diff options
Diffstat (limited to 'open_issues/performance')
| -rw-r--r-- | open_issues/performance/degradation.mdwn | 14 | ||||
| -rw-r--r-- | open_issues/performance/ipc_virtual_copy.mdwn | 358 |
2 files changed, 371 insertions, 1 deletions
diff --git a/open_issues/performance/degradation.mdwn b/open_issues/performance/degradation.mdwn index 5db82e31..db759308 100644 --- a/open_issues/performance/degradation.mdwn +++ b/open_issues/performance/degradation.mdwn @@ -18,7 +18,7 @@ Thomas Schwinge) > tree, reboot, build it again (1st): back to 11 h. Remove build tree, build > it again (2nd): 12 h 40 min. Remove build tree, build it again (3rd): 15 h. -IRC, freenode, #hurd, 2011-07-23 +IRC, freenode, #hurd, 2011-07-23: < antrik> tschwinge: yes, the system definitely gets slower with time. after running for a couple of weeks, it needs at least twice as @@ -26,3 +26,15 @@ IRC, freenode, #hurd, 2011-07-23 < antrik> I don't know whether this is only related to swap usage, or there are some serious fragmentation issues < braunr> antrik: both could be induced by fragmentation + +--- + +During [[IPC_virtual_copy]] testing: + +IRC, freenode, #hurd, 2011-09-02: + + <manuel> interestingly, running it several times has made the performance + drop quite much (i'm getting 400-500MB/s with 1M now, compared to nearly + 800 fifteen minutes ago) + <braunr> manuel: i observed the same behaviour + [...] diff --git a/open_issues/performance/ipc_virtual_copy.mdwn b/open_issues/performance/ipc_virtual_copy.mdwn new file mode 100644 index 00000000..00fa7180 --- /dev/null +++ b/open_issues/performance/ipc_virtual_copy.mdwn @@ -0,0 +1,358 @@ +[[!meta copyright="Copyright © 2011 Free Software Foundation, Inc."]] + +[[!meta license="""[[!toggle id="license" text="GFDL 1.2+"]][[!toggleable +id="license" text="Permission is granted to copy, distribute and/or modify this +document under the terms of the GNU Free Documentation License, Version 1.2 or +any later version published by the Free Software Foundation; with no Invariant +Sections, no Front-Cover Texts, and no Back-Cover Texts. A copy of the license +is included in the section entitled [[GNU Free Documentation +License|/fdl]]."]]"""]] + +IRC, freenode, #hurd, 2011-09-02: + + <slpz> what's the usual throughput for I/O operations (like "dd + if=/dev/zero of=/dev/null") in one of those Xen based Hurd machines + (*bber)? + <braunr> good question + <braunr> slpz: but don't use /dev/zero and /dev/null, as they don't have + anything to do with true I/O operations + <slpz> braunr: in fact, I want to test the performance of IPC's virtual + copy operations + <braunr> ok + <slpz> braunr: sorry, the "I/O" was misleading + <braunr> use bs=4096 then i guess + <slpz> bs > 2k + <braunr> ? + <slpz> braunr: everything about 2k is copied by vm_map_copyin/copyout + <slpz> s/about/above/ + <slpz> braunr: MiG's stubs check for that value and generate complex (with + out_of_line memory) messages if datalen is above 2k, IIRC + <braunr> ok + <braunr> slpz: found it, thanks + <tschwinge> tschwinge@strauss:~ $ dd if=/dev/zero of=/dev/null bs=4k & p=$! + && sleep 10 && kill -s INFO $p && sleep 1 && kill $p + <tschwinge> [1] 13469 + <tschwinge> 17091+0 records in + <tschwinge> 17090+0 records out + <tschwinge> 70000640 bytes (70 MB) copied, 17.1436 s, 4.1 MB/s + <tschwinge> Note, however 10 s vs. 17 s! + <tschwinge> And this is slow compared to heal hardware: + <tschwinge> thomas@coulomb:~ $ dd if=/dev/zero of=/dev/null bs=4k & p=$! && + sleep 10 && kill -s INFO $p && sleep 1 && kill $p + <tschwinge> [1] 28290 + <tschwinge> 93611+0 records in + <tschwinge> 93610+0 records out + <tschwinge> 383426560 bytes (383 MB) copied, 9.99 s, 38.4 MB/s + <braunr> tschwinge: