Re: [fedora-arm] builder io isue
On 12/25/2011 06:16 AM, Brendan Conoboy wrote: On 12/24/2011 09:25 PM, Dennis Gilmore wrote: Just throwing out what im seeing. lets see what ideas we can come up with. performance is better than before. and seneca has done a great job and put a lot of hard work into the reorg and im thankful for that. We just have another bottleneck to address. Allocating builders to individual rather than a single raid volume will help dramatically. Care to explain why? Since hfp/sfp F15 builds happen at the same time, having 2 hfp disks and 2 sfp disks is a good start. Split it up again by some other criteria- you want to try to ensure that any one builder only causes one spindle to be used and any 2 builders each only cause 2 spindles to be used (1-to-1), and so forth. As long as any single spindle isn't doing multiple simultaneous mock inits it should scale pretty well. You can largely avoid this by making sure your RAID and FS are aligned. Most mock IOPS are small, and provided layers are aligned right, should scale very well on RAID 0/10 (RAID5/6 will inevitably cripple just about anything when it comes to writes regardless of what you do, unless the vast majority of your writes is a multiple of your stripe-width, which is unlikely). Allocate the large files the builders are using all at once (no sparse files) to ensure large read/writes don't require seeks. Is this a "proper" SAN or just another Linux box with some disks in it? Is NFS backed by a SAN "volume"? Use the async nfs export option if it isn't already in use. +1 Disable fs journaling (normally dangerous, but this is throw-away space). Not really dangerous - the only danger is that you might have to wait for fsck to do it's thing on an unclean shutdown (which can take hours on a full TB scale disk, granted). Speaking of "dangerous" tweaks, you could LD_PRELOAD libeatmydata (add to a profile.d file in the mock config, and add the package to the buildsys-build group). That will eat fsync() calls which will smooth out commits and make a substantial difference to performance. Since it's scratch space anyway it doesn't matter WRT safety. Use noatime mounts if that won't break package builds checking filesystem conformance. Build of zsh will break on NFS whateveryou do. It will also break on a local FS with noatime. There may be other packages that suffer from this issue but I don't recall them off the top of my head. Anyway, that is an issue for a build policy - have one builder using block level storage with atime and the rest on NFS. Once all that is done, tweak the number of nfsds such that there are as many as possible without most of them going into deep sleep. Perhaps somebody else can suggest some optimal sysctl and ext4fs settings? As mentioned in a previous post, have a look here: http://www.altechnative.net/?p=96 Deadline scheduler might also help on the NAS/SAN end, plus all the usual tweaks (e.g. make sure write caches on the disks are enabled, if the disks support write-read-verify disable it, etc.) Gordan ___ arm mailing list arm@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/arm
Re: [fedora-arm] builder io isue
On 12/25/2011 05:25 AM, Dennis Gilmore wrote: When we were testing build were happening really fast, once we loaded up the build jobs things have become really slow http://arm.koji.fedoraproject.org/koji/taskinfo?taskID=238672 started at 1:19 utc and at 5:18 utc four hours later the buildrequires are still being installed. australia is completely io bound. i think we need to see how we can spread the io load. ~30 hosts reading and writing to the 4 spindles just saturates the disk io. Is your RAID aligned properly? I have found that making sure that your RAID and FS are aligned properly makes a big difference to performance, at least on my workloads. Have a look here: http://www.altechnative.net/?p=96 Just taking the default options when setting up the RAID and FS typically hammer the IOPS performance down to that of a single disk. i guess options are find a way to add more spindles. move /var/lib/mock to sdcard, see if we can get some kind of san that has alot of smaller fast disks. That will make it an order of magnitude worse. A typical class 10 SD card only manages about 20 random write IOPS. get some 2.5" usb drives one for each builder. some other idea? is there anyway we could add 4-8 more disks to australia. the size of the matters little. gaining more iops by adding more spindles would help. You could just replace the disks with some decent SSDs and be done with it (aligning the RAID and FS as explained on the above linked page still applies just the same). Especially since 15K rpm disks are now highly uncompetitively priced against SSDs. Gordan ___ arm mailing list arm@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/arm
