Ceph Crimson Weekly, 29 Oct 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: 2019-10-29 :: Crimson SeaStor OSD Weekly Meeting

Description

.

A

A

Did you say anything.

B

Do you attended the sister summit.

B

So I thought approved to later from my manager and the ticket sales and Edith already and but I'm sorry too, to see if I can buy the ticket on site Oh.

A

If you I, wish, you told you that I probably could have talked with avi I.

A

Think we can, if there isn't a problem in the official channel I. Probably there is a way to convince him to let you in but.

B

A

We have activity the manager of prison, a fair wage of Scylla.

B

How to connect him.

A

Hey I have a harvest Emily, you can say it in advocate it's in the system. Mailing list.

B

A

There we can see made, don't let me they know too important and we are important to sister, so I hope you will allow it.

A

I said hello: hey good morning.

C

Hello, hey guys.

D

Take me put everything.

C

Here, that's weak, I didn't I, didn't make much I.

C

Was reading the paper and the specs we gotta do to refresh my understanding of a session 13.

C

The this day is that, right, I I think I completely agree with with them that, even though we have a we have here, we need to pretend it a left ear that an overworked or try to to who walks the way that the FDA working works better in the case of the pending right always has a better performance.

C

Even we are. We are writing to and flash ways that are Akio or even the the first women the flesh without the FTL like soon during the flash.

C

All the open, channel, SSD and I also find that there are some ten simulators which we could be, which you could we could leverage it to on telling the behavior of uh of the flash with SF Kia or without it it could be helped away when we are prototyping, the OPA store.

E

There's good news on that front: I heard from Matt yesterday we ought to be able to get some prototype zone Ned as the NS SSDs, like real soon that he asked for an address to ship them to so I. Think we're or even we should happen weeks, I hope, although somehow these things always take longer than they're supposed to, but still.

C

Yeah, that's great I, also.

E

I'm gonna point out that yeah.

F

E

Believe that open channel is, if it's not dead precisely there are hyper scalars that are using it, but as an industry standard, it's not viable. The zone, native storage is where it's out or no zone namespace assistance. There we go.

C

So we're more likely targeting the don't, don't access the flashlight right.

E

There they're two different approaches to the same problem: it's just that this is ENS. One is the one that's standardized in nvme, and that seems to have one so good for us. We don't have to worry about the other total amorphous blob of whatever vendors want to do. We have like a standard right.

C

But the is this standard we're accepted by pipe made me make a potato. Yes,.

E

Yes, micron Western, Digital, Samsung and I forget the fourth. When I talk to at FMS are all reading devices. Now okay, open channel was dead, one was the NS.

C

I was just consuming the papers and the specs to intend him.

C

And also I updated, the system can play a little bit.

B

Space for the Crimson branch.

B

So many files and to meet error an adjuster trying to they fixed all the field ever no.

E

um One thing I want to suggest is: try to separate out like, for instance, do it one file at a time, birds it's for South context, but make that its own commit, because it's good just give me a whole bunch of like mechanical changes to a bunch of files. If you make out its own, commit it'll, be really really quick to review.

B

E

That make sense I.

B

Have two other declaration in in the manifest they declare the class self context. I have to remove that van, and that includes F contact start page in that that.

E

May not work in every case.

B

E

The reason for the pre declaration in most of those cases is not pop is not too polluted that that header with the Ceph context declaration. In those cases it's only passed by pointer, so it doesn't need the full implementation, just the just the declaration. So my suggestion is that you, you keep it pretty mechanical for files that are only part of clásicos. These simply switch it to namespace def common around the class declaration for files that are used by both use the top namespace.

B

In that file itself in.

E

Those files you just need to wrap that little class declaration in the namespace.

B

E

Just just right around the deck, the class or the fast production like it's just right now, class def contacts semicolon just wrap that in namespace F, colon colon column or if it's a file used also by alien or Princeton top namespace.

E

Does that make sense that way? It'll be you know, it'll always be correct. If it's, if it's included into the, if it's included in all three places, it'll get the right namespace in all three places. If it's only in classic, then don't worry about it doesn't matter. You just need to include this F common part or, if you prefer, you could just make it always use top to top namespace. It doesn't matter to me.

B

Crimson comment out H in that in those vows.

E

You could do that. You don't strip yeah you'll need to for any file that doesn't already have it. Oh yeah. You would need to do that, but that doesn't mean you don't also have to wrap the class declaration in the namespace declaration.

E

Like this I'll just put it into chat.

