From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-wr1-x430.google.com (mail-wr1-x430.google.com [IPv6:2a00:1450:4864:20::430]) (using TLSv1.2 with cipher ECDHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) by lists.bufferbloat.net (Postfix) with ESMTPS id 5CF693B29D for ; Sat, 22 Oct 2022 10:44:32 -0400 (EDT) Received: by mail-wr1-x430.google.com with SMTP id a10so8574703wrm.12 for ; Sat, 22 Oct 2022 07:44:32 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20210112; h=content-transfer-encoding:cc:to:subject:message-id:date:from :in-reply-to:references:mime-version:from:to:cc:subject:date :message-id:reply-to; bh=m7NiiXQXMAN/mRtHG8l5pKl5hRh2Azi+R+sKc0yi5QU=; b=qbxb4gReWHCOTql7owdkfziwLHuczsTLDx5L9UTWfG42qvIsMZBILAni0ZBirz7EFo Uod4znVF3K5T05gRugcVfvTMfz6Qg0prOLKzcbynbBH0oDoR2Tvn4+mYP/2k8ZW9LFLL GtIL703wbm/BvAq9Fba+87BgqgW/aPuM8kSc3AGW6W17AhOCdYwFA/MEot9KxBiZB5bc DlrN9Iun+zxqwY7rGXYNVfQPvnL6rHZNXQ7y/0r5duxEbCG6ykWlnwKjhKPINGidBBEP AnmkD9e9hENxTSCd+bINvSQ5PLbH9sF53etD109tm7HFXk7bHKkUiDNre2KR054PKWIT vHPg== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=content-transfer-encoding:cc:to:subject:message-id:date:from :in-reply-to:references:mime-version:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=m7NiiXQXMAN/mRtHG8l5pKl5hRh2Azi+R+sKc0yi5QU=; b=O6XL/tRHpxAeyOBdmBMR4hnqLuY5Y9ji5YAXcA9YIK42weGM/n3n+c9FHkzT7j9pWj NJDt/bZ7HQ6twSUqKooWbW82biTnCWPryT2gpy8QOyWWCFDx5Gos/gfka4QBRlO4Ul8u x7HI66XuVucVD5KN8LyX924HqfD58oylp88+uKVBNThv5PTjov2GBDD26ZNT5IWGJVt6 2ZhTqlInLT5ogkH4aZVIYw4M8LCZ2cmG25uMJVTaSfc4n3kIHvcmKBknJNuwrjJOtQIN BdJkfuj3hzyNwKLjao+rlQRFmd4fQ7FqolDuk73KZjYscomqeluIRoVMAgBF2K/rMK+x s01Q== X-Gm-Message-State: ACrzQf18DTKKY+Nk/syDH7esJnHrM+Mps/hv9V7NVMVANM7SgTzROC0o SQZ/DyFhTAOBN1cp2IIX7bk+6a1bMT5uo/FmQ4FHSETc7AE= X-Google-Smtp-Source: AMsMyM7CBhJMB35wW10GO+2UykylLiS5HY7pFkY/4Wash6Ed8FiIw7GelvQ6kItGyL8CWHPTndxy2AaVjqqZcOJWJ2k= X-Received: by 2002:a5d:5109:0:b0:22f:ed4:65da with SMTP id s9-20020a5d5109000000b0022f0ed465damr14951250wrt.688.1666449870536; Sat, 22 Oct 2022 07:44:30 -0700 (PDT) MIME-Version: 1.0 References: <87bkqatu61.fsf@toke.dk> <759c25c6fd54dceccc00eada5ccf5358d2d1c20c.camel@kau.se> In-Reply-To: From: Dave Taht Date: Sat, 22 Oct 2022 07:44:16 -0700 Message-ID: To: Herbert Wolverson Cc: "libreqos@lists.bufferbloat.net" Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable Subject: Re: [LibreQoS] In BPF pping - so far X-BeenThere: libreqos@lists.bufferbloat.net X-Mailman-Version: 2.1.20 Precedence: list List-Id: Many ISPs need the kinds of quality shaping cake can do List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Sat, 22 Oct 2022 14:44:32 -0000 On Sat, Oct 22, 2022 at 7:32 AM Herbert Wolverson via LibreQoS wrote: > > This morning I tested cpu-pping with live customers! > A little over 1,200 mapped IP addresses, about 600 mbps of real traffic f= lowing through a big > hierarchy of 52 sites. (600 is our "quiet time" traffic) > > It started very well: the updated xdp-cpumap system dropped in place and = the system worked as > before. xdp_pping started to show data with correct mappings. CPU load fr= om the mapping > system is within 1% of where it was before. > > After about 20 minutes of continuous execution, it started to run into so= me scaling issues. > The shaping system continued to run wonderfully, and CPU load was still f= ine. However, > it stopped reporting latency data! A bit of debugging showed that once yo= u exceed > 16,384 in-flight TCP streams it isn't handling the "map full" situation g= racefully - and > clearing the map from userspace isn't working correctly. So I hacked away= and hacked > away. > > Anyway, it turns out that it does in fact work fine at that scale. There'= s just a one-line > bug in the xdp_pping.c file. I forgot to actually *call* one line of pack= et cleanup code. > Adding that, and everything was awesome. > > The entire patch that fixed it consists of adding one line: > cleanup_packet_ts(packet_ts); > > Oops. :woot: > > Anyway, with that in place it's running superbly. I did identify a couple= of places in > which it's being overly verbose with debug information, so I've patched t= hat also. > After