[fedora-arm] Daily F15 ARM-Koji Build Status
Period: Sat Dec 24 13:05:02 UTC 2011 -- Sun Dec 25 13:05:03 UTC 2011 Errors UnbuiltDifferent WorkingBuilt --- 1118 [-8] 2494 [-30] 57 [0] 12 [-2]6734 [+40] * Different contains packages that are built but have a varying ENVR to that of PA-Koji * The error count is likely to drop as the packages are probably just waiting for a rebuild attempt * The numbers in the brackets show the difference recorded in the counts since the script was last run * https://fedoraproject.org/wiki/Architectures/ARM/F15_Koji_build * http://arm.koji.fedoraproject.org/~jchiappetta/stats.html ___ arm mailing list arm@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/arm
[fedora-arm] Fedora ARM F-15 Branched report: 20111125 changes
Compose started at Sun Dec 25 19:54:45 UTC 2011 Loaded plugins: langpacks, presto, refresh-packagekit New package: artha-1.0.1-3.fc15 A handy thesaurus based on WordNet New package: aspell-es-1.11-2.fc15 Spanish dictionaries for Aspell New package: aspell-ru-0.99f7-7.fc15 Russian dictionaries for Aspell New package: bidiv-1.5-10.fc15 Display logical Hebrew on unidirectional terminals New package: bind-dyndb-ldap-0.2.0-3.fc15 LDAP back-end plug-in for BIND New package: bltk-1.0.9-11.fc15 The BLTK measures notebook battery life under any workload New package: cd-discid-1.1-2.fc15 Utility to get CDDB discid information New package: couchdb-glib-0.7.2-1.fc15 A glib api to access CouchDB servers New package: dotconf-1.3-2.fc15 Libraries to parse configuration files New package: dvgrab-3.5-2.fc15 Utility to capture video from a DV camera New package: dwdiff-1.7-4.fc15 Front end to diff for comparing on a per word basis New package: ez-ipupdate-3.0.11-0.23.b8.fc15 Client for Dynamic DNS Services New package: fbdesk-1.4.1-7.fc15 Icon Manager for Fluxbox New package: ftnchek-3.3.1-12.fc15 Static analyzer for Fortran 77 programs New package: gdb-heap-0.5-4.fc15 Extensions to gdb for debugging dynamic memory allocation New package: ghasher-1.2.1-10.fc15 GUI hasher for GTK+ 2 New package: globalplatform-5.0.0-10.fc15 Library for access to OP 2.0.1 and GP 2.1.1 conforming smart cards New package: gnome-device-manager-0.2-6.fc15 Graphical Device Manager Application New package: gnome-libs-1.4.2-18.fc15 The main GNOME1 libraries New package: gtk-vnc-0.4.3-1.fc15 A GTK2 widget for VNC clients New package: gtkmm24-2.24.0-3.fc15 C++ interface for GTK2 (a GUI library for X) New package: hostapd-0.7.3-2.fc15 IEEE 802.11 AP, IEEE 802.1X/WPA/WPA2/EAP/RADIUS Authenticator New package: http-parser-0.3-6.20100911git.fc15 HTTP request/response parser for C New package: httrack-3.43.9-3.fc15 Website copier and offline browser New package: ifstatus-1.1.0-7.fc15 Command line real time interface graphs using ncurses New package: isomaster-1.3.7-2.fc15 An easy to use GUI CD image editor New package: ladspa-vco-plugins-0.3.0-9.fc15 Anti-aliased pulse and sawtooth oscillators New package: libUnihan-0.5.3-6.fc15 C library for Unihan character database in fifth normal form New package: libXrandr-1.3.1-2.fc15 X.Org X11 libXrandr runtime library New package: libeio-3.65-4.fc15 Event-based fully asynchronous I/O library New package: libgnome-2.32.1-2.fc15 GNOME base library New package: libgnomeui-2.24.5-2.fc15 GNOME base GUI library New package: libibmad-1.3.4-2.fc15 OpenFabrics Alliance InfiniBand MAD library New package: libibumad-1.3.4-2.fc15 OpenFabrics Alliance InfiniBand umad (user MAD) library New package: liblo-0.26-2.fc15 Open Sound Control library New package: libwpd-0.9.1-2.fc15 Library for reading and converting WordPerfect documents New package: libxnm-0.1.3-4.fc15 A library for parsing the XNM format New package: lv2-swh-plugins-1.0.15-5.20091118git.fc15 LV2 ports of LADSPA swh plugins New package: lxdm-0.3.0-4.fc15 Lightweight X11 Display Manager New package: maloc-1.5-1.fc15 Minimal Abstraction Layer for Object-oriented C New package: mutt-1.5.21-3.fc15 A text mode mail user agent New package: nxtvepg-2.8.1-6.fc15 A nexTView EPG decoder and browser New package: ocaml-json-static-0.9.8-4.fc15 OCaml JSON validator and converter (syntax extension) New package: olpc-contents-2.6-3.fc15 OLPC contents manifest tools New package: openbox-3.4.11.2-8.fc15 A highly configurable and standards-compliant X11 window manager New package: perl-BerkeleyDB-0.43-5.fc15 Interface to Berkeley DB New package: perl-Math-Random-ISAAC-XS-1.004-2.fc15 C implementation of the ISAAC PRNG algorithm New package: perl-Mozilla-LDAP-1.5.3-5.fc15 LDAP Perl module that wraps the OpenLDAP C SDK New package: perl-Net-DBus-GLib-0.33.0-7.fc15 Perl extension for the DBus GLib bindings New package: pngcrush-1.6.10-7.fc15 Optimizer for PNG (Portable Network Graphics) files New package: pngnq-0.5-9.fc15 Pngnq is a tool for quantizing PNG images in RGBA format New package: poster-20060221-8.fc15 Scales PostScript images to span multiple pages New package: pyhunspell-0.1-4.fc15 Python bindings for hunspell New package: python-isprelink-0.1.2-
