Date: Fri, 05 Feb 2021 05:31:59 +0000 From: bugzilla-noreply@freebsd.org To: bugs@FreeBSD.org Subject: [Bug 253261] AWS EC2 disk I/O performance is extremely bad Message-ID: <bug-253261-227@https.bugs.freebsd.org/bugzilla/>
next in thread | raw e-mail | index | archive | help
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=3D253261 Bug ID: 253261 Summary: AWS EC2 disk I/O performance is extremely bad Product: Base System Version: 12.2-RELEASE Hardware: amd64 OS: Any Status: New Severity: Affects Many People Priority: --- Component: kern Assignee: bugs@FreeBSD.org Reporter: raj@gusw.net I am trying to move all my AWS systems back from Amazon Linux to FreeBSD. B= ut I encounter massive slowness. People have reported about slow ENA network driver before.=20 Even worse is the performance of the NVMe nvd disks. This is so bad that my attempt to go with FreeBSD is now completely dead. And it makes me really s= ad. FreeBSD was once "the power to serve" and now it feels like I'm running Ult= rix on a VAX for the fun of it but for any real work I need Linux. So sad! The issue can be reproduced by anyone on AWS. Just get a FreeBSD 12 (latest) instance and allocate a gp2 of 1 TB or a gp3 of smaller size for which you specify 3000 IOPS and 250 MB/s IO capacity. Let's say that disk is /dev/nvd2 Now do this: dd if=3D/dev/zero of=3D/dev/nvd2 bs=3D100M status=3Dprogress and you see that it's writing with the whopping speed of 70 MB/s.=20 That used to be good, but it is no longer good. Compare Amazon Linux doing = the same thing at 300 MB/s. Now, when you put a file system over it, zfs or ufs, then instantly the performance gets better: newfs /dev/nvd2 mount /dev/nvd2 /mnt dd if=3D/dev/zero of=3D/mnt/test bs=3D100M status=3Dprogress now that works at about 250 MB/s. Decent. So, problem solved? No! It turns out if I create a PostgreSQL database over this setup, then ag= ain there is massive delay on the read and write and throughput will drop to ev= en worse than 70 MB/s. Creating one index takes 10 times as long as that same = on the Linux system. The dd issue is probably the best manifestation for fixing at least part of= the problem.=20 Is this about tuning? Even if it was, there is a severe problem with how the system is set up by default. But I'm afraid it's not even about tuning so b= ad is it. --=20 You are receiving this mail because: You are the assignee for the bug.=
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?bug-253261-227>