reducing the overly eager warning about not being able to read a TC= P header, > CPU performance improved by another 2% on average. I note I am VERY interested, science wize, about the actual incidence of malformed TCP headers, mistaken or incorrect ecn-related congestion responses, etc. This is stuff endpoints just arbitrarily drop on the floor. Always interested in the anomalies. https://www.youtube.com/watch?v=3DK1jasTyGLr8 > > Longer-term (i.e. not on a Saturday morning, when I'd rather be playing w= ith my > daughter!), I think I'll look at raising some of the buffer sizes. > > Thanks, > Herbert > > On Wed, Oct 19, 2022 at 11:13 AM Dave Taht wrote: >> >> PS - today's (free) p99 conference is *REALLY AWESOME*. https://www.p99c= onf.io/ >> >> On Wed, Oct 19, 2022 at 9:13 AM Dave Taht wrote: >> > >> > flent outputs a flent.gz file that I can parse and plot 20 differnt >> > ways. Also the graphing tools work on osx >> > >> > On Wed, Oct 19, 2022 at 9:11 AM Herbert Wolverson via LibreQoS >> > wrote: >> > > >> > > That's true. The 12th gen does seem to have some "special" features.= .. makes for a nice writing platform >> > > (this box is primarily my "write books and articles" machine). I'll = be doing a wider test on a more normal >> > > platform, probably at the weekend (with real traffic, hence the dela= y - have to find a time in which I >> > > minimize disruption) >> > > >> > > On Wed, Oct 19, 2022 at 10:49 AM dan wrote: >> > >> >> > >> Those 'efficiency' threads in Intel 12th gen should probably be add= ressed as well. You can't turn them off in BIOS. >> > >> >> > >> On Wed, Oct 19, 2022 at 8:48 AM Robert Chac=C3=B3n via LibreQoS wrote: >> > >>> >> > >>> Awesome work on this! >> > >>> I suspect there should be a slight performance bump once Hyperthre= ading is disabled and efficient power management is off. >> > >>> Hyperthreading/SMT always messes with HTB performance when I leave= it on. Thank you for mentioning that - I now went ahead and added instruct= ions on disabling hyperthreading on the Wiki for new users. >> > >>> Super promising results! >> > >>> Interested to see what throughput is with xdp-cpumap-tc vs cpumap-= pping. So far in your VM setup it seems to be doing very well. >> > >>> >> > >>> On Wed, Oct 19, 2022 at 8:06 AM Herbert Wolverson via LibreQoS wrote: >> > >>>> >> > >>>> Also, I forgot to mention that I *think* the current version has = removed the requirement that the inbound >> > >>>> and outbound classifiers be placed on the same CPU. I know interd= uo was particularly keen on packing >> > >>>> upload into fewer cores. I'll add that to my list of things to te= st. >> > >>>> >> > >>>> On Wed, Oct 19, 2022 at 9:01 AM Herbert Wolverson wrote: >> > >>>>> >> > >>>>> I'll definitely take a look - that does look interesting. I don'= t have X11 on any of my test VMs, but >> > >>>>> it looks like it can work without the GUI. >> > >>>>> >> > >>>>> Thanks! >> > >>>>> >> > >>>>> On Wed, Oct 19, 2022 at 8:58 AM Dave Taht = wrote: >> > >>>>>> >> > >>>>>> could I coax you to adopt flent? >> > >>>>>> >> > >>>>>> apt-get install flent netperf irtt fping >> > >>>>>> >> > >>>>>> You sometimes have to compile netperf yourself with --enable-de= mo on >> > >>>>>> some systems. >> > >>>>>> There are a bunch of python libs neede for the gui, but only on= the client. >> > >>>>>> >> > >>>>>> Then you can run a really gnarly test series and plot the resul= ts over time. >> > >>>>>> >> > >>>>>> flent --socket-stats --step-size=3D.05 -t 'the-test-conditions'= -H >> > >>>>>> the_server_name rrul # 110 other tests >> > >>>>>> >> > >>>>>> >> > >>>>>> On Wed, Oct 19, 2022 at 6:44 AM Herbert Wolverson via LibreQoS >> > >>>>>> wrote: >> > >>>>>> > >> > >>>>>> > Hey, >> > >>>>>> > >> > >>>>>> > Testing the current version ( https://github.com/thebracket/c= pumap-pping-hackjob ), it's doing better than I hoped. This build has share= d (not per-cpu) maps, and a userspace daemon (xdp_pping) to extract and res= et stats. >> > >>>>>> > >> > >>>>>> > My testing environment has grown a bit: >> > >>>>>> > * ShaperVM - running Ubuntu Server and LibreQoS, with the new= cpumap-pping-hackjob version of xdp-cpumap. >> > >>>>>> > * ExtTest - running Ubuntu Server, set as 10.64.1.1. Hosts an= iperf server. >> > >>>>>> > * ClientInt1 - running Ubuntu Server (minimal), set as 10.64.