aren't you just flush the god damn cache, which is Linus' point.
You don't need to arbitrarily disable the cache for certain devices to have good performance. It's just needlessly complicating it when the infrastructure to deal with this already exists.
You can already enable DAX if you want, but file system level support for it is very limited yet (only supported on xfs IIRC).
I think the main issue about DAX, which I kind of agree with, is that it bypasses the standard block-level interface and introduces an entirely separate path in the kernel. This wouldn't be a problem in of itself, but when you add the different file systems on top of it is where it becomes iffy; all of them assumes that the page cache is present and in use.
>the guy is bitching about CoC violations literally right after this message
How long until the tranny brigade gets to Linus?
>I think you should actually read
I have already, Linus were forced to moderate himself. He's wrong about the current state of NVMe disks though.
>Even so it's still the case that DIO is slower than caching even on NVMes.
It depends on
1) the NVMe disk in question, optane disks and flash memory disks are a lot faster than NAND-based SSDs for example
2) the amount of data being transfered, modern NVMe disks support IO operations that involve DMA transfers of several hundred megabytes simultaneously, which easily exceeds whatever can fit in the cache
3) Keeping the cache consistent when NVMe supports parallel IO operation is pain, and goes directly against the design of NVMe
>if someone's actually using NVDIMM for something particularly performance sensitive, they're also using DIO exclusively
Only if they use pmem interface, if you go through the block interface you need DAX which has limited support at the moment.
>in the rare case that they aren't you just flush the god damn cache, which is Linus' point.
This involves software in the data path, as well as potential for cache thrashing.
>You don't need to arbitrarily disable the cache for certain devices to have good performance.
You're not "disabling" the cache, you're simply bypassing it. Seriously, look up DAX.
There is nothing to talk about - memory access is indeed orders of magnitude faster that any IO, so the cache will be relevant for any IO device, no matter which technology it is using.
>memory access is indeed orders of magnitude faster that any IO
"IO" in this regard is DMA, which is direct memory access. Keep in mind that an NVMe drive is capable of reading and writing several hundred MBs directly from memory.
>so the cache will be relevant for any IO device, no matter which technology it is using.
That's painfully wrong.
Not him, but we're not talking about memory caches here, but the so-called page cache, which is a software structure used in the block driver in the kernel and is designed around the fact that spinning disks have a considerable delay before they are able to start transferring, something that isn't true any more with fast NVMe drives.
LOL JUST CACHE EVERYTHING
Isn't that what we've been doing since ever ?
I can't wait for Quantum Caching
Bump
We have to be careful with how much data we process.
>you violated le coc
Serves Linus right, he ack'd off on the CoC in the first place.
>MOOOOOOOOOOOM LINUS HURTIE HURTIE MY FEELINGSIES
Can someone think of the fefes please.
>lkml.org
imma have to go with dave on this one.
cache appeared cause shit was slow, ideally you should not need cache cause DIO is fast enough.
but, i guess it all depends on the cost of caching, including the development costs for the software doing the caching and the RAM.
would you cache 500TB of frequently viewed youtube videos to feed them instantly?
would you DIO a 5TB db distributed between 50 branch offices?
>2019
>linus torvaldsonn being relevant at all
Lmao, glad to see Linus has still got it. You can only suppress a man's autism for so long, particularly someone with autism as intense as Linus--those "empathy" classes his daughter put him through were never going to do jack shit. I mean, the guy is practically a self-admitted sociopath (in other words, a typical Finn).
What a fucking pussy.
>"Linus, nobody can talk about direct IO without you screaming and
tossing all your toys out of the crib. If you can't be civil or you
find yourself writing a some condescending "caching 101" explanation
to someone who has spent the last 15+ years working with filesystems
and caches, then you're far better off not saying anything."
This is so fucking true, Linus is so full of himself, the guy is truly a cancer for the linux development community.
>to someone who has spent the last 15+ years working with filesystems
If someone has spent 15 years on it and still doesn't get it, then that person is surely incompetent
So he wasn't brainwashed by trannies and is still vulgar and offensive?
Thank G-d.
In this case it's pretty obvious that Linus hasn't paid attention the last 4-5 years.
Chinner is/was a prominent engineer at sillicon graphics and has dozen of patents under his name. he is also the main dev for XFS on linux since years.
he had to be re educated
it's me or your font rendering is mediocre?
>he is also the main dev for XFS on linux since years.
And is the main driving force behind DAX.
Pretty sure you only use the apostrophe like then when it's a word in plural possessive form, even if there's already an 's' at the end. Correct me if I'm wrong.
As long as he only talks like that towards white heterosexual cis males it's all good. If they're in any way conservative he'll even get brownie points.
I'm Norwegian so I'm not beyond making mistakes like that, you might be right, but I don't really know.
linus shall retire and linux shall be put under the governance of a 16 seats council that would be seated by at&t, cisco, intel, ibm, google, microsoft, tencent, ... the council would then have to elect/pick commissioners for each division to oversee (filesystems, core kernel (scheduler, memory manager, ...), network stacks, graphic stacks, ...). one man to rull-them-all is archaic and so middle-age.
Ja, men da har du ingen unnskyldning for det er jo eg og.