E

E

At the top, with.

B

E

Ignore that line.

B

That namespace top namespace comment is included in the self chrome. Is a comment on age right? That's.

E

That's how it should work yeah if we set the separate.

B

Okay, our self chrome is in comment on edge, not East and abscess context. Dot, H yeah.

E

It's possible, you could have gotten to work the other way, but it would have increased compile times and in a few places it would actually have caused bugs or you would have caused compile errors. Keifa does that make sense, you're looking quizzical.

B

Classic structure I'm.

C

Trying to follow for a year.

E

Essentially, what I'm suggesting is that yeah.

B

E

One of these things that's used in more than one place. We actually want to change the way they're used in classic OST, because otherwise this will be terribly confusing. We'll have symbols that exist both in the top, unqualified namespace and in deeper namespaces and they'll all be visible, which I think we very much don't want, and it's worth doing some work to of avoid all so honestly, this improves the code. We've done this in a few places already, just as a side effect of modifying the code. It usually improves readability.

C

Is contact it's already in the common name space? It's not currently.

E

B

E

B

The other direction directory, we all declare mission death the concept context, so you suggest that I remove the self contact cluster of contacts and include our seven crimson common age in that in, though the fails right.

E

Just so that you can get at the top namespace backrub right.

B

E

Just so that you can get at the top namespace macro, that's that's! What it's there for.

E

Does anyone have a better way of doing this I'm really not like a seat? Boswell's guy I mean that kind of damp, but there it would not surprise me if there were a cleaner, less awful way to do this. Yeah.

C

I, don't need to take text about it to figure reading by the way, a cleaner way to do it, but I would like postponed after the after the Meetup. Okay.

E

Well, that's fair! What we're talking about next week anyway, yeah.

C

Prepare for they are approaching a discussion on sister. That's the four most important I think.

E

Yeah, that's what we.

B

C

C

Know so far, but I again, I'll suggest we we we consume some some papers and and to beauty understanding over your own step, oppress perspective of design novices door. So we can have a bit of discussion.

B

Have the meeting and in November fifth right to this discussion? Yes, yeah.

E

So in in the meantime, you know, Keith is suggesting that we should all be reading a lot and coming.

B

E

Like ideas for the.

B

C

Even even if you don't have a opinion, you need to carry yourself ready to understand what what are the scope and what we are facing and where we are. So why do you would be finding Europe if it would be difficult for you to catch up with the discussion.

D

Hello last week, I spent a bunch of var against the sister repo sorry Network. Just can you hear me now.

A

A

You might want to remove the camera to turn the camera off. Maybe.

D

With last week, I sent a bunch of peers to the step, / sister Reaper. Basically, it's work somehow related to a writer that the main the main thing the evidence of temporary spawning cannot pass has been reworked to actually don't change the observable behavior of Afghan and and enwrapped. Actually, this comic is longer than the term the the important part. Still it's and it's it's research, the benefit we saw during a writer testing.

D

Also I was to the term deponent talk. We had on Monday now I'm doing slides for for the sisters dammit my talk, girl got accepted and beside of that also I'm doing a reading for the sake of discussions we have there. That's me.

A

Question Radek the I don't see the same for requesting sister itself.

A

Which one he is the future temporaries.

D

Have you seen similar patches on made increased.

A

D

Okay, I sent there those those patches, the chef, / sister Rita, for initial review for internal review. After getting through it, I'm going to send to sister death.

D

Well idea: I used to have to have a something something like a gift for Abby. We will need to. We need to make some discs some profile, that's related to the buffer, fit in efficient consumption of DPD, KY and I mean input buffer factory and also a spiracle integrations. A lot of potential modifications to sister.

C

We can treat change them for for both for the merger of the approaching the input buffer drug-free craft refractory, but because it's more conversa, then the future changes I believe.

D

Something like that right running the period that PA one of the linked one of the peers: okay, I got it already. 18 G in the chat window.

A

This one I saw.

C

Okay, how's it going.

A

Medic, you finished.

D

Hey I wanted to add the the rest is sent against the same Decemeber.

D

Other PRS are sent against the same Reaper there's a bunch of them. Yeah.

A

I was I, went from one to another.

A

A

You asked what I did yes.

C

I asked for ask me if you have any update.

A

Well, I'm mostly doing going over code and reading I I, don't know if you saw this this presentation. This is regarding this matches with what time said in the beginning. One second.

A

A

Was I think it? It was two weeks and was uploaded about two weeks ago and then match it, matches what you said the same.