= 1.2. Hosts iperf client. >> > >>>>>> > * ClientInt2 - running Ubuntu Server (minimal), set as 10.64.= 1.3. Hosts iperf client. >> > >>>>>> > >> > >>>>>> > ClientInt1, ClientInt2 and one interface (LAN facing) of Shap= erVM are on a virtual switch. >> > >>>>>> > ExtTest and the other interface (WAN facing) of ShaperVM are = on a different virtual switch. >> > >>>>>> > >> > >>>>>> > These are all on a host machine running Windows 11, a core i7= 12th gen, 32 Gb RAM and fast SSD setup. >> > >>>>>> > >> > >>>>>> > TEST 1: DUAL STREAMS, LOW THROUGHPUT >> > >>>>>> > >> > >>>>>> > For this test, LibreQoS is configured: >> > >>>>>> > * Two APs, each with 5gbit/s max. >> > >>>>>> > * 100.64.1.2 and 100.64.1.3 setup as CPEs, each limited to ab= out 100mbit/s. They map to 1:5 and 2:5 respectively (separate CPUs). >> > >>>>>> > * Set to use Cake >> > >>>>>> > >> > >>>>>> > On each client, roughly simultaneously run: iperf -c 100.64.1= .1 -t 500 (for a long run). Running xdp_pping yields correct results: >> > >>>>>> > >> > >>>>>> > [ >> > >>>>>> > {"tc":"1:5", "avg" : 4, "min" : 3, "max" : 5, "samples" : 11}= , >> > >>>>>> > {"tc":"2:5", "avg" : 4, "min" : 3, "max" : 5, "samples" : 11}= , >> > >>>>>> > {}] >> > >>>>>> > >> > >>>>>> > Or when I waited a while to gather/reset: >> > >>>>>> > >> > >>>>>> > [ >> > >>>>>> > {"tc":"1:5", "avg" : 4, "min" : 3, "max" : 6, "samples" : 60}= , >> > >>>>>> > {"tc":"2:5", "avg" : 4, "min" : 3, "max" : 5, "samples" : 60}= , >> > >>>>>> > {}] >> > >>>>>> > >> > >>>>>> > The ShaperVM shows no errors, just periodic logging that it i= s recording data. CPU is about 2-3% on two CPUs, zero on the others (as ex= pected). >> > >>>>>> > >> > >>>>>> > After 500 seconds of continual iperfing, each client reported= a throughput of 104 Mbit/sec and 6.06 GBytes of data transmitted. >> > >>>>>> > >> > >>>>>> > So for smaller streams, I'd call this a success. >> > >>>>>> > >> > >>>>>> > TEST 2: DUAL STREAMS, HIGH THROUGHPUT >> > >>>>>> > >> > >>>>>> > For this test, LibreQoS is configured: >> > >>>>>> > * Two APs, each with 5gb/s max. >> > >>>>>> > * 100.64.1.2 and 100.64.1.3 setup as CPEs, each limited to 5G= bit/s! Mapped to 1:5 and 2:5 respectively (separate CPUs). >> > >>>>>> > >> > >>>>>> > Run iperfc -c 100.64.1.1 -t 500 on each client at the same ti= me. >> > >>>>>> > >> > >>>>>> > xdp_pping shows results, too: >> > >>>>>> > >> > >>>>>> > [ >> > >>>>>> > {"tc":"1:5", "avg" : 4, "min" : 1, "max" : 7, "samples" : 58}= , >> > >>>>>> > {"tc":"2:5", "avg" : 7, "min" : 3, "max" : 11, "samples" : 58= }, >> > >>>>>> > {}] >> > >>>>>> > >> > >>>>>> > [ >> > >>>>>> > {"tc":"1:5", "avg" : 5, "min" : 4, "max" : 8, "samples" : 13}= , >> > >>>>>> > {"tc":"2:5", "avg" : 8, "min" : 7, "max" : 10, "samples" : 13= }, >> > >>>>>> > {}] >> > >>>>>> > >> > >>>>>> > The ShaperVM shows two CPUs pegging between 70 and 90 percent= . >> > >>>>>> > >> > >>>>>> > After 500 seconds of continual iperfing, each client reported= a throughput of 2.72 Gbits/sec (158 GBytes) and 3.89 Gbits/sec and 226 GBy= tes. >> > >>>>>> > >> > >>>>>> > Maxing out HyperV like this is inducing a bit of latency (whi= ch is to be expected), but it's not bad. I also forgot to disable hyperthre= ading, and looking at the host performance it is sometimes running the seco= nd virtual CPU on an underpowered "fake" CPU. >> > >>>>>> > >> > >>>>>> > So for two large streams, I think we're doing pretty well als= o! >> > >>>>>> > >> > >>>>>> > TEST 3: DUAL STREAMS, SINGLE CPU >> > >>>>>> > >> > >>>>>> > This test is designed to try and blow things up. It's the sam= e as test 2, but both CPEs are set to the same CPU (1), using TC handles 1:= 5 and 1:6. >> > >>>>>> > >> > >>>>>> > ShaperVM CPU1 maxed out in the high 90s, the other CPUs were = idle. The pping stats start to show a bit of degradation in performance for= pounding it so hard: >> > >>>>>> > >> > >>>>>> > [ >> > >>>>>> > {"tc":"1:6", "avg" : 10, "min" : 9, "max" : 19, "samples" : 2= 