>the page cache is rapidly becoming a legacy structure
no it doesn't.
data locality is a solution to many problems... and nobody gives a shit about your nvme feelings.
you wouldn't say such cianigger shit if you knew that ssds are fast because of the onboard cache using dram.
pcie is a packet driven protocol and not directly communicating with the device.
what if your gpu or network card suddenly interrupts the transaction and starts filling the bus with packets?
Linus is a retard
linus may have failed as a father on his 3/4 children, but he won't fail on his firstborn.
Jeg er også norsk, og du tar feil angående page-cachen btw.
1) he's citing Dave's original email
2) The page cache isn't a hardware cache, it's maintained by software and when it was originally implemented it relied on the assumption at spinning disks had a start up overhead
>ssds are fast because of the onboard cache using dram.
There are RAM disks that have NVMe controllers as well, these are fast. Optane disks also have very low latency, without having onboard DRAM.
>pcie is a packet driven protocol and not directly communicating with the device.
While it is "packet" driven, in the sense that TLPs are sort-of like packets, PCIe operates on memory addressing. From a software perspective, it's all just memory accesses. I don't know what you mean by "not directly communicating with the device", but the CPU is more or less that by using MMIO for reading/writing to control registers, and the device is capable of accessing system memory directly through DMA.
>what if your gpu or network card suddenly interrupts the transaction and starts filling the bus with packets?
PCIe doesn't behave like IP/packet-switching, and should be regarded more like line switching tbqh. In PCIe, devices have credits to deal with flow control, as well as limitations on how many outstanding requests it is allowed to keep open. Not to mention that each point-to-point link is fully bidirectional, only in the case where multiple devices share the same upstream link would you see an effect on bandwidth.
Keep in mind that he's not a native English speaker. He's been in the US for some time but it's still easy to make mistakes.
så ka farsken sku me gjort uten havet, sku me bær båtan?
kek
Norfags begone.
I see those empathy classes really paid off for Linus, he's learning so fast.
HOLD NU KJEFT
>Jeg er også norsk, og du tar feil angående page-cachen btw.
Ja, jeg gjorde det og innser det nå. Med DMA i bildet blir saken litt annerledes, mtp. at det bare blir et steg unodvendig kopiering.
>Keep in mind that he's not a native English speaker. He's been in the US for some time but it's still easy to make mistakes.
He was talking about my post and I've only set my foot in the US for a few weeks.
It's perfectly fine to say Linus's point.
but Linus' point would imply Linus is a plural of something ending in s.
You shut your dirty shithole, Dave.
I am fucking tired of your shit.
Don't you think it's time to stop sending me fucking emails?
I DIDN'T SPEND 50000 DOLLARS ON EMPATHY CLASSES JUST SO I COULD GO AND KILL SOME ANNOYING FAGGOT WHO SPAMS MY EMAIL
No, Dave. You need to stop.
You are wasting
my time,
my energy,
my money,
my email server space.
You're a fucking disgrace, Dave.
I know people like you.
I grew up in a neighbourhood filled with retards like you.
You see, Dave,
here in Finland,
alcholism is a big, big problem.
Many of them even carry superstitious beliefs like the Sauna Fairy, not too mentioned all the rituals involving spirits without emetics in them. You know, the kind you die from in one night because you drink so much and go on a rampage murder spree killing windows users like yourself.
Just fuck off, Dave.
Get the fuck off my email server.
Get the fuck off this planet.
I heartily kek'd, but "here in finland" is incorrect, Linus lives in the US.
Linus wants you to believe he lives in the US so it's harder to find him.
kek
>L1
>L2
>L3
>ram
oh no no no
then
>DRAM on SSD
>NAND on SSD
or
>128MB cache on HDD
>Spinning rush on HDD
Linux is a bloated piece of shit. I hope systemd and microshit destroy it.
Based
>get told
>"m-muh CoC"
who could have seen that coming
How was he told, exactly? Linus had no arguments, other than yelling and being angry, and the last response from Linus shows how weak his argument is.
Just fuck off already, Dave.
>has patents
DAVE
FOR THE LAST TIME
DONT MAKE ME CANCEL MY EMPATHY CLASSES
holds? not a native speaker
linus's "MUH CACHES" has potential to become a Jow Forums approved meme.
>having patents is good thing
so sjw's are dave's now? im cool with that
DIO
>Linus had no arguments
arguments in favour of what? he only said that caching is important, dave only tried to damage control after he forgot about it and got told
In favour of the page cache for NVMe drives. See and
Joke's on you. He didn't make XNU (... did he?!)
The more Torvalds' emails I read the more I respect him.
>Yes, your last run of insulting rants on this topic resulted in me pointing out your CoC violations because you were unable to listen or discuss the subject matter in a civil manner. And you've started right where you left off last time....
Picture of Dave angrily tapping away at his laptop typing this reply.
>your COC violations
Holy fucking shit, is Dave a transgender? Cause he sure as fuck ain't no man.
BSD shills on suicide watch.
Just a reminder, linus wanted this, and thought he was sticking it to the Nazis by implementing COC like a fucking retard. He deserves it
Nah, kill yourself, Dave.
>long greasy homosexual hair intensifies