[fedora-arm] Fedora ARM F-15 Branched report: 201111251 changes
Compose started at Sun Dec 25 22:46:21 UTC 2011 Loaded plugins: langpacks, presto, refresh-packagekit New package: acpi-1.5-2.fc15 Command-line ACPI client New package: aspell-0.60.6-14.fc15 Spell checker New package: bindfs-1.8.3-4.fc15 Fuse filesystem to mirror a directory New package: dvipdfm-0.13.2d-42.fc15 A DVI to PDF translator New package: ftplib-3.1-7.fc15 Library of FTP routines New package: indent-2.2.11-3.fc15 A GNU program for formatting C code New package: lensfun-0.2.5-5.fc15 Library to rectify defects introduced by photographic lenses New package: mod_security-2.5.12-4.fc15 Security module for the Apache HTTP Server New package: mon-1.2.0-7.fc15 General-purpose resource monitoring system New package: ocaml-csv-1.1.7-9.fc15 OCaml library for reading and writing CSV files New package: perl-B-Compiling-0.02-4.fc15 Expose PL_compiling to perl New package: perl-PerlIO-gzip-0.18-10.fc15 Perl extension to provide a PerlIO layer to gzip/gunzip New package: perl-Unicode-Map8-0.13-4.fc15 Mapping table between 8-bit chars and Unicode for Perl New package: psad-2.1.7-2.fc15 Port Scan Attack Detector (psad) watches for suspect traffic Updated Packages: kernel-2.6.41.5-5.fc15 -- * Tue Dec 20 2011 Josh Boyer - Fix config options in arm configs after latest commits - Backport upstream fix for b44_poll oops (rhbz #741117) * Mon Dec 19 2011 Dave Jones - x86, dumpstack: Fix code bytes breakage due to missing KERN_CONT * Thu Dec 15 2011 Dennis Gilmore - build imx highbank and kirkwood kernel variants on arm - add patch for tegra usb storage resets - omap config cleanups from dmarlin * Thu Dec 15 2011 Josh Boyer - 2.6.41.5-4 - Add patch to fix Intel wifi regression in 3.1.5 (rhbz 767173) - Add patch from Jeff Layton to fix suspend with NFS (rhbz #717735) - Backport ALPS touchpad patches from input/next branch (rhbz #590880) * Thu Dec 15 2011 Dave Jones - 2.6.41.5-3 - Disable Intel IOMMU by default. * Fri Dec 09 2011 Josh Boyer 2.6.41.5-1 - Linux 3.1.5 (Fedora 2.6.41.5) * Thu Dec 08 2011 Chuck Ebbert 2.6.41.5-0.rc2.1 - Linux 3.1.5-rc2 (Fedora 2.6.41.5-rc2) - Fix wrong link speed on some sky2 network adapters (rhbz #757839) * Wed Dec 07 2011 Chuck Ebbert - Linux 3.1.5-rc1 (Fedora 2.6.41.5-rc1) - Comment out merged patches: xfs-Fix-possible-memory-corruption-in-xfs_readlink.patch rtlwifi-fix-lps_lock-deadlock.patch * Tue Dec 06 2011 Chuck Ebbert - Disable uas until someone can fix it (rhbz #717633) * Tue Dec 06 2011 Josh Boyer - Add reworked pci ASPM patch from Matthew Garrett Summary: Added Packages: 14 Removed Packages: 0 Modified Packages: 1 Compose finshed at Sun Dec 25 23:36:49 UTC 2011 ___ arm mailing list arm@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/arm
Re: [fedora-arm] builder io isue
On 12/25/2011 03:47 AM, Gordan Bobic wrote: On 12/25/2011 06:16 AM, Brendan Conoboy wrote: Allocating builders to individual rather than a single raid volume will help dramatically. Care to explain why? Sure, see below. Is this a "proper" SAN or just another Linux box with some disks in it? Is NFS backed by a SAN "volume"? As I understand it, the server is a Linux host using raid0 with 512k chunks across 4 sata drives. This md device is then formatted with some filesystem (ext4?). Directories on this filesystem are then exported to individual builders such that each builder has its own private space. These private directories contain a large file that is used as a loopback ext4fs (IE, the builder mounts the nfs share, then loopback mounts the file on that nfs share as an ext4fs). This is where /var/lib/mock comes from. Just to be clear, if you looked at nfs mounted directory on a build host you would see a single large file that represented a filesystem, making traditional ext?fs tuning a bit more complicated. The structural complication is that we have something like 30-40 systems all vying for the attention of those 4 spindles. It's really important that each builder not cause more than one disk to perform an operation because seeks are costly, and if just 2 disks get called up by a single builder, 50% of the storage resources will be taken up by a single host until the operation completes. With 40 hosts, you'll just end up thrashing (with considerably fewer hosts, too).. Raid0 gives great throughput, but it's at the cost of latency. With so many 100mbit builders, throughput is less important and latency is key. Roughly put, the two goals for good performance in this scenario are: 1. Make sure each builder only activates one disk per operation. 