4}, >> > >>>>>> > {"tc":"1:5", "avg" : 10, "min" : 8, "max" : 18, "samples" : 2= 4}, >> > >>>>>> > {}] >> > >>>>>> > >> > >>>>>> > For whatever reason, it smoothed out over time: >> > >>>>>> > >> > >>>>>> > [ >> > >>>>>> > {"tc":"1:6", "avg" : 10, "min" : 9, "max" : 12, "samples" : 5= 0}, >> > >>>>>> > {"tc":"1:5", "avg" : 10, "min" : 8, "max" : 13, "samples" : 5= 0}, >> > >>>>>> > {}] >> > >>>>>> > >> > >>>>>> > Surprisingly (to me), I didn't encounter errors. Each client = received 2.22 Gbit/s performance, over 129 Gbytes of data. >> > >>>>>> > >> > >>>>>> > TEST 4: DUAL STREAMS, 50 SUB-STREAMS >> > >>>>>> > >> > >>>>>> > This test is also designed to break things. Same as test 3, b= ut using iperf -c 100.64.1.1 -P 50 -t 120 - 50 substreams, to try and reall= y tax the flow tracking. (Shorter time window because I really wanted to go= and find coffee) >> > >>>>>> > >> > >>>>>> > ShaperVM CPU sat at around 80-97%, tending towards 97%. pping= results show that this torture test is worsening performance, and there's = always lots of samples in the buffer: >> > >>>>>> > >> > >>>>>> > [ >> > >>>>>> > {"tc":"1:6", "avg" : 23, "min" : 19, "max" : 27, "samples" : = 49}, >> > >>>>>> > {"tc":"1:5", "avg" : 24, "min" : 19, "max" : 27, "samples" : = 49}, >> > >>>>>> > {}] >> > >>>>>> > >> > >>>>>> > This test also ran better than I expected. You can definitely= see some latency creeping in as I make the system work hard. Each VM showe= d around 2.4 Gbit/s in total performance at the end of the iperf session. T= here's definitely some latency creeping in, which is expected - but I'm not= sure I expected quite that much. >> > >>>>>> > >> > >>>>>> > WHAT'S NEXT & CONCLUSION >> > >>>>>> > >> > >>>>>> > I noticed that I forgot to turn off efficient power managemen= t on my VMs and host, and left Hyperthreading on by mistake. So that hurts = overall performance. >> > >>>>>> > >> > >>>>>> > The base system seems to be working pretty solidly, at least = for small tests.Next up, I'll be removing extraneous debug reporting code, = removing some code paths that don't do anything but report, and looking for= any small optimization opportunities. I'll then re-run these tests. Once t= hat's done, I hope to find a maintenance window on my WISP and try it with = actual traffic. >> > >>>>>> > >> > >>>>>> > I also need to re-run these tests without the pping system to= provide some before/after analysis. >> > >>>>>> > >> > >>>>>> > On Tue, Oct 18, 2022 at 1:01 PM Herbert Wolverson wrote: >> > >>>>>> >> >> > >>>>>> >> It's probably not entirely thread-safe right now (ran into s= ome issues reading per_cpu maps back from userspace; hopefully, I'll get th= at figured out) - but the commits I just pushed have it basically working o= n single-stream testing. :-) >> > >>>>>> >> >> > >>>>>> >> Setup cpumap as usual, and periodically run xdp-pping. This = gives you per-connection RTT information in JSON: >> > >>>>>> >> >> > >>>>>> >> [ >> > >>>>>> >> {"tc":"1:5", "avg" : 5, "min" : 5, "max" : 5, "samples" : 1}= , >> > >>>>>> >> {}] >> > >>>>>> >> >> > >>>>>> >> (With the extra {} because I'm not tracking the tail and hav= en't done comma removal). The tool also empties the various maps used to ga= ther data, acting as a "reset" point. There's a max of 60 samples per queue= , in a ringbuffer setup (so newest will start to overwrite the oldest). >> > >>>>>> >> >> > >>>>>> >> I'll start trying to test on a larger scale now. >> > >>>>>> >> >> > >>>>>> >> On Mon, Oct 17, 2022 at 3:34 PM Robert Chac=C3=B3n wrote: >> > >>>>>> >>> >> > >>>>>> >>> Hey Herbert, >> > >>>>>> >>> >> > >>>>>> >>> Fantastic work! Super exciting to see this coming together,= especially so quickly. >> > >>>>>> >>> I'll test it soon. >> > >>>>>> >>> I understand and agree with your decision to omit certain f= eatures (ICMP tracking,DNS tracking, etc) to optimize performance for our u= se case. Like you said, in order to merge the functionality without a perfo= rmance hit, merging them is sort of the only way right now. Otherwise there= would be a lot of redundancy and lost throughput for an ISP's use. Though = hopefully long term there will be a way to keep