is the first result on xen vm ? + <tschwinge> I think so. + <braunr> :/ + <slpz> tschwinge: Thanks! Could you please try with a higher block size, + something like 128k or 256k? + <tschwinge> strauss is on a machine that also hosts a buildd, I think. + <braunr> oh ok + <pinotree> yes, aside either rossini or mozart + <tschwinge> And I can confirm that with dd if=/dev/zero of=/dev/null bs=4k + running, a parallel sleep 10 takes about 20 s (on strauss). + +[[open_issues/time]] + + <braunr> slpz: i'll set up xen hosts soon and can try those tests while + nothing else runs to have more accurate results + <tschwinge> tschwinge@strauss:~ $ dd if=/dev/zero of=/dev/null bs=256k & + p=$! && sleep 10 && kill -s INFO $p && sleep 1 && kill $p + <tschwinge> [1] 13482 + <tschwinge> 4566+0 records in + <tschwinge> 4565+0 records out + <tschwinge> 1196687360 bytes (1.2 GB) copied, 13.6751 s, 87.5 MB/s + <braunr> slpz: gains are logarithmic beyond the page size + <tschwinge> thomas@coulomb:~ $ dd if=/dev/zero of=/dev/null bs=256k & p=$! + && sleep 10 && kill -s INFO $p && sleep 1 && kill $p + <tschwinge> [1] 28295 + <tschwinge> 6335+0 records in + <tschwinge> 6334+0 records out + <tschwinge> 1660420096 bytes (1.7 GB) copied, 9.99 s, 166 MB/s + <tschwinge> This time a the sleep 10 decided to take 13.6 s. + ``Interesting.'' + <slpz> tschwinge: Thanks again. The results for the Xen machine are not bad + though. I can't obtain a throughput over 50MB/s with KVM. + <tschwinge> slpz: Want more data (bs)? Just tell. + <braunr> slpz: i easily get more than that + <braunr> slpz: what buffer size do you use ? + <slpz> tschwinge: no, I just wanted to see if Xen has an upper limit beyond + KVM's. Thank you. + <slpz> braunr: I try with different sizes until I find the maximum + throughput for a certain amount of requests (count) + <slpz> braunr: are you working with KVM? + <braunr> yes + <braunr> slpz: my processor is a model name : Intel(R) Core(TM)2 Duo + CPU E7500 @ 2.93GHz + <braunr> Linux silvermoon 2.6.32-5-amd64 #1 SMP Tue Jun 14 09:42:28 UTC + 2011 x86_64 GNU/Linux + <braunr> (standard amd64 squeeze kernel) + <slpz> braunr: and KVM's version? + <braunr> squeeze (0.12.5) + <braunr> bbl + <gnu_srs> 212467712 bytes (212 MB) copied, 9.95 s, 21.4 MB/s on kvm for me! + <slpz> gnu_srs: which block size? + <gnu_srs> 4k, and 61.7 MB/s with 256k + <slpz> gnu_srs: could you try with 512k and 1M? + <gnu_srs> 512k: 56.0 MB/s, 1024k: 40.2 MB/s Looks like the peak is around a + few 100k + <slpz> gnu_srs: thanks! + <slpz> I've just obtained 1.3GB/s with bs=512k on other (newer) machine + <braunr> on which hw/vm ? + <slpz> I knew this is a cpu-bound test, but I couldn't imagine faster + processors could make this difference + <slpz> braunr: Intel(R) Core(TM) i5 CPU 650 @ 3.20GHz + <slpz> braunr: KVM + <braunr> ok + <braunr> how much time did you wait before reading the result ? + <slpz> that was 20x times better than the same test on my Intel(R) + Core(TM)2 Duo CPU T7500 @ 2.20GHz + <slpz> braunr: I've repeated the test with a fixed "count" + <gnu_srs> My box is: Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz: Max + is 67 MB/s around 140k block size + <braunr> yes but how much time did dd run ? + <gnu_srs> 10 s plus/minus a few fractions of a second, + <braunr> try waiting 30s + <slpz> braunr: didn't check, let me try again + <braunr> my kvm peaks at 130 MiB/s with bs 512k / 1M + <gnu_srs> 2029690880 bytes (2.0 GB) copied, 30.02 s, 67.6 MB/s, bs=140k + <braunr> gnu_srs: i'm very surprised with slpz's result of 1.3 GiB/s + <slpz> braunr: over 60 s running, same performance + <braunr> nice + <braunr> i wonder what makes it so fast + <braunr> how much cache ? + <gnu_srs> Me