I think you are right Garret, but why not squeeze all possible performance with some very simple changes?...
Btw here is what I've used successfully. SSG-6047R-E1R24L / SUPER Barebone 4U SuperStorage Server CM8063501288301 / 2.10GHz Xeon E5-2620 v2 Six-Core x2 MEM-DR316L-SL04-ER16 / 16GB DDR3 PC3-12800 (1600MHz) 240pin DIM EA 4 10022680 / x 12 MG03SCA400 / TOSHIBA x 8 wiredzone.com On Jul 1, 2016 11:41 AM, "Garrett D'Amore" <[email protected]> wrote: > for a database workload id recommend investing in either a good SSD for an > SLOG or just going pure SSD for the pool the database is on. once that is > done i suspect any performance differences between virtual and bare metal > will be small enough to ignore. > > Sent from my iPhone > > > On Jul 1, 2016, at 8:15 AM, Humberto Ramirez <[email protected]> wrote: > > > > Specifically on the performance of Windows VM's a few things you can try: > > > > - Check the property "model" of your disks and make sure is "virtio" > > If its IDE add an extra disk with virtio enabled, install virtio, > > shutdown, make the change on the original disk and start again > > > > vmadm get {UUID} | json disks > > > > > > - Set "compression": "lz4" "block_size": 131072 (This one can > > only be set at creation) > > > > - Set zfs sync=disabled on the VM's HD ie: zfs set > > sync=disabled zones/{UUID}-disk0 > > This is dangerous but a good way to determine if ZIL is the culprit > > > > - Play with different versions of the KVM virtio drivers I got up to > > 8%~12% better performance than the ones from Joyent, > > > > > https://fedorapeople.org/groups/virt/virtio-win/direct-downloads/archive-virtio/ > > > > > > For security, backup and take a snapshot of the disk(s) before trying > > any of this. > > > > > > > > > > I would recommend to test performance with CrystalDisk mark from > > inside the VM with every change (You could get some artificial results > > due to caching etc.) > > > > ----------------------------------------------------------------------- > > CrystalDiskMark 5.0.2 x64 (C) 2007-2015 hiyohiyo > > Crystal Dew World : http://crystalmark.info/ > > ----------------------------------------------------------------------- > > * MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s] > > * KB = 1000 bytes, KiB = 1024 bytes > > > > Sequential Read (Q= 32,T= 1) : 1197.028 MB/s > > Sequential Write (Q= 32,T= 1) : 1385.574 MB/s > > Random Read 4KiB (Q= 32,T= 1) : 74.490 MB/s [ 18186.0 IOPS] > > Random Write 4KiB (Q= 32,T= 1) : 107.978 MB/s [ 26361.8 IOPS] > > Sequential Read (T= 1) : 687.428 MB/s > > Sequential Write (T= 1) : 482.795 MB/s > > Random Read 4KiB (Q= 1,T= 1) : 15.460 MB/s [ 3774.4 IOPS] > > Random Write 4KiB (Q= 1,T= 1) : 13.381 MB/s [ 3266.8 IOPS] > > > > Test : 100 MiB [C: 22.9% (57.3/249.9 GiB)] (x1) [Interval=5 sec] > > Date : 2015/10/21 14:55:31 > > OS : Windows Server 2008 > > > > > > zpool status > > > > NAME STATE READ WRITE CKSUM > > zones ONLINE 0 0 0 > > mirror-0 ONLINE 0 0 0 > > c0t5000039588CB2F5Ad0 ONLINE 0 0 0 > > c0t5000039598D2E102d0 ONLINE 0 0 0 > > mirror-1 ONLINE 0 0 0 > > c0t5000039598D2E15Ed0 ONLINE 0 0 0 > > c0t5000039598E304BAd0 ONLINE 0 0 0 > > mirror-2 ONLINE 0 0 0 > > c0t5000039598E304C2d0 ONLINE 0 0 0 > > c0t5000039598E318DEd0 ONLINE 0 0 0 > > mirror-3 ONLINE 0 0 0 > > c0t5000039598E318F2d0 ONLINE 0 0 0 > > c0t5000039598E319C6d0 ONLINE 0 0 0 > > > > > > > > > > > > > > > > > > > > > > > > > >> On Thu, Jun 30, 2016 at 11:37 AM, <[email protected]> wrote: > >> hi all. i know the basic question is a tired one, so feel free to > ignore. i'm hoping someone can provide a recommendation based on real-world > experience with smartos and this particular software stack: windows server > 2008/2012 with sybase ads, for a small office client/server application. > >> > >> i have a small customer (15 desktops) whose business is totally managed > by a single client/server application. the latest version of the > application has moved to using a sybase ads backend. the end-user software > runs on a terminal server, as before, so i already have that covered. what > i do not have covered is the hardware for the new sybase ads portion. the > vendor requires windows 2008/2012 (standard or essentials) for their > propriety installation/configuration of the sybase database and their > related update/backup/management tools. > >> > >> i have smartos successfully running in several places with small > services (samba3 domain controller, webserver, zimbra on linux vm, and a > few windows vms running terminal services for applications). these work > great, with disk performance being the only remotely questionable aspect. i > do not have experience with large i/o workloads. there are a *lot* of > conversations on the list with varying recommendations/suggestions for > different situations. the most straight forward of the suggestions are "use > the hardware joyent uses in their datacenters" and "test it yourself and go > from there" ... > >> > >> this being a small customer (and myself just being a consultant) makes > affording enterprise-grade hardware a challenge. so i'm hoping someone can > recommend a known-working configuration that would allow a windows server > kvm running sybase database to perform decently on smartos. the less > expensive the better, of course. > >> > >> and if your recommendation is to skip smartos and go with a bare metal > install, i'd like to hear that as well. it's not what i want to end up > with, but that doesn't mean it's not the right answer. not everything works > well virtualized, and with the kvm serial-disk-access issue i've read about > i realize that might be the case here. > >> > >> thanks for your insight. > >> > >> -dewey > > > > > > ------------------------------------------- smartos-discuss Archives: https://www.listbox.com/member/archive/184463/=now RSS Feed: https://www.listbox.com/member/archive/rss/184463/25769125-55cfbc00 Modify Your Subscription: https://www.listbox.com/member/?member_id=25769125&id_secret=25769125-7688e9fb Powered by Listbox: http://www.listbox.com
