Hi ,
i have been browsing the discussion / kb and googled alot on this subject and didnt find any clear directions concerning this issue. The issue being, the speed will not exceed between 3 and 4MB/sec (thats about 24 to 32mbit) network performence when we upload/download from the datastore using the VI client or the http browser. Even the VMWare convert 3 and 4 wont exceed that network speed.
BUT
Usual network traffic from/to the VM running in it, can achieve speed of 60MB/sec (500mbit) between each other and top the full gigabit (100-125MB/sec) if we do network transfer with another host/physical machine. So it is clearly not a disk IO or NIC issue, since my VM performs very well, and yes the VM are using the same vswitch as the host management.
Here are the test i did:
The Server is a proliant ML350 G5, 10GB ram, and the data store is a 900GB Raid 10 array composed of 4 x 450GB SAS 15K RPM disks
i have 2 VM guest (both windows 2003 x86) running on a VMware ESXi 3.5 U4 with HP management agents integreted
Both guest use enhanced vmxnet NIC.
Test #1: Transfer of a 600mb .iso file from a physical server to VM guest #1 took 7 seconds 650mb/7 = 92MB/sec (0.73gbit/s)
Test #2: Transfer from that same 600mb iso file from guest #1 to guest #2 took 12 seconds 650mb/12 = 54MB/sec (0.43gbit/s)
Test #3: Installed VI client on guest #2, logged into esxi, opened datastore to upload a file and upload that same .iso file took 210 seconds 650mb/210 = 3MB/sec (0.024gbit/sec or 24mbit/s.....)
Is this by design or normal or? I know that some peoples enable ssh on the host and then use scp that has way better performences, that can work out for file moving but what happens with converter then who has same performence as when we use datastore file upload/download? Anyone else having this same issue?
Thanks
Seb
VI client will not win Formula 1 prize, no doubts.
Try to use Veaam FastSCP (free) to manage files.
---
VMware vExpert '2009
I'm very happy to know that it doesn't happens just with me. Last weekend I took 17 hours to convert a Physical Windows 2003 machine to ESXi.
The source Machine had one RAID 5 volume with 3 SCSI disks. Two partitions 12GB and 286GB.
Converter version: 4.0.0 build 146302
It was a hot conversion. 12Gb partitions wasn´t resized, and converter did a block-level cloning. It was very fast. (ok, everyone will say, block level is much faster then file-level, but in practice, it never did any diference to me)
286GB (159GB used) partition was resized to 255GB, and converter did a file-level cloning. Here the transfer was horrible. Max of 3MB/s.
No packages are dropped. No packages loss.
The source machine and the ESXi host was in the same switch.
Even with 3MB/s or less network traffic, the disks on the source machine was 100% busy, working very hard.
Windows performance monitor showed 20MB/s disk READ I/O, network traffic 3MB/s ou less.
Sysinternals FileMon running in source machine didn't show wich files converter-agent was copying/accessing. (strange!)
Convert to ESXi is an impossible mission!
The way I solved this is to run the VI Clent in a VM that I use for managment, Also I run Converter 4 in the same VM. I have great speeds now and it is very stable.
I tried that and obtained same performence. The only quick way i found to transfer large files was to enable ftp via the unsupported console. With alot of research i have found that tso setting might have something to do with this, to disable tso on management interface, we have to re-create vmkernel NIC with the flag -t from the unsupported console (esxcfg-vmknic -t). I didnt try this yet as i dont want to test this on a production server and i dont have any machine in a lab right now.
Update: Just tested ESXi 4.0, seems to have the same problem unfortunatly.
Can we enable ftp?