too, I cannot get better values than around 67 MB/s + <braunr> gnu_srs: same questions + <slpz> braunr: 4096KB, same as my laptop + <braunr> slpz: l2 ? l3 ? + <gnu_srs> kvm: cache=writeback, CPU: 4096 KB + <braunr> gnu_srs: this has nothing to do with the qemu option, it's about + the cpu + <slpz> braunr: no idea, it's the first time I touch this machine. I going + to see if I find the model in processorfinder + <braunr> under my host linux system, i get a similar plot, that is, + performance drops beyond bs=1M + <gnu_srs> braunr: OK, bu I gave you the cache size too, same as slpz. + <braunr> i wonder what dd actually does + <braunr> read() and writes i guess + <slpz> braunr: read/write repeatedly, nothing fancy + <braunr> slpz: i don't think it's a good test for virtual copy + <braunr> io_read_request, vm_deallocate, io_write_request, right + <braunr> slpz: i really wonder what it is about i5 that improves speed so + much + <slpz> braunr: me too + <slpz> braunr: L2: 2x256KB, L3: 4MB + <slpz> and something calling "SmartCache" + <gnu_srs> slpz: where did you find these values? + <slpz> gnu_srs: ark.intel.com and wikipedia + <gnu_srs> aha, cpuinfo just gives cache size. + <slpz> that "SmartCache" thing seems to be just L2 cache sharing between + cores. Shouldn't make a different since we're using only one core, and I + don't see KVM hooping between them. + <manuel> with bs=256k: 7004487680 bytes (7.0 GB) copied, 10 s, 700 MB/s + <manuel> (qemu/kvm, 3 * Intel(R) Xeon(R) E5504 2GHz, cache size 4096 KB) + <slpz> manuel: did you try with 512k/1M? + <manuel> bs=512k: 7730626560 bytes (7.7 GB) copied, 10 s, 773 MB/s + <manuel> bs=1M: 7896825856 bytes (7.9 GB) copied, 10 s, 790 MB/s + <slpz> manuel: those are pretty good numbers too + <braunr> xeon processor + <gnu_srs> lshw gave me: L1 Cache 256KiB, L2 cache 4MiB + <slpz> sincerely, I've never seen Hurd running this fast. Just checked + "uname -a" to make sure I didn't take the wrong image :-) + <manuel> for bs=256k, 60s: 40582250496 bytes (41 GB) copied, 60 s, 676 MB/s + <braunr> slpz: i think you can assume processor differences alter raw + copies too much to get any valuable results about virtual copy operations + <braunr> you need a specialized test program + <manuel> and bs=512k, 60s, 753 MB/s + <slpz> braunr: I'm using the mach_perf suite from OSFMach to do the + "serious" testing. I just wanted a non-synthetic test to confirm the + readings. + +[[!taglink open_issue_gnumach]] -- have a look at *mach_perf*. + + <braunr> manuel: how much cache ? 2M ? + <braunr> slpz: ok + <braunr> manuel: hmno, more i guess + <manuel> braunr: /proc/cpuinfo says cache size : 4096 KB + <braunr> ok + <braunr> manuel: performance should drop beyond bs=2M + <braunr> but that's not relevant anyway + <gnu_srs> Linux: bs=1M, 10.8 GB/s + <slpz> I think this difference is too big to be only due to a bigger amount + of CPU cycles... + <braunr> slpz: clearly + <slpz> gnu_srs: your host system has 64 or 32 bits? + <slpz> braunr: I'm going to investigate a bit + <slpz> but this accidental discovery just made my day. We're able to run + Hurd at decent speeds on newer hardware! + <braunr> slpz: what result do you get with the same test on your host + system ? + <manuel> interestingly, running it several times has made the performance + drop quite much (i'm getting 400-500MB/s with 1M now, compared to nearly + 800 fifteen minutes ago) + +[[Degradataion]]. + + <slpz> braunr: probably an almost infinite throughput, but I don't consider + that a valid test, since in Linux, the write operation to "/dev/null" + doesn't involve memory copying/moving + <braunr> manuel: i observed the same behaviour + <gnu_srs> slpz: Host system is 64 bit + <braunr> slpz: it doesn't on the hurd either + <braunr> slpz: (under 2k, that is) + <braunr> over* + <slpz> braunr: humm, you're right, as the null translator doesn't "touch" + the memory, CoW rules apply + <braunr> slpz: the only thing which actually copies things around is dd + <braunr> probably by simply calling read() + <braunr> which gets its result from a VM copy operation, but copies the + content to the caller provided buffer + <braunr> then vm_deallocate() the data from the storeio (zero) translator + <braunr> if storeio isn't too dumb, it doesn't even touch the transfered + buffer (as anonymous vm_map()ped memory is already cleared) + +[[!taglink open_issue_documentation]] + + <braunr> so this is a good test for measuring (profiling?) our ipc overhead + <braunr> and possibly the vm mapping operations (which could partly explain + why the results get worse over time) + <braunr> manuel: can you run vminfo | wc -l on your gnumach process ? + <slpz> braunr: Yes, unless some special situation apply, like the source + address/offset being unaligned, or if the translator decides to return + the result in a different buffer (which I assume is not the case for + storeio/zero) + <manuel> braunr: 35 + <braunr> slpz: they can't be unaligned, the vm code asserts that + <braunr> manuel: ok, this is normal + <slpz> braunr: address/offset from read() + <braunr> slpz: the caller provided buffer you mean ? + <slpz> braunr: yes, and the offset of the memory_object, if it's a pager + based translator + <braunr> slpz: highly unlikely, the compiler chooses appropriate alignments + for such buffers + <slpz> braunr: in those cases, memcpy is used over vm_copy + <braunr> slpz: and the glibc memcpy() optimized versions can usually deal + with that + <braunr> slpz: i don't get your point about memory objects + <braunr> slpz: requests on memory objects always have aligned values too + <slpz> braunr: sure, but can't deal with the user requesting non + page-aligned sizes + <braunr> slpz: we're considering our dd tests, for which we made sure sizes + were page aligned + <slpz> braunr: oh, I was talking in a general sense, not just in this dd + tests, sorry + <slpz> by the way, dd on the host tops at 12 GB/s with bs=2M + <braunr> that's consistent with our other results + <braunr> slpz: you mean, even on your i5 processor with 1.3 GiB/s on your + hurd kvm ? + <slpz> braunr: yes, on the GNU/Linux which is running as host + <braunr> slpz: well that's not consistent + <slpz> braunr: consistent with what? + <braunr> slpz: i get roughly the same result on my host, but ten times less + on my hurd kvm + <braunr> slpz: what's your kernel/kvm versions ? + <slpz> 2.6.32-5-amd64 (debian's build) 0.12.5 + <braunr> same here + <braunr> i'm a bit clueless + <braunr> why do i only get 130 MiB/s where you get 1.3 .. ? :) + <slpz> well, on my laptop, where Hurd on KVM tops on 50 MB/s, Linux gets a + bit more than 10 GB/s + <braunr> see + <braunr> slpz: reduce bs to 256k and test again if you have time please + <slpz> braunr: on which system? + <braunr> slpz: the fast one + <braunr> (linux host) + <slpz> braunr: Hurd? + <slpz> ok + <slpz> 12 GB/s + <braunr> i get 13.3 + <slpz> same for 128k, only at 64k starts dropping + <slpz> maybe, on linux we're being limited by memory speed, while on Hurd's + this test is (much) more CPU-bound? + <braunr> slpz: maybe + <braunr> too bad processor stalls aren't easy to measure + <slpz> braunr: that's very true. It's funny when you read a paper which + measures performance by cycles on an old RISC processor. That's almost + impossible to do (with reliability) nowadays :-/ + <slpz> I wonder which throughput can achieve Hurd running bare-metal on + this