all projects working indepe= ndently but interoperably with a plugin system of some kind. >> > >>>>>> >>> >> > >>>>>> >>> By the way, I'm making some headway on LibreQoS v1.3. Focus= ing on optimizations for high sub counts (8000+ subs) as well as stateful c= hanges to the queue structure. >> > >>>>>> >>> I'm working to set up a physical lab to test high throughpu= t and high client count scenarios. >> > >>>>>> >>> When testing beyond ~32,000 filters we get "no space left o= n device" from xdp-cpumap-tc, which I think relates to the bpf map size lim= itation you mentioned. Maybe in the coming months we can take a look at tha= t. >> > >>>>>> >>> >> > >>>>>> >>> Anyway great work on the cpumap-pping program! Excited to s= ee more on this. >> > >>>>>> >>> >> > >>>>>> >>> Thanks, >> > >>>>>> >>> Robert >> > >>>>>> >>> >> > >>>>>> >>> On Mon, Oct 17, 2022 at 12:45 PM Herbert Wolverson via Libr= eQoS wrote: >> > >>>>>> >>>> >> > >>>>>> >>>> Hey, >> > >>>>>> >>>> >> > >>>>>> >>>> My current (unfinished) progress on this is now available = here: https://github.com/thebracket/cpumap-pping-hackjob >> > >>>>>> >>>> >> > >>>>>> >>>> I mean it about the warnings, this isn't at all stable, de= bugged - and can't promise that it won't unleash the nasal demons >> > >>>>>> >>>> (to use a popular C++ phrase). The name is descriptive! ;-= ) >> > >>>>>> >>>> >> > >>>>>> >>>> With that said, I'm pretty happy so far: >> > >>>>>> >>>> >> > >>>>>> >>>> * It runs only on the classifier - which xdp-cpumap-tc has= nicely shunted onto a dedicated CPU. It has to run on both >> > >>>>>> >>>> the inbound and outbound classifiers, since otherwise it= would only see half the conversation. >> > >>>>>> >>>> * It does assume that your ingress and egress CPUs are map= ped to the same interface; I do that anyway in BracketQoS. Not doing >> > >>>>>> >>>> that opens up a potential world of pain, since writes to= the shared maps would require a locking scheme. Too much locking, and you = lose all of the benefit of using multiple CPUs to begin with. >> > >>>>>> >>>> * It is pretty wasteful of RAM, but most of the shaper sys= tems I've worked with have lots of it. >> > >>>>>> >>>> * I've been gradually removing features that I don't want = for BracketQoS. A hypothetical future "useful to everyone" version wouldn't= do that. >> > >>>>>> >>>> * Rate limiting is working, but I removed the requirement = for a shared configuration provided from userland - so right now it's alway= s set to report at 1 second intervals per stream. >> > >>>>>> >>>> >> > >>>>>> >>>> My testbed is currently 3 Hyper-V VMs - a simple "client" = and "world", and a "shaper" VM in between running a slightly hacked-up Libr= eQoS. >> > >>>>>> >>>> iperf from "client" to "world" (with Libre set to allow 10= gbit/s max, via a cake/HTB queue setup) is around 5 gbit/s at present, on m= y >> > >>>>>> >>>> test PC (the host is a core i7, 12th gen, 12 cores - 64gb = RAM and fast SSDs) >> > >>>>>> >>>> >> > >>>>>> >>>> Output currently consists of debug messages reading: >> > >>>>>> >>>> cpumap/0/map:4-1371 [000] D..2. 515.399222: bpf_tra= ce_printk: (tc) Flow open event >> > >>>>>> >>>> cpumap/0/map:4-1371 [000] D..2. 515.399239: bpf_tra= ce_printk: (tc) Send performance event (5,1), 374696 >> > >>>>>> >>>> cpumap/0/map:4-1371 [000] D..2. 515.399466: bpf_tra= ce_printk: (tc) Flow open event >> > >>>>>> >>>> cpumap/0/map:4-1371 [000] D..2. 515.399475: bpf_tra= ce_printk: (tc) Send performance event (5,1), 247069 >> > >>>>>> >>>> cpumap/0/map:4-1371 [000] D..2. 516.405151: bpf_tra= ce_printk: (tc) Send performance event (5,1), 5217155 >> > >>>>>> >>>> cpumap/0/map:4-1371 [000] D..2. 517.405248: bpf_tra= ce_printk: (tc) Send performance event (5,1), 4515394 >> > >>>>>> >>>> cpumap/0/map:4-1371 [000] D..2. 518.406117: bpf_tra= ce_printk: (tc) Send performance event (5,1), 4481289 >> > >>>>>> >>>> cpumap/0/map:4-1371 [000] D..2. 519.406255: bpf_tra= ce_printk: (tc) Send performance event (5,1), 4255268 >> > >>>>>> >>>> cpumap/0/map:4-1371 [000] D..2. 520.407864: bpf_tra= ce_printk: (tc) Send performance event (5,1), 5249493 >> > >>>>>> >>>> cpumap/0/map:4-1371 [000] D..2. 