diff options
Diffstat (limited to 'open_issues/performance/io_system')
| -rw-r--r-- | open_issues/performance/io_system/read-ahead.mdwn | 1176 |
1 files changed, 1176 insertions, 0 deletions
diff --git a/open_issues/performance/io_system/read-ahead.mdwn b/open_issues/performance/io_system/read-ahead.mdwn index d6a98070..710c746b 100644 --- a/open_issues/performance/io_system/read-ahead.mdwn +++ b/open_issues/performance/io_system/read-ahead.mdwn @@ -16,6 +16,9 @@ License|/fdl]]."]]"""]] # [[community/gsoc/project_ideas/disk_io_performance]] +# [[gnumach_page_cache_policy]] + + # 2011-02 [[Etenil]] has been working in this area. @@ -389,3 +392,1176 @@ License|/fdl]]."]]"""]] with appropriate frame size. Is that right? <youpi> question of taste, better ask on the list <mcsim> ok + + +## IRC, freenode, #hurd, 2012-06-09 + + <mcsim> hello. What fictitious pages in gnumach are needed for? + <mcsim> I mean why real page couldn't be grabbed straight, but in sometimes + fictitious page is grabbed first and than converted to real? + <braunr> mcsim: iirc, fictitious pages are needed by device pagers which + must comply with the vm pager interface + <braunr> mcsim: specifically, they must return a vm_page structure, but + this vm_page describes device memory + <braunr> mcsim: and then, it must not be treated like normal vm_page, which + can be added to page queues (e.g. page cache) + + +## IRC, freenode, #hurd, 2012-06-22 + + <mcsim> braunr: Ah. Patch for large storages introduced new callback + pager_notify_evict. User had to define this callback on his own as + pager_dropweak, for instance. But neal's patch change this. Now all + callbacks could have any name, but user defines structure with pager ops + and supplies it in pager_create. + <mcsim> So, I just changed notify_evict to confirm it to new style. + <mcsim> braunr: I want to changed interface of mo_change_attributes and + test my changes with real partitions. For both these I have to update + ext2fs translator, but both partitions I have are bigger than 2Gb, that's + why I need apply this patch.z + <mcsim> But what to do with mo_change_attributes? I need somehow inform + kernel about page fault policy. + <mcsim> When I change mo_ interface in kernel I have to update all programs + that use this interface and ext2fs is one of them. + + <mcsim> braunr: Who do you think better to inform kernel about fault + policy? At the moment I've added fault_strategy parameter that accepts + following strategies: randow, sequential with single page cluster, + sequential with double page cluster and sequential with quad page + cluster. OSF/mach has completely another interface of + mo_change_attributes. In OSF/mach mo_change_attributes accepts structure + of parameter. This structure could have different formats depending o + <mcsim> This rpc could be useful because it is not very handy to update + mo_change_attributes for kernel, for hurd libs and for glibc. Instead of + this kernel will accept just one more structure format. + <braunr> well, like i wrote on the mailing list several weeks ago, i don't + think the policy selection is of concern currently + <braunr> you should focus on the implementation of page clustering and + readahead + <braunr> concerning the interface, i don't think it's very important + <braunr> also, i really don't like the fact that the policy is per object + <braunr> it should be per map entry + <braunr> i think it mentioned that in my mail too + <braunr> i really think you're wasting time on this + <braunr> http://lists.gnu.org/archive/html/bug-hurd/2012-04/msg00064.html + <braunr> http://lists.gnu.org/archive/html/bug-hurd/2012-04/msg00029.html + <braunr> mcsim: any reason you completely ignored those ? + <mcsim> braunr: Ok. I'll do clustering for map entries. + <braunr> no it's not about that either :/ + <braunr> clustering is grouping several pages in the same transfer between + kernel and pager + <braunr> the *policy* is held in map entries + <antrik> mcsim: I'm not sure I properly understand your question about the + policy interface... but if I do, it's IMHO usually better to expose + individual parameters as RPC arguments explicitly, rather than hiding + them in an opaque structure... + <antrik> (there was quite some discussion about that with libburn guy) + <mcsim> antrik: Following will be ok? kern_return_t vm_advice(map, address, + length, advice, cluster_size) + <mcsim> Where advice will be either random or sequential + <antrik> looks fine to me... but then, I'm not an expert on this stuff :-) + <antrik> perhaps "policy" would be clearer than "advice"? + <mcsim> madvise has following prototype: int madvise(void *addr, size_t + len, int advice); + <mcsim> hmm... looks like I made a typo. Or advi_c_e is ok too? + <antrik> advise is a verb; advice a noun... there is a reason why both + forms show up in the madvise prototype :-) + <mcsim> so final variant should be kern_return_t vm_advise(map, address, + length, policy, cluster_size)? + <antrik> mcsim: nah, you are probably right that its better to keep + consistency with madvise, even if the name of the "advice" parameter + there might not be ideal... + <antrik> BTW, where does cluster_size come from? from the filesystem? + <antrik> I see merits both to naming the parameter "policy" (clearer) or + "advice" (more consistent) -- you decide :-) + <mcsim> antrik: also there is variant strategy, like with inheritance :) + I'll choose advice for now. + <mcsim> What do you mean under "where does cluster_size come from"? + <antrik> well, madvise doesn't have this parameter; so the value must come + from a different source? + <mcsim> in madvise implementation it could fixed value or somehow + calculated basing on size of memory range. In OSF/mach cluster size is + supplied too (via mo_change_attributes). + <antrik> ah, so you don't really know either :-) + <antrik> well, my guess is that it is derived from the cluster size used by + the filesystem in question + <antrik> so for us it would always be 4k for now + <antrik> (and thus you can probably leave it out alltogether...) + <antrik> well, fatfs can use larger clusters + <antrik> I would say, implement it only if it's very easy to do... if it's + extra effort, it's probably not worth it + <mcsim> There is sense to make cluster size bigger for ext2 too, since most + likely consecutive clusters will be within same group. + <mcsim> But anyway I'll handle this later. + <antrik> well, I don't know what cluster_size does exactly; but by the + sound of it, I'd guess it makes an assumption that it's *always* better + to read in this cluster size, even for random access -- which would be + simply wrong for 4k filesystem clusters... + <antrik> BTW, I agree with braunr that madvice() is optional -- it is way + way more important to get readahead working as a default policy first + + +## IRC, freenode, #hurd, 2012-07-01 + + <mcsim> youpi: Do you think you could review my code? + <youpi> sure, just post it to the list + <youpi> make sure to break it down into logical pieces + <mcsim> youpi: I pushed it my branch at gnumach repository + <mcsim> youpi: or it is still better to post changes to list? + <youpi> posting to the list would permit feedback from other people too + <youpi> mcsim: posix distinguishes normal, sequential and random + <youpi> we should probably too + <youpi> the system call should probably be named "vm_advise", to be a verb + like allocate etc. + <mcsim> youpi: ok. A have a talk with antrik regarding naming, I'll change + this later because compiling of glibc take a lot of time. + <youpi> mcsim: I find it odd that vm_for_every_page allocates non-existing + pages + <youpi> there should probably be at least a flag to request it or not + <mcsim> youpi: normal policy is synonym to default. And this could be + treated as either random or sequential, isn't it? + <braunr> mcsim: normally, no + <youpi> yes, the normal policy would be the default + <youpi> it doesn't mean random or sequential + <youpi> it's just to be a compromise between both + <youpi> random is meant to make no read-ahead, since that'd be spurious + anyway + <youpi> while by default we should make readahead + <braunr> and sequential makes even more aggressive readahead, which usually + implies a greater number of pages to fetch + <braunr> that's all + <youpi> yes + <youpi> well, that part is handled by the cluster_size parameter actually + <braunr> what about reading pages preceding the faulted paged ? + <mcsim> Shouldn't sequential clean some pages (if they, for example, are + not precious) that are placed before fault page? + <braunr> ? + <youpi> that could make sense, yes + <braunr> you lost me + <youpi> and something that you wouldn't to with the normal policy + <youpi> braunr: clear what has been read previously + <braunr> ? + <youpi> since the access is supposed to be sequential + <braunr> oh + <youpi> the application will proabably not re-read what was already read + <braunr> you mean to avoid caching it ? + <youpi> yes + <braunr> inactive memory is there for that + <youpi> while with the normal policy you'd assume that the application + might want to go back etc. + <youpi> yes, but you can help it + <braunr> yes + <youpi> instead of making other pages compete with it + <braunr> but then, it's for precious pages + <youpi> I have to say I don't know what a precious page it + <youpi> s + <youpi> does it mean dirty pages? + <braunr> no + <braunr> precious means cached pages + <braunr> "If precious is FALSE, the kernel treats the data as a temporary + and may throw it away if it hasn't been changed. If the precious value is + TRUE, the kernel treats its copy as a data repository and promises to + return it to the manager; the manager may tell the kernel to throw it + away instead by flushing and not cleaning the data" + <braunr> hm no + <braunr> precious means the kernel must keep it + <mcsim> youpi: According to vm_for_every_page. What kind of flag do you + suppose? If object is internal, I suppose not to cross the bound of + object, setting in_end appropriately in vm_calculate_clusters. + <mcsim> If object is external we don't know its actual size, so we should + make mo request first. And for this we should create fictitious pages. + <braunr> mcsim: but how would you implement this "cleaning" with sequential + ? + <youpi> mcsim: ah, ok, I thought you were allocating memory, but it's just + fictitious pages + <youpi> comment "Allocate a new page" should be fixed :) + <mcsim> braunr: I don't now how I will implement this specifically (haven't + tried yet), but I don't think that this is impossible + <youpi> braunr: anyway it's useful as an example where normal and + sequential would be different + <braunr> if it can be done simply + <braunr> because i can see more trouble than gains in there :) + <mcsim> braunr: ok :) + <braunr> mcsim: hm also, why fictitious pages ? + <braunr> fictitious pages should normally be used only when dealing with + memory mapped physically which is not real physical memory, e.g. device + memory + <mcsim> but vm_fault could occur when object represent some device memory. + <braunr> that's exactly why there are fictitious pages + <mcsim> at the moment of allocating of fictitious page it is not know what + backing store of object is. + <braunr> really ? + <braunr> damn, i've got used to UVM too much :/ + <mcsim> braunr: I said something wrong? + <braunr> no no + <braunr> it's just that sometimes, i'm confusing details about the various + BSD implementations i've studied + <braunr> out-of-gsoc-topic question: besides network drivers, do you think + we'll have other drivers that will run in userspace and have to implement + memory mapping ? like framebuffers ? + <braunr> or will there be a translation layer such as storeio that will + handle mapping ? + <youpi> framebuffers typically will, yes + <youpi> that'd be antrik's work on drm + <braunr> hmm + <braunr> ok + <youpi> mcsim: so does the implementation work, and do you see performance + improvement? + <mcsim> youpi: I haven't tested it yet with large ext2 :/ + <mcsim> youpi: I'm going to finish now moving of ext2 to new interface, + than other translators in hurd repository and than finish memory policies + in gnumach. Is it ok? + <youpi> which new interface? + <mcsim> Written by neal. I wrote some temporary code to make ext2 work with + it, but I'm going to change this now. + <youpi> you mean the old unapplied patch? + <mcsim> yes + <youpi> did you have a look at Karim's work? + <youpi> (I have to say I never found the time to check how it related with + neal's patch) + <mcsim> I found only his work in kernel. I didn't see his work in applying + of neal's patch. + <youpi> ok + <youpi> how do they relate with each other? + <youpi> (I have never actually looked at either of them :/) + <mcsim> his work in kernel and neal's patch? + <youpi> yes + <mcsim> They do not correlate with each other. + <youpi> ah, I must be misremembering what each of them do + <mcsim> in kam's patch was changes to support sequential reading in reverse + order (as in OSF/Mach), but posix does not support such behavior, so I + didn't implement this either. + <youpi> I can't find the pointer to neal's patch, do you have it off-hand? + <mcsim> http://comments.gmane.org/gmane.os.hurd.bugs/351 + <youpi> thx + <youpi> I think we are not talking about the same patch from Karim + <youpi> I mean lists.gnu.org/archive/html/bug-hurd/2010-06/msg00023.html + <mcsim> I mean this patch: + http://lists.gnu.org/archive/html/bug-hurd/2010-06/msg00024.html + <mcsim> Oh. + <youpi> ok + <mcsim> seems, this is just the same + <youpi> yes + <youpi> from a non-expert view, I would have thought these patches play + hand in hand, do they really? + <mcsim> this patch is completely for kernel and neal's one is completely + for libpager. + <youpi> i.e. neal's fixes libpager, and karim's fixes the kernel + <mcsim> yes + <youpi> ending up with fixing the whole path? + <youpi> AIUI, karim's patch will be needed so that your increased readahead + will end up with clustered page request? + <mcsim> I will not use kam's patch + <youpi> is it not needed to actually get pages in together? + <youpi> how do you tell libpager to fetch pages together? + <youpi> about the cluster size, I'd say it shouldn't be specified at + vm_advise() level + <youpi> in other OSes, it is usually automatically tuned + <youpi> by ramping it up to a maximum readahead size (which, however, could + be specified) + <youpi> that's important for the normal policy, where there are typically + successive periods of sequential reads, but you don't know in advance for + how long + <mcsim> braunr said that there are legal issues with his code, so I cannot + use it. + <braunr> did i ? + <braunr> mcsim: can you give me a link to the code again please ? + <youpi> see above :) + <braunr> which one ? + <youpi> both + <youpi> they