2. Make sure each io operation causes the minimum amount of seeking. You're right that good alignment and block sizes and whatnot will help this cause, but there is still greater likelihood of io operations traversing spindle boundaries periodically in the best situation. You'd need a chunk size about equal to the fs image file size to pull that off. Perhaps an lvm setup with strictly defined layouts with each lvcreate would make it a bit more manageable, but for simplicity's sake I advocate simply treating the 4 disks like 4 disks, exported according to expected usage patterns. In the end, if all this is done and the builders are delayed by deep sleeping nfsds, the only options are to move /var/lib/mock to local storage or increase the number of spindles on the server. Disable fs journaling (normally dangerous, but this is throw-away space). Not really dangerous - the only danger is that you might have to wait for fsck to do it's thing on an unclean shutdown (which can take hours on a full TB scale disk, granted). I mean dangerous in the sense that if the server goes down, there might be data loss, but the builders using the space won't know that. This is particularly true if nfs exports are async. Speaking of "dangerous" tweaks, you could LD_PRELOAD libeatmydata (add to a profile.d file in the mock config, and add the package to the buildsys-build group). That will eat fsync() calls which will smooth out commits and make a substantial difference to performance. Since it's scratch space anyway it doesn't matter WRT safety. Sounds good to me :-) Build of zsh will break on NFS whateveryou do. It will also break on a local FS with noatime. There may be other packages that suffer from this issue but I don't recall them off the top of my head. Anyway, that is an issue for a build policy - have one builder using block level storage with atime and the rest on NFS. Since loopback files representing filesystems are being used with nfs as the storage mechanism, this would probably be a non-issue. You just can't have the builder mount its loopback fs noatime (hadn't thought of that previously). Once all that is done, tweak the number of nfsds such that there are as many as possible without most of them going into deep sleep. Perhaps somebody else can suggest some optimal sysctl and ext4fs settings? As mentioned in a previous post, have a look here: http://www.altechnative.net/?p=96 Deadline scheduler might also help on the NAS/SAN end, plus all the usual tweaks (e.g. make sure write caches on the disks are enabled, if the disks support write-read-verify disable it, etc.) Definitely worth testing. Well ordered IO is critical here. -- Brendan Conoboy / Red Hat, Inc. / b...@redhat.com ___ arm mailing list arm@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/arm
Re: [fedora-arm] builder io isue
On 12/26/2011 03:57 AM, Brendan Conoboy wrote: On 12/25/2011 03:47 AM, Gordan Bobic wrote: On 12/25/2011 06:16 AM, Brendan Conoboy wrote: Allocating builders to individual rather than a single raid volume will help dramatically. Care to explain why? Sure, see below. Is this a "proper" SAN or just another Linux box with some disks in it? Is NFS backed by a SAN "volume"? As I understand it, the server is a Linux host using raid0 with 512k chunks across 4 sata drives. This md device is then formatted with some filesystem (ext4?). Directories on this filesystem are then exported to individual builders such that each builder has its own private space. These private directories contain a large file that is used as a loopback ext4fs (IE, the builder mounts the nfs share, then loopback mounts the file on that nfs share as an ext4fs). This is where /var/lib/mock comes from. Just to be clear, if you looked at nfs mounted directory on a build host you would see a single large file that represented a filesystem, making traditional ext?fs tuning a bit more complicated. Why not just mount direct via NFS? It'd be a lot quicker, not to mention easier to tune. It'd work for building all but a handful of packages (e.g. zsh), but you could handle that by having a single builder that uses a normal fs that has a policy pointing the packages