521.406664: bpf_tra= ce_printk: (tc) Send performance event (5,1), 3795993 >> > >>>>>> >>>> cpumap/0/map:4-1371 [000] D..2. 522.407469: bpf_tra= ce_printk: (tc) Send performance event (5,1), 3949519 >> > >>>>>> >>>> cpumap/0/map:4-1371 [000] D..2. 523.408126: bpf_tra= ce_printk: (tc) Send performance event (5,1), 4365335 >> > >>>>>> >>>> cpumap/0/map:4-1371 [000] D..2. 524.408929: bpf_tra= ce_printk: (tc) Send performance event (5,1), 4154910 >> > >>>>>> >>>> cpumap/0/map:4-1371 [000] D..2. 525.410048: bpf_tra= ce_printk: (tc) Send performance event (5,1), 4405582 >> > >>>>>> >>>> cpumap/0/map:4-1371 [000] D..2. 525.434080: bpf_tra= ce_printk: (tc) Send flow event >> > >>>>>> >>>> cpumap/0/map:4-1371 [000] D..2. 525.482714: bpf_tra= ce_printk: (tc) Send flow event >> > >>>>>> >>>> >> > >>>>>> >>>> The times haven't been tweaked yet. The (5,1) is tc handle= major/minor, allocated by the xdp-cpumap parent. >> > >>>>>> >>>> I get pretty low latency between VMs; I'll set up a test w= ith some real-world data very soon. >> > >>>>>> >>>> >> > >>>>>> >>>> I plan to keep hacking away, but feel free to take a peek. >> > >>>>>> >>>> >> > >>>>>> >>>> Thanks, >> > >>>>>> >>>> Herbert >> > >>>>>> >>>> >> > >>>>>> >>>> On Mon, Oct 17, 2022 at 10:14 AM Simon Sundberg wrote: >> > >>>>>> >>>>> >> > >>>>>> >>>>> Hi, thanks for adding me to the conversation. Just a coup= le of quick >> > >>>>>> >>>>> notes. >> > >>>>>> >>>>> >> > >>>>>> >>>>> On Mon, 2022-10-17 at 16:13 +0200, Toke H=C3=B8iland-J=C3= =B8rgensen wrote: >> > >>>>>> >>>>> > [ Adding Simon to Cc ] >> > >>>>>> >>>>> > >> > >>>>>> >>>>> > Herbert Wolverson via LibreQoS writes: >> > >>>>>> >>>>> > >> > >>>>>> >>>>> > > Hey, >> > >>>>>> >>>>> > > >> > >>>>>> >>>>> > > I've had some pretty good success with merging xdp-pp= ing ( >> > >>>>>> >>>>> > > https://github.com/xdp-project/bpf-examples/blob/mast= er/pping/pping.h ) >> > >>>>>> >>>>> > > into xdp-cpumap-tc ( https://github.com/xdp-project/x= dp-cpumap-tc ). >> > >>>>>> >>>>> > > >> > >>>>>> >>>>> > > I ported over most of the xdp-pping code, and then ch= anged the entry point >> > >>>>>> >>>>> > > and packet parsing code to make use of the work alrea= dy done in >> > >>>>>> >>>>> > > xdp-cpumap-tc (it's already parsed a big chunk of the= packet, no need to do >> > >>>>>> >>>>> > > it twice). Then I switched the maps to per-cpu maps, = and had to pin them - >> > >>>>>> >>>>> > > otherwise the two tc instances don't properly share d= ata. >> > >>>>>> >>>>> > > >> > >>>>>> >>>>> >> > >>>>>> >>>>> I guess the xdp-cpumap-tc ensures that the same flow is p= rocessed on >> > >>>>>> >>>>> the same CPU core at both ingress or egress. Otherwise, i= f a flow may >> > >>>>>> >>>>> be processed by different cores on ingress and egress the= per-CPU maps >> > >>>>>> >>>>> will not really work reliably as each core will have a di= fferent view >> > >>>>>> >>>>> on the state of the flow, if there's been a previous pack= et with a >> > >>>>>> >>>>> certain TSval from that flow etc. >> > >>>>>> >>>>> >> > >>>>>> >>>>> Furthermore, if a flow is always processed on the same co= re (on both >> > >>>>>> >>>>> ingress and egress) I think per-CPU maps may be a bit was= teful on >> > >>>>>> >>>>> memory. From my understanding the keys for per-CPU maps a= re still >> > >>>>>> >>>>> shared across all CPUs, it's just that each CPU gets its = own value. So >> > >>>>>> >>>>> all CPUs will then have their own data for each flow, but= it's only the >> > >>>>>> >>>>> CPU processing the flow that will have any relevant data = for the flow >> > >>>>>> >>>>> while the remaining CPUs will just have an empty state fo= r that flow. >> > >>>>>> >>>>> Under the same assumption that packets within the same fl= ow are always >> > >>>>>> >>>>> processed on the same core there should generally not be = any >> > >>>>>> >>>>> concurrency issues with having a global (non-per-CPU) eit= her as packets >> > >>>>>> >>>>> from the same flow cannot be processed concurrently then = (and thus no >> > >>>>>> >>>>> concurrent access to the same value in the map). I am how= ever still >> > >>>>>> >>>>> very unclear on if there's any considerable