8-STABLE Slow Write Speeds on ESXI 4.0

Ivan Voras ivoras at freebsd.org
Mon Aug 9 16:38:36 UTC 2010


On 9 August 2010 18:11, Jeremy Chadwick <freebsd at jdc.parodius.com> wrote:

> I thought Intel VT-d was supposed to help address things like this?

Probably - http://www.intel.com/technology/itj/2006/v10i3/2-io/7-conclusion.htm
says it should help unmodified guests, but I don't know for sure. I do
know that Nehalems run faster on VMWare, probably because "nested
paging" or whatever it's called helps context switches on syscalls.

> I can confirm on VMware Workstation 7.1, not ESXi, that disk I/O
> performance isn't that great.  I only test with a Host OS of Windows XP
> SP3, and for the Guest OS's hard disk driver use the LSI SATA/SAS
> option.  I can't imagine IDE/ATA being faster, since (at least
> Workstation) emulates an Intel ICH2.

Yes, disk IO was always slow with VMWare. VirtualBox cheats by
emulating ATA controllers (ICH6) instead of SCSI and turning on disk
cache - it's noticably faster than VMWare.

> I was under the impression that ESXi provided native access to the
> hardware in the system (vs. Workstation which emulates everything)?

I think it can be configured this way, but then you'd need a separate
LUN for the VM drive, bypassing vmware's usual storage (vmfs) and all
the goodies that come with it. OTOH, there are paravirtualized drivers
for Linux and Windows in 4.0 which should help, but I haven't tried
them yet.

> The controller seen by FreeBSD in the OP's system is:
>
> mpt0: <LSILogic SAS/SATA Adapter> port 0x4000-0x40ff mem 0xd9c04000-0xd9c07fff,0xd9c10000-0xd9c1ffff irq 18 at device 0.0 on pci3
> mpt0: [ITHREAD]
> mpt0: MPI Version=1.5.0.0
>
> Which looks an awful lot like what I see on Workstation 7.1.
>
> FWIW, Workstation 7.1 is fairly adamant about stating "if you want
> faster disk I/O, pre-allocate the disk space rather than let disk use
> grow dynamically".  I've never tested this however.

Yes, this statement has always been true.

> How does Linux's I/O perform with the same setup?

I've tested Linux, Windows and FreeBSD on VMWare 3.5 last year and the
results (IOPS) were:

ESXi-FreeBSD	174
ESXi-Linux	221
ESXI-Windows	98
Xen-FreeBSD	72
Xen-Linux	148
Xen-Linux-PV	244
HyperV-FreeBSD	61
HyperV-Linux	69
HyperV-Windows	58

(I couldn't get Windows to run on Xen; "Linux-PV" is Linux as
paravirtualized Xen guest).


More information about the freebsd-stable mailing list