that fail self-tests on NFS at it. The structural complication is that we have something like 30-40 systems all vying for the attention of those 4 spindles. It's really important that each builder not cause more than one disk to perform an operation because seeks are costly, and if just 2 disks get called up by a single builder, 50% of the storage resources will be taken up by a single host until the operation completes. With 40 hosts, you'll just end up thrashing (with considerably fewer hosts, too).. Raid0 gives great throughput, but it's at the cost of latency. With so many 100mbit builders, throughput is less important and latency is key. 512KB chunks sound vastly oversized for this sort of a workload. But if you are running ext4 on top of loopback file on top of NFS, no wonder the performance sucks. Roughly put, the two goals for good performance in this scenario are: 1. Make sure each builder only activates one disk per operation. Sounds like a better way to ensure that would be to re-architect the storage solution more sensibly. If you really want to use block level storage, use iSCSI on top of raw partitions. Providing those partitions are suitably aligned (e.g. for 4KB physical sector disks, erase block sizes, underlying RAID, etc.), your FS on top of those iSCSI exports will also end up being properly aligned, and the stride, stripe-width and block group size will all still line up properly. But with 40 builders, each builder only hammering one disk, you'll still get 10 builders hammering each spindle and causing a purely random seek pattern. I'd be shocked if you see any measurable improvement from just splitting up the RAID. 2. Make sure each io operation causes the minimum amount of seeking. You're right that good alignment and block sizes and whatnot will help this cause, but there is still greater likelihood of io operations traversing spindle boundaries periodically in the best situation. You'd need a chunk size about equal to the fs image file size to pull that off. Using the fs image over loopback over NFS sounds so eyewateringly wrong that I'm just going to give up on this thread if that part is immutable. I don't think the problem is significantly fixable if that approach remains. Perhaps an lvm setup with strictly defined layouts with each lvcreate would make it a bit more manageable, but for simplicity's sake I advocate simply treating the 4 disks like 4 disks, exported according to expected usage patterns. I don't see why you think that seeking within a single disk is any less problematic than seeking across multiple disks. That will only happen when the file exceeds the chunk size, and that will typically happen only at the end when linking - there aren't many cases where a single code file is bigger than a sensible chunk size (and in a 4-disk RAID0 case, you're pretty much forced to use a 32KB chunk size if you intend for the block group beginnings to be distributed across spindles). In the end, if all this is done and the builders are delayed by deep sleeping nfsds, the only options are to move /var/lib/mock to local storage or increase the number of spindles on the server. And local storage will be what? SD cards? There's only one model line of SD cards I have seen to date that actually produce random-write results that begin to approach a ~5000 rpm disk (up to 100 IOPS), and those are SLC and quite expensive. Having spent the last few months patching, fixing up and rebuilding RHEL6 packages for ARM, I have a pretty good understanding of what works for backing storage and what doesn't - and SD
Re: [fedora-arm] builder io isue
On 12/26/2011 05:06 AM, Gordan Bobic wrote: Disable fs journaling (normally dangerous, but this is throw-away space). Not really dangerous - the only danger is that you might have to wait for fsck to do it's thing on an unclean shutdown (which can take hours on a full TB scale disk, granted). I mean dangerous in the sense that if the server goes down, there might be data loss, but the builders using the space won't know that. This is particularly true if nfs exports are async. Strictly speaking, journal is about preventing the integrity of the FS so you don't have to fsck it after an unclean shutdown, not about preventing data loss as such. But I guess you could argue the two are related. Obviously, I meant protecting rather than preventing the integrity. :^) *facepalm* Gordan ___ arm mailing list arm@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/arm
Re: [fedora-arm] builder io isue