only differ by a typo + <braunr> mcsim: i don't remember saying that, do you have any link ? + <braunr> or log ? + <mcsim> sorry, can you rephrase "ending up with fixing the whole path"? + <mcsim> cluster_size in vm_advise also could be considered as advise + <braunr> no + <braunr> it must be the third time we're talking about this + <youpi> mcsim: I mean both parts would be needed to actually achieve + clustered i/o + <braunr> again, why make cluster_size a per object attribute ? :( + <youpi> wouldn't some objects benefit from bigger cluster sizes, while + others wouldn't? + <youpi> but again, I believe it should rather be autotuned + <youpi> (for each object) + <braunr> if we merely want posix compatibility (and for a first attempt, + it's quite enough), vm_advise is good, and the kernel selects the + implementation (and thus the cluster sizes) + <braunr> if we want finer grained control, perhaps a per pager cluster_size + would be good, although its efficiency depends on several parameters + <braunr> (e.g. where the page is in this cluster) + <braunr> but a per object cluster size is a large waste of memory + considering very few applications (if not none) would use the "feature" + .. + <braunr> (if any*) + <youpi> there must be a misunderstanding + <youpi> why would it be a waste of memory? + <braunr> "per object" + <youpi> so? + <braunr> there can be many memory objects in the kernel + <youpi> so? + <braunr> so such an overhead must be useful to accept it + <youpi> in my understanding, a cluster size per object is just a mere + integer for each object + <youpi> what overhead? + <braunr> yes + <youpi> don't we have just thousands of objects? + <braunr> for now + <braunr> remember we're trying to remove the page cache limit :) + <youpi> that still won't be more than tens of thousands of objects + <youpi> times an integer + <youpi> that's completely neglectible + <mcsim> braunr: Strange, Can't find in logs. Weird things are happening in + my memory :/ Sorry. + <braunr> mcsim: i'm almost sure i never said that :/ + <braunr> but i don't trust my memory too much either + <braunr> youpi: depends + <youpi> mcsim: I mean both parts would be needed to actually achieve + clustered i/o + <mcsim> braunr: I made I call vm_advise that applies policy to memory range + (vm_map_entry to be specific) + <braunr> mcsim: good + <youpi> actually the cluster size should even be per memory range + <mcsim> youpi: In this sense, yes + <youpi> k + <mcsim> sorry, Internet connection lags + <braunr> when changing a structure used to create many objects, keep in + mind one thing + <braunr> if its size gets larger than a threshold (currently, powers of + two), the cache used by the slab allocator will allocate twice the + necessary amount + <youpi> sure + <braunr> this is the case with most object caching allocators, although + some can have specific caches for common sizes such as 96k which aren't + powers of two + <braunr> anyway, an integer is negligible, but the final structure size + must be checked + <braunr> (for both 32 and 64 bits) + <mcsim> braunr: ok. + <mcsim> But I didn't understand what should be done with cluster size in + vm_advise? Should I delete it? + <braunr> to me, the cluster size is a pager property + <youpi> to me, the cluster size is a map property + <braunr> whereas vm_advise indicates what applications want + <youpi> you could have several process accessing the same file in different + ways + <braunr> youpi: that's why there is a policy + <youpi> isn't cluster_size part of the policy? + <braunr> but if the pager abilities are limited, it won't change much + <braunr> i'm not sure + <youpi> cluster_size is the amount of readahead, isn't it? + <braunr> no, it's the amount of data in a single transfer + <mcsim> Yes, it is. + <braunr> ok, i'll have to check your code + <youpi> shouldn't transfers permit unbound amounts of data? + <mcsim> braunr: than I misunderstand what readahead is + <braunr> well then cluster size is per policy :) + <braunr> e.g. random => 0, normal => 3, sequential => 15 + <braunr> why make it per map entry ? + <youpi> because it depends on what the application doezs + <braunr> let me check the code + <youpi> if it's accessing randomly, no need for big transfers + <youpi> just page transfers will be fine + <youpi> if accessing sequentially, rather use whole MiB of transfers + <youpi> and these behavior can be for the same file + <braunr> mcsim: the call is vm_advi*s*e + <braunr> mcsim: the call is vm_advi_s_e + <braunr> not advice + <youpi> yes, he agreed earlier + <braunr> ok + <mcsim> cluster_size is the amount of data that I try to read at one time. + <mcsim> at singe mo_data_request + <mcsim> *single + <youpi> which, to me, will depend on the actual map + <braunr> ok so it is the transfer size + <youpi> and should be autotuned, especially for normal behavior + <braunr> youpi: it makes no sense to have both the advice and the actual + size per map entry + <youpi> to get big readahead with all apps + <youpi> braunr: the size is not only dependent on the advice, but also on + the application behavior + <braunr> youpi: how does this application tell this ? + <youpi> even for sequential, you shouldn't necessarily use very big amounts + of transfers + <braunr> there is no need for the advice if there is a cluster size + <youpi> there can be, in the case of sequential, as we said, to clear + previous pages + <youpi> but otherwise, indeed + <youpi> but for me it's the converse + <youpi> the cluster size should be tuned anyway + <braunr> and i'm against giving the cluster size in the advise call, as we + may want to prefetch previous data as well + <youpi> I don't see how that collides + <braunr> well, if you consider it's the transfer size, it doesn't + <youpi> to me cluster size is just the size of a window + <braunr> if you consider it's the amount of pages following a faulted page, + it will + <braunr> also, if your policy says e.g. "3 pages before, 10 after", and + your cluster size is 2, what happens ? + <braunr> i would find it much simpler to do what other VM variants do: + compute the I/O sizes directly from the policy + <youpi> don't they autotune, and use the policy as a maximum ? + <braunr> depends on the implementations + <youpi> ok, but yes I agree + <youpi> although casting the size into stone in the policy looks bogus to + me + <braunr> but making cluster_size part of the kernel interface looks way too + messy + <braunr> it is + <braunr> that's why i would have thought it as part of the pager properties + <braunr> the pager is the true component besides the kernel that is + actually involved in paging ... + <youpi> well, for me the flexibility should still be per application + <youpi> by pager you mean the whole pager, not each file, right? + <braunr> if a pager can page more because e.g. it's a file system with big + block sizes, why not fetch more ? + <braunr> yes + <braunr> it could be each file + <braunr> but only if we have use for it + <braunr> and i don't see that currently + <youpi> well, posix currently doesn't provide a way to set it + <youpi> so it would be useless atm + <braunr> i was thinking about our hurd pagers + <youpi> could we perhaps say that the policy maximum could be a fraction of + available memory? + <braunr> why would we want that ? + <youpi> (total memory, I mean) + <youpi> to make it not completely cast into stone + <youpi> as have been in the past in gnumach + <braunr> i fail to understand :/ + <youpi> there must be a misunderstanding then + <youpi> (pun not intended) + <braunr> why do you want to limit the policy maximum ? + <youpi> how to decide it? + <braunr> the pager sets it + <youpi> actually I don't see how a pager could decide it + <youpi> on what ground does it make the decision? + <youpi> readahead should ideally be as much as 1MiB + <braunr> 02:02 < braunr> if a pager can page more because e.g. it's a file + system with big block sizes, why not fetch more ? + <braunr> is the example i have in mind + <braunr> otherwise some default values + <youpi> that's way smaller than 1MiB, isn't it? + <braunr> yes + <braunr> and 1 MiB seems a lot to me :) + <youpi> for readahead, not really + <braunr> maybe for sequential + <youpi> that's what we care about! + <braunr> ah, i thought we cared about normal + <youpi> "as much as 1MiB", I said + <youpi> I don't mean normal :) + <braunr> right + <braunr> but again, why limit ? + <braunr> we could have 2 or more ? + <youpi> at some point you don't get more efficiency + <youpi> but eat more memory + <braunr> having the pager set the amount allows us to easily adjust it over + time + <mcsim> braunr: Do you think that readahead should be implemented in + libpager? + <youpi> than needed + <braunr> mcsim: no + <braunr> mcsim: err + <braunr> mcsim: can't answer + <youpi> mcsim: do you read the log of what you have missed during + disconnection? + <braunr> i'm not sure about what libpager does actually + <mcsim> yes + <braunr> for me it's just mutualisation of code used by pagers + <braunr> i don't know the details + <braunr> youpi: yes + <braunr> youpi: that's why we want these values not hardcoded in the kernel + <braunr> youpi: so that they can be adjusted by our shiny user space OS + <youpi> (btw apparently linux uses minimum 16k, maximum 128 or 256k) + <braunr> that's more reasonable + <youpi> that's just 4 times less :) + <mcsim> braunr: You say that pager should decide how much data should be + read ahead, but each pager can't implement it on it's own as there will + be too much overhead. So the only way is to implement this in libpager. + <braunr> mcsim: gni ? + <braunr> why couldn't they ? + <youpi> mcsim: he means the size, not the actual implementation + <youpi> the maximum size, actually + <braunr> actually, i would imagine it as the pager giving per policy + parameters + <youpi> right + <braunr> like how many before and after + <youpi> I agree, then + <braunr> the kernel could limit, sure, to avoid letting pagers use + completely insane values + <youpi> (and that's just a max, the kernel autotunes below that) + <braunr> why not + <youpi> that kernel limit could be a fraction of memory, then? + <braunr> it could, yes + <braunr> i see what you mean now + <youpi> mcsim: did you understand our discussion? + <youpi> don't hesitate to ask for clarification + <mcsim> I supposed cluster_size to be such parameter. And advice will help + to interpret this parameter (whether data should be read after fault page + or some data should be cleaned before) + <youpi> mcsim: we however believe that it's rather the pager than the + application that would tell that + <youpi> at least for the default values + <youpi> posix doesn't have a way to specify it, and I don't think it will + in the future + <braunr> and i don't think our own hurd-specific programs will need more + than that + <braunr> if they do, we can slightly change the interface to make it a per + object property + <braunr> i've checked the slab properties, and it seems we can safely add + it per object + <braunr> cf http://www.sceen.net/~rbraun/slabinfo.out + <braunr> so it would still be set by the pager, but if depending on the + object, the pager could set different values + <braunr> youpi: do you think the pager should just provide one maximum size + ? or per policy sizes ? + <youpi> I'd say per policy size + <youpi> so people can increase sequential size like crazy when they know + their sequential applications need it, without disturbing the normal + behavior + <braunr> right + <braunr> so the last decision is per pager or per object + <braunr> mcsim: i'd say whatever makes your implementation simpler :) + <mcsim> braunr: how kernel knows that object are created by specific pager? + <braunr> that's the kind of things i'm referring to with "whatever makes + your implementation simpler" + <braunr> but usually, vm_objects have an ipc port and some properties + relatedto their pagers + <braunr> -usually + <braunr> the problem i had in mind was the locking protocol but our spin + locks are noops, so it will be difficult to detect deadlocks + <mcsim> braunr: and for every policy there should be variable in vm_object + structure with appropriate cluster_size? + <braunr> if you want it per object, yes + <braunr> although i really don't think we want it + <youpi> better keep it per pager for now + <braunr> let's imagine youpi finishes his 64-bits support, and i can + successfully remove the page cache limit + <braunr> we'd jump from 1.8 GiB at most to potentially dozens of GiB of RAM + <braunr> and 1.8, mostly unused + <braunr> to dozens almost completely used, almost all the times for the + most interesting use cases + <braunr> we may have lots and lots of objects to keep around + <braunr> so if noone really uses the feature ... there is no point + <youpi> but also lots and lots of memory to spend on it :) + <youpi> a lot of objects are just one page, but a lof of them are not + <braunr> sure + <braunr> we wouldn't be doing that otherwise :) + <braunr> i'm just saying there is no reason to add the overhead of several + integers for each object if they're simply not used at all + <braunr> hmm, 64-bits, better page cache, clustered paging I/O :> + <braunr> (and readahead included in the last ofc) + <braunr> good night ! + <mcsim> than, probably, make system-global max-cluster_size? This will save + some memory. Also there is usually no sense in reading really huge chunks + at once. + <youpi> but that'd be tedious to set + <youpi> there are only a few pagers, that's no wasted memory + <youpi> the user being able to set it for his own pager is however a very + nice feature, which can be very useful for databases, image processing, + etc. + <mcsim> In conclusion I have to implement following: 3 memory policies per + object and per vm_map_entry. Max cluster size for every policy should be + set per pager. + <mcsim> So, there should be 2 system calls for setting memory policy and + one for setting cluster sizes. + <mcsim> Also amount of data to transfer should be tuned automatically by + every page fault. + <mcsim> youpi: Correct me, please, if I'm wrong. + <youpi> I believe that's what we ended up to decide, yes + + +## IRC, freenode, #hurd, 2012-07-02 + + <braunr> is it safe to say that all memory objects implemented by external + pagers have "file" semantics ? + <braunr> i wonder if the current memory manager interface is suitable for + device pagers + <mcsim> braunr: What does "file" semantics mean? + <braunr> mcsim: anonymous memory doesn't have the same semantics as a file + for example + <braunr> anonymous memory that is discontiguous in physical memory can be + contiguous in swap + <braunr> and its location can change with time + <braunr> whereas with a memory object, the data exchanged with pagers is + identified with its offset + <braunr> in (probably) all other systems, this way of