machine... + <antrik> both the Xeon and the i5 use cores based on the Nehalem + architecture + <antrik> apparently Nehalem is where Intel first introduces nested page + tables + <antrik> which pretty much explains the considerably lower overhead of VM + magic + <cjuner> antrik, what are nested page tables? (sounds like the 4-level page + tables we already have on amd64, or 2-level or 3-level on x86 pae) + <antrik> page tables were always 2-level on x86 + <antrik> that's unrelated + <antrik> nested page tables means there is another layer of address + translation, so the VMM can do it's own translation and doesn't care what + the guest system does => no longer has to intercept all page table + manipulations + <braunr> antrik: do you imply it only applies to virtualized systems ? + <antrik> braunr: yes + <slpz> antrik: Good guess. Looks like Intel's EPT are doing the trick by + allowing the guest OS deal with its own page faults + <slpz> antrik: next monday, I'll try disabling EPT support in KVM on that + machine (the fast one). That should confirm your theory empirically. + <slpz> this also means that there're too many page faults, as we should be + doing virtual copies of memory that is not being accessed + <slpz> and looking at how the value of "page faults" in "vmstat" increases, + shows that page faults are directly proportional to the number of pages + we are asking from the translator + <slpz> I've also tried doing a long read() directly, to be sure that "dd" + is not doing something weird, and it shows the same behaviour. + <braunr> slpz: dd does copy buffers + <braunr> slpz: i told you, it's not a good test case for pure virtual copy + evaluation + <braunr> antrik: do you know if xen benefits from nested page tables ? + <antrik> no idea + +[[!taglink open_issue_xen]] + + <slpz> braunr: but my small program doesn't, and still provokes a lot of + page faults + <braunr> slpz: are you certain it doesn't ? + <slpz> braunr: looking at google, it looks like recent Xen > 3.4 supports + EPT + <braunr> ok + <braunr> i'm ordering my new server right now, core i5 :) + <slpz> braunr: at least not explicitily. I need to look at MiG stubs again, + I don't remember if they do something weird. + <antrik> braunr: sandybridge or nehalem? :-) + <braunr> antrik: no idea + <antrik> does it tell a model number? + <braunr> not yet + <braunr> but i don't have a choice for that, so i'll order it first, check + after + <antrik> hehe + <antrik> I'm not sure it makes all that much difference anyways for a + server... unless you are running it at 100% load ;-) + <braunr> antrik: i'm planning on running xen guests suchs as new buildd + <antrik> hm... note though that some of the nehalem-generation i5s were + dual-core, while all the new ones are quad + <braunr> it's a quad + <antrik> the newer generation has better performance per GHz and per + Watt... but considering that we are rather I/O-limited in most cases, it + probably won't make much difference + <antrik> not sure whether there are further virtualisation improvements + that could be relevant... + <braunr> buildds spend much time running gcc, so even such improvements + should help + <braunr> there, server ordered :) + <braunr> antrik: model name : Intel(R) Core(TM) i5-2400 CPU @ 3.10GHz + +IRC, freenode, #hurd, 2011-09-06: + + <slpz> youpi: what machines are being used for buildd? Do you know if they + have EPT/RVI? + <youpi> we use PV Xen there + <slpz> I think Xen could also take advantage of those technologies. Not + sure if only in HVM or with PV too. + <youpi> only in HVM + <youpi> in PV it does not make sense: the guest already provides the + translated page table + <youpi> which is just faster than anything else |