performance i= mpact between >> > >>>>>> >>>>> global and per-CPU map versions if the same key is not ac= cessed >> > >>>>>> >>>>> concurrently. >> > >>>>>> >>>>> >> > >>>>>> >>>>> > > Right now, output >> > >>>>>> >>>>> > > is just stubbed - I've still got to port the perfmap = output code. Instead, >> > >>>>>> >>>>> > > I'm dumping a bunch of extra data to the kernel debug= pipe, so I can see >> > >>>>>> >>>>> > > roughly what the output would look like. >> > >>>>>> >>>>> > > >> > >>>>>> >>>>> > > With debug enabled and just logging I'm now getting a= bout 4.9 Gbits/sec on >> > >>>>>> >>>>> > > single-stream iperf between two VMs (with a shaper VM= in the middle). :-) >> > >>>>>> >>>>> > >> > >>>>>> >>>>> > Just FYI, that "just logging" is probably the biggest s= ource of >> > >>>>>> >>>>> > overhead, then. What Simon found was that sending the d= ata from kernel >> > >>>>>> >>>>> > to userspace is one of the most expensive bits of eppin= g, at least when >> > >>>>>> >>>>> > the number of data points goes up (which is does as add= itional flows are >> > >>>>>> >>>>> > added). >> > >>>>>> >>>>> >> > >>>>>> >>>>> Yhea, reporting individual RTTs when there's lots of them= (you may get >> > >>>>>> >>>>> upwards of 1000 RTTs/s per flow) is not only problematic = in terms of >> > >>>>>> >>>>> direct overhead from the tool itself, but also becomes de= manding for >> > >>>>>> >>>>> whatever you use all those RTT samples for (i.e. need to = log, parse, >> > >>>>>> >>>>> analyze etc. a very large amount of RTTs). One way to dea= l with that is >> > >>>>>> >>>>> of course to just apply some sort of sampling (the -r/--r= ate-limit and >> > >>>>>> >>>>> -R/--rtt-rate >> > >>>>>> >>>>> > >> > >>>>>> >>>>> > > So my question: how would you prefer to receive this = data? I'll have to >> > >>>>>> >>>>> > > write a daemon that provides userspace control (perio= dic cleanup as well as >> > >>>>>> >>>>> > > reading the performance stream), so the world's kinda= our oyster. I can >> > >>>>>> >>>>> > > stick to Kathie's original format (and dump it to a n= amed pipe, perhaps?), >> > >>>>>> >>>>> > > a condensed format that only shows what you want to u= se, an efficient >> > >>>>>> >>>>> > > binary format if you feel like parsing that... >> > >>>>>> >>>>> > >> > >>>>>> >>>>> > It would be great if we could combine efforts a bit her= e so we don't >> > >>>>>> >>>>> > fork the codebase more than we have to. I.e., if "upstr= eam" epping and >> > >>>>>> >>>>> > whatever daemon you end up writing can agree on data fo= rmat etc that >> > >>>>>> >>>>> > would be fantastic! Added Simon to Cc to facilitate thi= s :) >> > >>>>>> >>>>> > >> > >>>>>> >>>>> > Briefly what I've discussed before with Simon was to ha= ve the ability to >> > >>>>>> >>>>> > aggregate the metrics in the kernel (WiP PR [0]) and ha= ve a userspace >> > >>>>>> >>>>> > utility periodically pull them out. What we discussed w= as doing this >> > >>>>>> >>>>> > using an LPM map (which is not in that PR yet). The ide= a would be that >> > >>>>>> >>>>> > userspace would populate the LPM map with the keys (pre= fixes) they >> > >>>>>> >>>>> > wanted statistics for (in LibreQOS context that could b= e one key per >> > >>>>>> >>>>> > customer, for instance). Epping would then do a map loo= kup into the LPM, >> > >>>>>> >>>>> > and if it gets a match it would update the statistics i= n that map entry >> > >>>>>> >>>>> > (keeping a histogram of latency values seen, basically)= . Simon's PR >> > >>>>>> >>>>> > below uses this technique where userspace will "reset" = the histogram >> > >>>>>> >>>>> > every time it loads it by swapping out two different ma= p entries when it >> > >>>>>> >>>>> > does a read; this allows you to control the sampling ra= te from >> > >>>>>> >>>>> > userspace, and you'll just get the data since the last = time you polled. >> > >>>>>> >>>>> >> > >>>>>> >>>>> Thank's Toke for summarzing both the current state and th= e plan going >> > >>>>>> >>>>> forward. I will just note that this PR (and all my other = work with >> > >>>>>> >>>>> ePPing/BPF-PPing/XDP-PPing/I-suck-at-names-PPing) will be= more or less >> > >>>>>> >>>>> on