specifying data is + common to all files, whatever the file system + <braunr> linux uses the struct vm_file name, while in BSD/Solaris they are + called vnodes (the link between a file system inode and virtual memory) + <braunr> my question is : can we implement external device pagers with the + current interface, or is this interface really meant for files ? + <braunr> also + <braunr> mcsim: something about what you said yesterday + <braunr> 02:39 < mcsim> In conclusion I have to implement following: 3 + memory policies per object and per vm_map_entry. Max cluster size for + every policy should be set per pager. + <braunr> not per object + <braunr> one policy per map entry + <braunr> transfer parameters (pages before and after the faulted page) per + policy, defined by pagers + <braunr> 02:39 < mcsim> So, there should be 2 system calls for setting + memory policy and one for setting cluster sizes. + <braunr> adding one call for vm_advise is good because it mirrors the posix + call + <braunr> but for the parameters, i'd suggest changing an already existing + call + <braunr> not sure which one though + <mcsim> braunr: do you know how mo_change_attributes implemented in + OSF/Mach? + <braunr> after a quick reading of the reference manual, i think i + understand why they made it per object + <braunr> mcsim: no + <braunr> did they change the call to include those paging parameters ? + <mcsim> it accept two parameters: flavor and pointer to structure with + parameters. + <mcsim> flavor determines semantics of structure with parameters. + <mcsim> + http://www.darwin-development.org/cgi-bin/cvsweb/osfmk/src/mach_kernel/vm/memory_object.c?rev=1.1 + <mcsim> structure can have 3 different views and what exect view will be is + determined by value of flavor + <mcsim> So, I thought about implementing similar call that could be used + for various purposes. + <mcsim> like ioctl + <braunr> "pointer to structure with parameters" <= which one ? + <braunr> mcsim: don't model anything anywhere like ioctl please + <mcsim> memory_object_info_t attributes + <braunr> ioctl is the very thing we want NOT to have on the hurd + <braunr> ok attributes + <braunr> and what are the possible values of flavour, and what kinds of + attributes ? + <mcsim> and then appears something like this on each case: behave = + (old_memory_object_behave_info_t) attributes; + <braunr> ok i see + <mcsim> flavor could be OLD_MEMORY_OBJECT_BEHAVIOR_INFO, + MEMORY_OBJECT_BEHAVIOR_INFO, MEMORY_OBJECT_PERFORMANCE_INFO etc + <braunr> i don't really see the point of flavour here, other than + compatibility + <braunr> having attributes is nice, but you should probably add it as a + call parameter, not inside a structure + <braunr> as a general rule, we don't like passing structures too much + to/from the kernel, because handling them with mig isn't very clean + <mcsim> ok + <mcsim> What policy parameters should be defined by pager? + <braunr> i'd say number of pages to page-in before and after the faulted + page + <mcsim> Only pages before and after the faulted page? + <braunr> for me yes + <braunr> youpi might have different things in mind + <braunr> the page cleaning in sequential mode is something i wouldn't do + <braunr> 1/ applications might want data read sequentially to remain in the + cache, for other sequential accesses + <braunr> 2/ applications that really don't want to cache anything should + use O_DIRECT + <braunr> 3/ it's complicated, and we're in july + <braunr> i'd rather have a correct and stable result than too many unused + features + <mcsim> braunr: MADV_SEQUENTIAL Expect page references in sequential order. + (Hence, pages in the given range can be aggressively read ahead, and may + be freed soon after they are accessed.) + <mcsim> this is from linux man + <mcsim> braunr: Can I at least make keeping in mind that it could be + implemented? + <mcsim> I mean future rpc interface + <mcsim> braunr: From behalf of kernel pager is just a port. + <mcsim> That's why it is not clear for me how I can make in kernel + per-pager policy + <braunr> mcsim: you can't + <braunr> 15:19 < braunr> after a quick reading of the reference manual, i + think i understand why they made it per object + <braunr> + http://pubs.opengroup.org/onlinepubs/009695399/functions/posix_madvise.html + <braunr> POSIX_MADV_SEQUENTIAL + <braunr> Specifies that the application expects to access the specified + range sequentially from lower addresses to higher addresses. + <braunr> linux might free pages after their access, why not, but this is + entirely up to the implementation + <mcsim> I know, when but applications might want data read sequentially to + remain in the cache, for other sequential accesses this kind of access + could be treated rather normal or random + <braunr> we can do differently + <braunr> mcsim: no + <braunr> sequential means the access will be sequential + <braunr> so aggressive readahead (e.g. 0 pages before, many after), should + be used + <braunr> for better performance + <braunr> from my pov, it has nothing to do with caching + <braunr> i actually sometimes expect data to remain in cache + <braunr> e.g. before playing a movie from sshfs, i sometimes prefetch it + using dd + <braunr> then i use mplayer + <braunr> i'd be very disappointed if my data didn't remain in the cache :) + <mcsim> At least these pages could be placed into inactive list to be first + candidates for pageout. + <braunr> that's what will happen by default + <braunr> mcsim: if we need more properties for memory objects, we'll adjust + the call later, when we actually implement them + <mcsim> so, first call is vm_advise and second is changed + mo_change_attributes? + <braunr> yes + <mcsim> there will appear 3 new parameters in mo_c_a: policy, pages before + and pages after? + <mcsim> braunr: With vm_advise I didn't understand one thing. This call is + defined in defs file, so that should mean that vm_advise is ordinal rpc + call. But on the same time it is defined as syscall in mach internals (in + mach_trap_table). + <braunr> mcsim: what ? + <braunr> were is it "defined" ? (it doesn't exit in gnumach currently) + <mcsim> Ok, let consider vm_map + <mcsim> I define it both in mach_trap_table and in defs file. + <mcsim> But why? + <braunr> uh ? + <braunr> let me see + <mcsim> Why defining in defs file is not enough? + <mcsim> and previous question: there will appear 3 new parameters in + mo_c_a: policy, pages before and pages after? + <braunr> mcsim: give me the exact file paths please + <braunr> mcsim: we'll discuss the new parameters after + <mcsim> kern/syscall_sw.c + <braunr> right i see + <mcsim> here mach_trap_table in defined + <braunr> i think they're not used + <braunr> they were probably introduced for performance + <mcsim> and ./include/mach/mach.defs + <braunr> don't bother adding vm_advise as a syscall + <braunr> about the parameters, it's a bit more complicated + <braunr> you should add 6 parameters + <braunr> before and after, for the 3 policies + <braunr> but + <braunr> as seen in the posix page, there could be more policies .. + <braunr> ok forget what i said, it's stupid + <braunr> yes, the 3 parameters you had in mind are correct + <braunr> don't forget a "don't change" value for the policy though, so the + kernel ignores the before/after values if we don't want to change that + <mcsim> ok + <braunr> mcsim: another reason i asked about "file semantics" is the way we + handle the cache + <braunr> mcsim: file semantics imply data is cached, whereas anonymous and + device memory usually isn't + <braunr> (although having the cache at the vm layer instead of the pager + layer allows nice things like the swap cache) + <mcsim> But this shouldn't affect possibility of implementing of device + pager. + <braunr> yes it may + <braunr> consider how a fault is actually handled by a device + <braunr> mach must use weird fictitious pages for that + <braunr> whereas it would be better to simply let the pager handle the + fault as it sees fit + <mcsim> setting may_cache to false should resolve the issue + <braunr> for the caching problem, yes + <braunr> which is why i still think it's better to handle the cache at the + vm layer, unlike UVM which lets the vnode pager handle its own cache, and + removes the vm cache completely + <mcsim> The only issue with pager interface I see is implementing of + scatter-gather DMA (as current interface does not support non-consecutive + access) + <braunr> right + <braunr> but that's a performance issue + <braunr> my problem with device pagers is correctness + <braunr> currently, i think the kernel just asks pagers for "data" + <braunr> whereas a device pager should really map its device memory where + the fault happen + <mcsim> braunr: You mean that every access to memory should cause page + fault? + <mcsim> I mean mapping of device memory + <braunr> no + <braunr> i mean a fault on device mapped memory should directly access a + shared region + <braunr> whereas file pagers only implement backing store + <braunr> let me explain a bit more + <braunr> here is what happens with file mapped memory + <braunr> you map it, access it (some I/O is done to get the page content in + physical memory), then later it's flushed back + <braunr> whereas with device memory, there shouldn't be any I/O, the device + memory should directly be mapped (well, some devices need the same + caching behaviour, while others provide direct access) + <braunr> one of the obvious consequences is that, when you map device + memory (e.g. a framebuffer), you expect changes in your mapped memory to + be effective right away + <braunr> while with file mapped memory, you need to msync() it + <braunr> (some framebuffers also need to be synced, which suggests greater + control is needed for external pagers) + <mcsim> Seems that I understand you. But how it is implemented in other + OS'es? Do they set something in mmu? + <braunr> mcsim: in netbsd, pagers have a fault operatin in addition to get + and put + <braunr> the device pager sets get and put to null and implements fault + only + <braunr> the fault callback then calls the d_mmap callback of the specific + driver + <braunr> which usually results in the mmu being programmed directly + <braunr> (e.g. pmap_enter or similar) + <braunr> in linux, i think raw device drivers, being implemented as + character device files, must provide raw read/write/mmap/etc.. functions + <braunr> so it looks pretty much similar + <braunr> i'd say our current external pager interface is insufficient for + device pagers + <braunr> but antrik may know more since he worked on ggi + <braunr> antrik: ^ + <mcsim> braunr: Seems he used io_map + <braunr> mcsim: where ar eyou looking at ? the incubator ? + <mcsim> his master's thesis + <braunr> ah the thesis + <braunr> but where ? :) + <mcsim> I'll give you a link + <mcsim> http://dl.dropbox.com/u/36519904/kgi_on_hurd.pdf + <braunr> thanks + <mcsim> see p 158 + <braunr> arg, more than 200 pages, and he says he's lazy :/ + <braunr> mcsim: btw, have a look at m_o_ready + <mcsim> braunr: This is old form of mo_change attributes + <mcsim> I'm not going to change it + <braunr> mcsim: these are actually the default object parameters right ? + <braunr> mcsim: if you don't change it, it means the kernel must set + default values until the pager changes them, if it does + <mcsim> yes. + <antrik> mcsim: madvise() on Linux has a separate flag to indicate that + pages won't be reused. thus I think it would *not* be a good idea to + imply it in SEQUENTIAL + <antrik> braunr: yes, my KMS code relies on mapping memory objects for the + framebuffer + <antrik> (it should be noted though that on "modern" hardware, mapping + graphics memory directly usually gives very poor performance, and drivers + tend to avoid it...) + <antrik> mcsim: BTW, it was most likely me who warned about legal issues + with KAM's work. AFAIK he never managed to get the copyright assignment + done :-( + <antrik> (that's not really mandatory for the gnumach work though... only + for the Hurd userspace parts) + <antrik> also I'd like to point out again that the cluster_size argument + from OSF Mach was probably *not* meant for advice from application + programs, but rather was supposed to reflect the cluster size of the + filesystem in question. at least that sounds much more plausible to me... + <antrik> braunr: I have no idea whay you mean by "device pager". device + memory is mapped once when the VM mapping is established; there is no + need for any fault handling... + <antrik> mcsim: to be clear, I think the cluster_size parameter is mostly + orthogonal to policy... and probably not very useful at all, as ext2 + almost always uses page-sized clusters. I'm strongly advise against + bothering with it in the initial implementation + <antrik> mcsim: to avoid confusion, better use a completely different name + for the policy-decided readahead size + <mcsim> antrik: ok + <antrik> braunr: well, yes, the thesis report turned out HUGE; but the + actual work I did on the KGI port is fairly tiny (not more than a few + weeks of actual hacking... everything else was just brooding) + <antrik> braunr: more importantly, it's pretty much the last (and only + non-trivial) work I did on the Hurd :-( + <antrik> (also, I don't think I used the word "lazy"... my problem is not + laziness per se; but rather inability to motivate myself to do anything + not providing near-instant gratification...) + <braunr> antrik: right + <braunr> antrik: i shouldn't consider myself lazy either + <braunr> mcsim: i agree with antrik, as i told you weeks ago + <braunr> about + <braunr> 21:45 < antrik> mcsim: to be clear, I think the cluster_size + parameter is mostly orthogonal to policy... and probably not very useful + at all, as ext2 almost always uses page-sized clusters. I'm strongly + advise against bothering with it + <braunr> in the initial implementation + <braunr> antrik: but how do you actually map device memory ? + <braunr> also, strangely enough, here is the comment in dragonflys + madvise(2) + <braunr> 21:45 < antrik> mcsim: to be clear, I think the cluster_size + parameter is mostly orthogonal to policy... and probably not very useful + at all, as ext2 almost always uses page-sized clusters. I'm strongly + advise against bothering with it + <braunr> in the initial implementation + <braunr> arg + <braunr> MADV_SEQUENTIAL Causes the VM system to depress the priority of + pages immediately preceding a given page when it is faulted in. + <antrik> braunr: interesting... + <antrik> (about SEQUENTIAL on dragonfly) + <antrik> as for mapping device memory, I just use to device_map() on the + mem device to map the physical address space into a memory object, and + then through vm_map into the driver (and sometimes application) address + space + <antrik> formally, there *is* a pager involved of course (implemented + in-kernel