On 12/25/2011 09:06 PM, Gordan Bobic wrote: Why not just mount direct via NFS? It'd be a lot quicker, not to mention easier to tune. It'd work for building all but a handful of packages (e.g. zsh), but you could handle that by having a single builder that uses a normal fs that has a policy pointing the packages that fail self-tests on NFS at it. I'm not acquainted with the rationale for the decision so perhaps somebody else can comment. Beyond the packages that demand a local filesystem, perhaps there were issues with .nfsXXX files, or some stability problem not seen when working with a single open file? Not sure. 512KB chunks sound vastly oversized for this sort of a workload. But if you are running ext4 on top of loopback file on top of NFS, no wonder the performance sucks. Well, 512KB chunks is oversize for traditional NFS use, but perhaps undersized for this unusual use case. Sounds like a better way to ensure that would be to re-architect the storage solution more sensibly. If you really want to use block level storage, use iSCSI on top of raw partitions. Providing those partitions are suitably aligned (e.g. for 4KB physical sector disks, erase block sizes, underlying RAID, etc.), your FS on top of those iSCSI exports will also end up being properly aligned, and the stride, stripe-width and block group size will all still line up properly. I understand there was an issue with iSCSI stability about a year ago. One of our engineers tried it on his trimslice recently and had no problems so it may be time to reevaluate its use. But with 40 builders, each builder only hammering one disk, you'll still get 10 builders hammering each spindle and causing a purely random seek pattern. I'd be shocked if you see any measurable improvement from just splitting up the RAID. Let's say 10 (40/4) builders are using one disk at the same time- that's not necessarily a doomsday scenario since their network speed is only 100mbps. The one situation you want to avoid is having numerous mock setups at one time, that will amount to a hammering. How much time on average is spent composing the chroot vs building? Sure, at some point builders will simply overwhelm any given disk, but what is that point? My guess is that 10 is really pushing it. 5 would be better. Using the fs image over loopback over NFS sounds so eyewateringly wrong that I'm just going to give up on this thread if that part is immutable. I don't think the problem is significantly fixable if that approach remains. Why is that? I don't see why you think that seeking within a single disk is any less problematic than seeking across multiple disks. That will only happen when the file exceeds the chunk size, and that will typically happen only at the end when linking - there aren't many cases where a single code file is bigger than a sensible chunk size (and in a 4-disk RAID0 case, you're pretty much forced to use a 32KB chunk size if you intend for the block group beginnings to be distributed across spindles). It's the chroot composition that makes me think seeking across multiple disks is an issue. And local storage will be what? SD cards? There's only one model line of SD cards I have seen to date that actually produce random-write results that begin to approach a ~5000 rpm disk (up to 100 IOPS), and those are SLC and quite expensive. Having spent the last few months patching, fixing up and rebuilding RHEL6 packages for ARM, I have a pretty good understanding of what works for backing storage and what doesn't - and SD cards are not an approach to take if performance is an issue. Even expensive, highly branded Class 10 SD cards only manage ~ 20 IOPS (80KB/s) on random writes. 80KB/s? Really? That sounds like bad alignment. Strictly speaking, journal is about preventing the integrity of the FS so you don't have to fsck it after an unclean shutdown, not about preventing data loss as such. But I guess you could argue the two are related. Right, sorry, was still thinking of async. I'm still not sure what is the point of using a loopback-ed file for storage instead of raw NFS. NFS mounted with nolock,noatime,proto=udp works exceedingly well for me with NFSv3. I didn't think udp was a good idea any longer. Well, deadline is about favouring reads over writes. Writes you can buffer as long as you have RAM to spare (expecially with libeatmydata LD_PRELOAD-ed). Reads, however, block everything until they complete. So favouring reads over writes may well get you ahead in terms of keeping he builders busy. It really begs the question: What are builers blocking on right now? I'd assumed chroot composition which is rather write heavy. -- Brendan Conoboy / Red Hat, Inc. / b...@redhat.com ___ arm mailing list arm@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/arm