A

A

E

Think that guy works with abu talib, probably same research, script.

A

Okay, interesting.

A

Only way, I'm also starting to work on measuring performance, they build a new version of system and trying to get the best measurement I can with regular ngon trains on me that try to see that I get the best line. That makes sense that I know how I'm I am testing.

A

List it I think I said that last week, I enlisted a the new intern. We have to help me testing. The problem is, is that he disappears. He comes only once or twice a week and this week II wasn't feeling well, so nothing much was done by him, yet so more or less. This is what what I'm doing and I'm nearing to have the parts of the shock console.

A

First, implementation for review for comments. I have some company I had some questions of regarding things I'm, seeing while developing this, but I'll save this for the face to face.

A

That's that's it for me. We.

C

Have up here for support in Compton.

C

A

Private branch I won't have time because I'm going to concentrate on reading for the meeting, I probably won't have time to have it published before the meeting, but I'll approach you with probably with questions during the with the automatic.

A

So that's it for me.

C

Now we can do some some dumb thing.

A

C

A

C

Interesting and.

F

This last week, I was this: PR is merged, it is a and now the criminal messenger is fully tested with async messenger.

F

Now, so, if we have further bugs, we can easily debug with with it and the country I'm I'm, working to enable native stack with messenger probe to and try to get some performance data compared with big stack and there's a lot of lab issues to install the operating system, adding a body of decay and still working on it currently I'm still unable to bind the NIC to to the kernel driver, I'm, still figuring it out.

F

C

So you are using the PMD driver right.

F

Yeah, okay, it's it's vfi! Oh okay!.

C

Yeah there's a couple ways to to use typically along with the with the Knicks, which is supported. That's a way, that's the way she started using. It I think yes for better performance, yeah.

F

C

D

Maybe not a question just a quick note on the DPD k, testing I think it's it's extremely important, not even because of the raw performance gain we could get because of switching to native stock. It's about verifying our the the very initial hypothesis we made about terms of problems in cross flow. As this context, mult switches discuss actually yeah because of because of using the POSIX tag, we have the very last.

D

This called interactions in entire Chrome OS d. It's actually it's quite an right now, more or less familiar, just let's say: ASIC messenger classical as ASIC messenger, combined with object stir and what is funny it's that we are. We are when you are our profiling, classical just messenger threat of classical OSD. You are you very similar of that similar IPC instruction per cycle ratio in comparison to what we have at the common criminal I'm I am Creek Emily I'm extremely curious.

D

How the IPC would like, after getting rid of those very last Cisco's we have because of the network interaction. It will be extremely important topic for also for for the Feast of discussion. I mean SPD K integration, to judge how important it really is where, when a stage we should, we should preceding it whether we want to focus on SPD K, even at the beginning phase, or maybe postpone it a bit I.

F

Think for a steep as pity kz as picketing already started to integrate systems SP k. So maybe we can leverage race work yeah. It's perfect.

D

C

Think another thing indicate could help is that let's memory copy, because we have full management of memory out with a native stack, but it's not a compatible with the public stack right exactly.

D

But well at the moment we are a good impression we are, we are in testing. We are mostly focused on very small chunks, for K mem copy will be problem for bigger for bigger things.

C

But the tallest eyes are the smaller, smaller size to use tip. You dub you service to its do a little bit of overhead right, because.

A

It weighs practically.

C

Moving the memory chunk from phone wire to a disk accumulates.

D

C

Any other questions or comments.

C

Next week we have quite, but it some of us will be in San, Francisco, I guess and we were not sure we will be able to to have the meeting by then.

C

A

F

C

Probably we will cancel the meeting next week and we'll continue on with meeting the week after next week.

C

Oh good yeah, good night.

F

A

C

E

A short update, um I've got the other context reference or at the outer context, patch up for PR waiting for their crimson perf thing to go through, but it sounds like we know what's wrong with her um generally still making slow progress in the crimson back for recovery stuff. That's a do.

C

You know damn I miss you I.

E

Did you guys hurry to stuff? That's all I have okay.

C

E

Generally, after context.

E

Yep, like I, said we are going to get DNS drives the next few weeks. That'll be good after context. Er should be pretty close to radio at least once crimson, perfect routes that I didn't break through put bands to working on back full recovery. It's not going quickly, but I am making progress. That's it.

C

C

Then go guys next week, I said mostly you up next week and see all of you in there the weak neck of the next week. Thank you.

A