.. should have written "making the cpus run hotter and clock lower and harming single-thread performance without significantly increasing total throughput, sometimes even harming total throughput" - and making it harder to see how much *actual* cpu is being used by a process (a process seemingly using 50% of a physical core might actually be using near 100% of a core, but you can't see it in htop because HT makes it look like 50%/the-ht-core-looks-ready-to-go-even-if-it-isn't) - comparatively AMD's SMT seems to nearly always increase total throughput.
On Fri, 22 Jul 2022 at 23:56, Hans Henrik Bergan <divinit...@gmail.com> wrote: > 10 days of radio silence? What's going on? anyway benchmarks (countsey of > bench.sh ) of the server/bandwidth > > -------------------- A Bench.sh Script By Teddysun ------------------- > Version : v2022-06-01 > Usage : wget -qO- bench.sh | bash > ---------------------------------------------------------------------- > CPU Model : Intel(R) Xeon(R) CPU E7- 4870 @ 2.40GHz > CPU Cores : 40 @ 2524.597 MHz > CPU Cache : 30720 KB > AES-NI : Enabled > VM-x/AMD-V : Enabled > Total Disk : 393.6 GB (48.7 GB Used) > Total Mem : 125.9 GB (12.0 GB Used) > Total Swap : 8.0 GB (0 Used) > System uptime : 10 days, 2 hour 11 min > Load average : 40.18, 40.10, 40.13 > OS : Ubuntu 20.04.4 LTS > Arch : x86_64 (64 Bit) > Kernel : 5.4.0-122-generic > TCP CC : cubic > Virtualization : Dedicated > Organization : AS31798 DataCity > Location : Kitchener / CA > Region : Ontario > ---------------------------------------------------------------------- > I/O Speed(1st run) : 812 MB/s > I/O Speed(2nd run) : 841 MB/s > I/O Speed(3rd run) : 821 MB/s > I/O Speed(average) : 824.7 MB/s > ---------------------------------------------------------------------- > Node Name Upload Speed Download Speed Latency > Speedtest.net 941.15 Mbps 829.87 Mbps 3.09 ms > Los Angeles, US 937.01 Mbps 903.15 Mbps 64.21 ms > Dallas, US 941.23 Mbps 898.69 Mbps 36.36 ms > Montreal, CA 888.80 Mbps 916.91 Mbps 11.09 ms > Paris, FR 886.36 Mbps 820.77 Mbps 90.50 ms > Amsterdam, NL 854.88 Mbps 770.73 Mbps 98.16 ms > Shanghai, CN 426.16 Mbps 833.23 Mbps 216.21 ms > Nanjing, CN 439.53 Mbps 542.91 Mbps 198.60 ms > Seoul, KR 465.37 Mbps 367.20 Mbps 216.88 ms > Singapore, SG 417.93 Mbps 132.76 Mbps 212.71 ms > Tokyo, JP 668.62 Mbps 676.35 Mbps 134.48 ms > ---------------------------------------------------------------------- > Finished in : 5 min 35 sec > Timestamp : 2022-07-22 21:35:37 UTC > ---------------------------------------------------------------------- > > - it incorrectly says 393GB because of partitioning, it's actually 1TB > > - I've intentionally disabled Hyperthreading because... in my > opinion/experience, hyperthreading implementations in older Intel CPUs > often did more harm than good, making the cpus run hotter and clock lower, > this is a (high-end) 2011 model. (I have the opposite experience with AMD > SMT fwiw) > > > > On Thu, 14 Jul 2022 at 01:18, Hans Henrik Bergan <divinit...@gmail.com> > wrote: > >> Fwiw Chris Haas / vendiadvertising.com has reached out, they're willing >> to sponsor raid(1) if my proposal is accepted. >> Haas is reading this mailing list, but does not wish to participate/post >> directly at present. >> (I have no prior relation, never heard of them before today) >> >> On Tue, 12 Jul 2022 at 22:31, Hans Henrik Bergan <divinit...@gmail.com> >> wrote: >> >>> i have a 40-core 128GB RAM server with 1 (and only 1) static ipv4 >>> address that I intend to keep around until at least 27 november 2025 (but >>> unsure after that), >>> it has more RAM/CPU than i need, and i could set up a virtual machine on >>> it and give VNC access to the VM in the form of >>> > ssh -L 9999:localhost:3389 user@ip >>> then point a RDP/VNC client to localhost:9999 >>> i can dedicate individual TCP ports to it, but not any of 22, 80, 443, >>> 5900, 5901, 8083, 9999, 587, 2525 >>> (after which i can recommend Teamviewer for a more comfortable access) >>> >>> but the server has several caveats: >>> - it's un-raided, rolling just a single 1TB SSD, if that SSD was to die, >>> everything on it would be lost, and it would probably take days to replace >>> the drive. >>> - it's located in a somewhat unstable environment, occasional power >>> outage, it was out on 30 june, 23 june, 27 january, and in 2021 was down on >>> 30 august. (it boots up and recovers automatically, but it has a reboot >>> time of roughly 300 seconds until ssh responds again) >>> - I can not offer a dedicated IP (but can offer some dedicated ports) >>> - May be decommissioned after 27-11-2025 >>> >>> If that sounds like a good idea, I'll want a ssh public key. >>> >>> >>> On Tue, 12 Jul 2022 at 15:38, Christoph M. Becker <cmbecke...@gmx.de> >>> wrote: >>> >>>> On 11.07.2022 at 21:04, Hans Henrik Bergan wrote: >>>> >>>> > Any of you windows.php.net guys familiar with SSH? Can one be >>>> created? >>>> >>>> That would be possible (needed to be set-up by Alex Schoenmaker), but >>>> while that could solve the upload issue, it wouldn't solve the other >>>> issue I've mentioned, namely that we need to avoid dynamically linking >>>> against incompatible versions of dependencies. And there is yet another >>>> issue, namely security. I think these issues can only be solved by some >>>> central place where the packages are build. >>>> >>>> > On Mon, 11 Jul 2022 at 19:52, Christoph M. Becker <cmbecke...@gmx.de> >>>> wrote: >>>> > >>>> >> On 11.07.2022 at 18:53, Hans Henrik Bergan wrote: >>>> >> >>>> >>> Is there any SSH public key associated with the Windows php >>>> developers? >>>> >> Or >>>> >>> with CMB? >>>> >> >>>> >> No. At least not regarding windows.php.net. >>>> >> >>>> >>> On Mon, 11 Jul 2022 at 18:25, Christoph M. Becker < >>>> cmbecke...@gmx.de> >>>> >> wrote: >>>> >>> >>>> >>>> On 11.07.2022 at 17:41, Hans Henrik Bergan wrote: >>>> >>>> >>>> >>>>> Do you mean it is unlikely that Github Actions will suffice? >>>> because >>>> >> that >>>> >>>>> would definitely be a great solution if it was possible/feasible . >>>> >>>> >>>> >>>> No. It is only unlikely that automated uploads to window.php.net >>>> will >>>> >>>> ever be possible using GH actions. >>>> >>>> >>>> >>>>> Anyway, with regards to "I don't think there are special >>>> requirements >>>> >>>>> regarding the hardware", can you give a rough estimate on RAM/disk >>>> >> space >>>> >>>>> requirements? >>>> >>>> >>>> >>>> The old machine had something like 4 or 8 GB RAM, and a 128 GB hard >>>> >> disk. >>>> >>>> >>>> >>> >>>> >> >>>> > >>>> >>>>