by the mem device), but it doesn't really do anything + interesting + <antrik> thinking about it, there *might* actually be page faults involved + when the address ranges are first accessed... but even then, the handling + is really trivial and not terribly interesting + <braunr> antrik: it does the most interesting part, create the physical + mapping + <braunr> and as trivial as it is, it requires a special interface + <braunr> i'll read about device_map again + <braunr> but yes, the fact that it's in-kernel is what solves the problem + here + <braunr> what i'm interested in is to do it outside the kernel :) + <antrik> why would you want to do that? + <antrik> there is no policy involved in doing an MMIO mapping + <antrik> you ask for the pysical memory region you are interested in, and + that's it + <antrik> whether the kernel adds the page table entries immediately or on + faults is really an implementation detail + <antrik> braunr: ^ + <braunr> yes it's a detail + <braunr> but do we currently have the interface to make such mappings from + userspace ? + <braunr> and i want to do that because i'd like as many drivers as possible + outside the kernel of course + <antrik> again, the userspace driver asks the kernel to establish the + mapping (through device_map() and then vm_map() on the resulting memory + object) + <braunr> hm i'm missing something + <braunr> + http://www.gnu.org/software/hurd/gnumach-doc/Device-Map.html#Device-Map + <= this one ? + <antrik> yes, this one + <braunr> but this implies the device is implemented by the kernel + <antrik> the mem device is, yes + <antrik> but that's not a driver + <braunr> ah + <antrik> it's just the interface for doing MMIO + <antrik> (well, any physical mapping... but MMIO is probably the only real + use case for that) + <braunr> ok + <braunr> i was thinking about completely removing the device interface from + the kernel actually + <braunr> but it makes sense to have such devices there + <antrik> well, in theory, specific kernel drivers can expose their own + device_map() -- but IIRC the only one that does (besides mem of course) + is maptime -- which is not a real driver either... + <braunr> oh btw, i didn't know you had a blog :) + <antrik> well, it would be possible to replace the device interface by + specific interfaces for the generic pseudo devices... I'm not sure how + useful that would be + <braunr> there are lots of interesting stuff there + <antrik> hehe... another failure ;-) + <braunr> failure ? + <antrik> well, when I realized that I'm speding a lot of time pondering + things, and never can get myself to actually impelemnt any of them, I had + the idea that if I write them down, there might at least be *some* good + from it... + <antrik> unfortunately it turned out that I need so much effort to write + things down, that most of the time I can't get myself to do that either + :-( + <braunr> i see + <braunr> well it's still nice to have it + <antrik> (notice that the latest entry is two years old... and I haven't + even started describing most of my central ideas :-( ) + <braunr> antrik: i tried to create a blog once, and found what i wrote so + stupid i immediately removed it + <antrik> hehe + <antrik> actually some of my entries seem silly in retrospect as well... + <antrik> but I guess that's just the way it is ;-) + <braunr> :) + <braunr> i'm almost sure other people would be interested in what i had to + say + <antrik> BTW, I'm actually not sure whether the Mach interfaces are + sufficient to implement GEM/TTM... we would certainly need kernel support + for GART (as for any other kind IOMMU in fact); but beyond that it's not + clear to me + <braunr> GEM ? TTM ? GART ? + <antrik> GEM = Graphics Execution Manager. part of the "new" DRM interface, + closely tied with KMS + <antrik> TTM = Translation Table Manager. does part of the background work + for most of the GEM drivers + <braunr> "The Graphics Execution Manager (GEM) is a computer software + system developed by Intel to do memory management for device drivers for + graphics chipsets." hmm + <antrik> (in fact it was originally meant to provide the actual interface; + but the Inter folks decided that it's not useful for their UMA graphics) + <antrik> GART = Graphics Aperture + <antrik> kind of an IOMMU for graphics cards + <antrik> allowing the graphics card to work with virtual mappings of main + memory + <antrik> (i.e. allowing safe DMA) + <braunr> ok + <braunr> all this graphics stuff looks so complex :/ + <antrik> it is + <antrik> I have a whole big chapter on that in my thesis... and I'm not + even sure I got everything right + <braunr> what is nvidia using/doing (except for getting the finger) ? + <antrik> flushing out all the details for KMS, GEM etc. took the developers + like two years (even longer if counting the history of TTM) + <antrik> Nvidia's proprietary stuff uses a completely own kernel interface, + which is of course not exposed or docuemented in any way... but I guess + it's actually similar in what it does) + <braunr> ok + <antrik> (you could ask the nouveau guys if you are truly + interested... they are doing most of their reverse engineering at the + kernel interface level) + <braunr> it seems graphics have very special needs, and a lot of them + <braunr> and the interfaces are changing often + <braunr> so it's not that much interesting currently + <braunr> it just means we'll probably have to change the mach interface too + <braunr> like you said + <braunr> so the answer to my question, which was something like "do mach + external pagers only implement files ?", is likely yes + <antrik> well, KMS/GEM had reached some stability; but now there are + further changes ahead with the embedded folks coming in with all their + dedicated hardware, calling for unified buffer management across the + whole pipeline (from capture to output) + <antrik> and yes: graphics hardware tends to be much more complex regarding + the interface than any other hardware. that's because it's a combination + of actual I/O (like most other devices) with a very powerful coprocessor + <antrik> and the coprocessor part is pretty much unique amongst peripherial + devices + <antrik> (actually, the I/O part is also much more complex than most other + hardware... but that alone would only require a more complex driver, not + special interfaces) + <antrik> embedded hardware makes it more interesting in that the I/O + part(s) are separate from the coprocessor ones; and that there are often + several separate specialised ones of each... the DRM/KMS stuff is not + prepared to deal with this + <antrik> v4l over time has evolved to cover such things; but it's not + really the right place to implement graphics drivers... which is why + there are not efforts to unify these frameworks. funny times... + + +## IRC, freenode, #hurd, 2012-07-03 + + <braunr> mcsim: vm_for_every_page should be static + <mcsim> braunr: ok + <braunr> mcsim: see http://gcc.gnu.org/onlinedocs/gcc/Inline.html + <braunr> and it looks big enough that you shouldn't make it inline + <braunr> let the compiler decide for you (which is possible only if the + function is static) + <braunr> (otherwise a global symbol needs to exist) + <braunr> mcsim: i don't know where you copied that comment from, but you + should review the description of the vm_advice call in mach.Defs + <mcsim> braunr: I see + <mcsim> braunr: It was vm_inherit :) + <braunr> mcsim: why