hold for a couple of weeks right now as I'm trying to = finish up a >> > >>>>>> >>>>> paper. >> > >>>>>> >>>>> >> > >>>>>> >>>>> > I was thinking that if we all can agree on the map form= at, then your >> > >>>>>> >>>>> > polling daemon could be one userspace "client" for that= , and the epping >> > >>>>>> >>>>> > binary itself could be another; but we could keep compa= tibility between >> > >>>>>> >>>>> > the two, so we don't duplicate effort. >> > >>>>>> >>>>> > >> > >>>>>> >>>>> > Similarly, refactoring of the epping code itself so it = can be plugged >> > >>>>>> >>>>> > into the cpumap-tc code would be a good goal... >> > >>>>>> >>>>> >> > >>>>>> >>>>> Should probably do that...at some point. In general I thi= nk it's a bit >> > >>>>>> >>>>> of an interesting problem to think about how to chain mul= tiple XDP/tc >> > >>>>>> >>>>> programs together in an efficent way. Most XDP and tc pro= grams will do >> > >>>>>> >>>>> some amount of packet parsing and when you have many chai= ned programs >> > >>>>>> >>>>> parsing the same packets this obviously becomes a bit was= teful. In the >> > >>>>>> >>>>> same time it would be nice if one didn't need to manually= merge >> > >>>>>> >>>>> multiple programs together into a single one like this to= get rid of >> > >>>>>> >>>>> this duplicated parsing, or at least make that process of= merging those >> > >>>>>> >>>>> programs as simple as possible. >> > >>>>>> >>>>> >> > >>>>>> >>>>> >> > >>>>>> >>>>> > -Toke >> > >>>>>> >>>>> > >> > >>>>>> >>>>> > [0] https://github.com/xdp-project/bpf-examples/pull/59 >> > >>>>>> >>>>> >> > >>>>>> >>>>> N=C3=A4r du skickar e-post till Karlstads universitet beh= andlar vi dina personuppgifter. >> > >>>>>> >>>>> When you send an e-mail to Karlstad University, we will p= rocess your personal data. >> > >>>>>> >>>> >> > >>>>>> >>>> _______________________________________________ >> > >>>>>> >>>> LibreQoS mailing list >> > >>>>>> >>>> LibreQoS@lists.bufferbloat.net >> > >>>>>> >>>> https://lists.bufferbloat.net/listinfo/libreqos >> > >>>>>> >>> >> > >>>>>> >>> >> > >>>>>> >>> >> > >>>>>> >>> -- >> > >>>>>> >>> Robert Chac=C3=B3n >> > >>>>>> >>> CEO | JackRabbit Wireless LLC >> > >>>>>> > >> > >>>>>> > _______________________________________________ >> > >>>>>> > LibreQoS mailing list >> > >>>>>> > LibreQoS@lists.bufferbloat.net >> > >>>>>> > https://lists.bufferbloat.net/listinfo/libreqos >> > >>>>>> >> > >>>>>> >> > >>>>>> >> > >>>>>> -- >> > >>>>>> This song goes out to all the folk that thought Stadia would wo= rk: >> > >>>>>> https://www.linkedin.com/posts/dtaht_the-mushroom-song-activity= -6981366665607352320-FXtz >> > >>>>>> Dave T=C3=A4ht CEO, TekLibre, LLC >> > >>>> >> > >>>> _______________________________________________ >> > >>>> LibreQoS mailing list >> > >>>> LibreQoS@lists.bufferbloat.net >> > >>>> https://lists.bufferbloat.net/listinfo/libreqos >> > >>> >> > >>> >> > >>> >> > >>> -- >> > >>> Robert Chac=C3=B3n >> > >>> CEO | JackRabbit Wireless LLC >> > >>> _______________________________________________ >> > >>> LibreQoS mailing list >> > >>> LibreQoS@lists.bufferbloat.net >> > >>> https://lists.bufferbloat.net/listinfo/libreqos >> > > >> > > _______________________________________________ >> > > LibreQoS mailing list >> > > LibreQoS@lists.bufferbloat.net >> > > https://lists.bufferbloat.net/listinfo/libreqos >> > >> > >> > >> > -- >> > This song goes out to all the folk that thought Stadia would work: >> > https://www.linkedin.com/posts/dtaht_the-mushroom-song-activity-698136= 6665607352320-FXtz >> > Dave T=C3=A4ht CEO, TekLibre, LLC >> >> >> >> -- >> This song goes out to all the folk that thought Stadia would work: >> https://www.linkedin.com/posts/dtaht_the-mushroom-song-activity-69813666= 65607352320-FXtz >> Dave T=C3=A4ht CEO, TekLibre, LLC > > _______________________________________________ > LibreQoS mailing list > LibreQoS@lists.bufferbloat.net > https://lists.bufferbloat.net/listinfo/libreqos --=20 This song goes out to all the folk that thought Stadia would work: https://www.linkedin.com/posts/dtaht_the-mushroom-song-activity-69813666656= 07352320-FXtz Dave T=C3=A4ht CEO, TekLibre, LLC