isn't NORMAL defined in vm_advise.h ? + <braunr> mcsim: i figured actually ;) + <mcsim> braunr: I was going to do it later when. + <braunr> mcsim: for more info on inline, see + http://www.kernel.org/doc/Documentation/CodingStyle + <braunr> arg that's an old one + <mcsim> braunr: I know that I do not follow coding style + <braunr> mcsim: this one is about linux :p + <braunr> mcsim: http://lxr.linux.no/linux/Documentation/CodingStyle should + have it + <braunr> mcsim: "Chapter 15: The inline disease" + <mcsim> I was going to fix it later during refactoring when I'll merge + mplaneta/gsoc12/working to mplaneta/gsoc12/master + <braunr> be sure not to forget :p + <braunr> and the best not to forget is to do it asap + <braunr> +way + <mcsim> As to inline. I thought that even if I specify function as inline + gcc makes final decision about it. + <mcsim> There was a specifier that made function always inline, AFAIR. + <braunr> gcc can force a function not to be inline, yes + <braunr> but inline is still considered as a strong hint + + +## IRC, freenode, #hurd, 2012-07-05 + + <mcsim1> braunr: hello. You've said that pager has to supply 2 values to + kernel to give it an advice how execute page fault. These two values + should be number of pages before and after the page where fault + occurred. But for sequential policy number of pager before makes no + sense. For random policy too. For normal policy it would be sane to make + readahead symmetric. Probably it would be sane to make pager supply + cluster_size (if it is necessary to supply any) that w + <mcsim1> *that will be advice for kernel of least sane value? And maximal + value will be f(free_memory, map_entry_size)? + <antrik> mcsim1: I doubt symmetric readahead would be a good default + policy... while it's hard to estimate an optimum over all typical use + cases, I'm pretty sure most situtations will benefit almost exclusively + from reading following pages, not preceeding ones + <antrik> I'm not even sure it's useful to read preceding pages at all in + the default policy -- the use cases are probably so rare that the penalty + in all other use cases is not justified. I might be wrong on that + though... + <antrik> I wonder how other systems handle that + <LarstiQ> antrik: if there is a mismatch between pages and the underlying + store, like why changing small bits of data on an ssd is slow? + <braunr> mcsim1: i don't see why not + <braunr> antrik: netbsd reads a few pages before too + <braunr> actually, what netbsd does vary on the version, some only mapped + in resident pages, later versions started asynchronous transfers in the + hope those pages would be there + <antrik> LarstiQ: not sure what you are trying to say + <braunr> in linux : + <braunr> 321 * MADV_NORMAL - the default behavior is to read clusters. + This + <braunr> 322 * results in some read-ahead and read-behind. + <braunr> not sure if it's actually what the implementation does + <antrik> well, right -- it's probably always useful to read whole clusters + at a time, especially if they are the same size as pages... that doesn't + mean it always reads preceding pages; only if the read is in the middle + of the cluster AIUI + <LarstiQ> antrik: basically what braunr just pasted + <antrik> and in most cases, we will want to read some *following* clusters + as well, but probably not preceding ones + * LarstiQ nods + <braunr> antrik: the default policy is usually rather sequential + <braunr> here are the numbers for netbsd + <braunr> 166 static struct uvm_advice uvmadvice[] = { + <braunr> 167 { MADV_NORMAL, 3, 4 }, + <braunr> 168 { MADV_RANDOM, 0, 0 }, + <braunr> 169 { MADV_SEQUENTIAL, 8, 7}, + <braunr> 170 }; + <braunr> struct uvm_advice { + <braunr> int advice; + <braunr> int nback; + <braunr> int nforw; + <braunr> }; + <braunr> surprising isn't it ? + <braunr> they may suggest sequential may be backwards too + <braunr> makes sense + <antrik> braunr: what are these numbers? pages? + <braunr> yes + <antrik> braunr: I suspect the idea behind SEQUENTIAL is that with typical + sequential access patterns, you will start at one end of the file, and + then go towards the other end -- so the extra clusters in the "wrong" + direction do not actually come into play + <antrik> only situation where some extra clusters are actually read is when + you start in the middle of a file, and thus do not know yet in which + direction the sequential read will go... + <braunr> yes, there are similar comments in the linux code + <braunr> mcsim1: so having before and after numbers seems both + straightforward and in par with other implementations + <antrik> I'm still surprised about the almost symmetrical policy for NORMAL + though + <antrik> BTW, is it common to use heuristics for automatically recognizing + random and sequential patterns in the absence of explicit madise? + <braunr> i don't know + <braunr> netbsd doesn't use any, linux seems to have different behaviours + for anonymous and file memory + <antrik> when KAM was working on this stuff, someone suggested that... + <braunr> there is a file_ra_state struct in linux, for per file read-ahead + policy + <braunr> now the structure is of course per file system, since they all use + the same address + <braunr> (which is why i wanted it to be per pager in the first place) + <antrik> mcsim1: as I said before, it might be useful for the pager to + supply cluster size, if it's different than page size. but right now I + don't think this is something worth bothering with... + <antrik> I seriously doubt it would be useful for the pager to supply any + other kind of policy + <antrik> braunr: I don't understand your remark about using the same + address... + <antrik> braunr: pre-mapping seems the obvious way to implement readahead + policy + <antrik> err... per-mapping + <braunr> the ra_state (read ahead state) isn't the policy + <braunr> the policy is per mapping, parts of the implementation of the + policy is per file system + <mcsim1> braunr: How do you look at following implementation of NORMAL + policy: We have fault page that is current. Than we have maximal size of + readahead block. First we find first absent pages before and after + current. Than we try to fit block that will be readahead into this + range. Here could be following situations: in range RBS/2 (RBS -- size of + readahead block) there is no any page, so readahead will be symmetric; if + current page is first absent page than all + <mcsim1> RBS block will consist of pages that are after current; on the + contrary if current page is last absent than readahead will go backwards. + <mcsim1> Additionally if current page is approximately in the middle of the + range we can decrease RBS, supposing that access is random. + <braunr> mcsim1: i think your gsoc project is about readahead, we're in + july, and you need to get the job done + <braunr> mcsim1: grab one policy that works, pages before and after are + good enough + <braunr> use sane default values, let the pagers decide if they want + something else + <braunr> and concentrate on the real work now + <antrik> braunr: I still don't see why pagers should mess with that... only + complicates matters IMHO + <braunr> antrik: probably, since they almost all use the default + implementation + <braunr> mcsim1: just use sane values inside the kernel :p + <braunr> this simplifies things by only adding the new vm_advise call and + not change the existing external pager interface |