Excessive delays in HDs (Local and Remote-iSCSI)

R

rikw22

New Member
Feb 23, 2011
3
0
1
Hello,
I'm having very slow in some virtual machines.
My current setup is as follows:
- The hard drives of the machines are stored in container files that are in storage, the connection is made via iSCSI.
- Both "/" machines are assembled locally on 4 HD's in RAID 5
- Both "/ var" are mounted on the storage via iSCSI
- The connection between the machines is done via Cat6 cable - Gigabit

Machine A
- Container virtualization (the files are saved in storage (iSCSI))
- The transfer rate for "/" (via iSCSI) reaches 200 MB / s

Machine B
- Full virtualization
- Transfer to "/" (HD local machine) = 150MB / s
- This machine contains only the partition "/"

Machine C
- Full virtualization
- Transfer to "/" (HD local machine) = 60MB / s
- Transfer to "/ var" (via iSCSI) = 14MB / s

Machine D
- Full virtualization
- Transfer to "/" (HD local machine) = 8.7 MB / s
- Transfer to "/ var" (via iSCSI) = 6.3 MB / s


The physical machine is a Dell PowerEdge R410 (16 x Intel (R) Xeon (R) CPU E5620@2.40GHz | 16GB Ram)



The transfer tests were performed using the command "dd" of Linux.
dd if = / dev / zero of = teste.iso bs = 1024 count = 512000

NOTE:
- The bottleneck is not in the processor and memory for the tests performed using the CPU and memory did not reach 50%.
- The two machines are physically the same switch Gigabit
- Proxmox 1.6


Does anyone have any idea of what can be?
 
65 - Machine A
root@Container:~# dd if=/dev/zero of=bigfile bs=1024k count=1024 conv=fdatasync
1024+0 records in
1024+0 records out
1073741824 bytes (1.1 GB) copied, 187.447 s, 5.7 MB/s


63 - Machine B
root@machineb[/]# dd if=/dev/zero of=bigfile bs=1024k count=1024 conv=fdatasync
1024+0 records in
1024+0 records out
1073741824 bytes (1,1 GB) copied, 103,961 seconds, 10,3 MB/s

61 - Machine D
machined:/# dd if=/dev/zero of=bigfile bs=1024k count=1024 conv=fdatasync
1024+0 registos dentro
1024+0 registos fora
1073741824 bytes (1,1 GB) copiados, 64,6794 s, 16,6 MB/s

machined:/var/vmail# dd if=/dev/zero of=bigfile bs=1024k count=1024 conv=fdatasync
1024+0 registos dentro
1024+0 registos fora
1073741824 bytes (1,1 GB) copiados, 176,869 s, 6,1 MB/s

Note:
- This transfer is very slow, given that the connection between the servers is Gigabit
- In a test on my local machine returned 109MB / s on a common desktop machine.
 
rikw22 said:
...

Note:
- This transfer is very slow, given that the connection between the servers is Gigabit
- In a test on my local machine returned 109MB / s on a common desktop machine.
Click to expand...
Hi,
109MB/s troughput to an iscsi-Volume? I can't believe that. 109 MB/s is fairly the maximum which is reachable with an gigabit connection (perhaps measurement with local buffers...).
But i'm not the right one for iscsi-speed - my tests are all not fast enough... my next test will done soon with 10G-Nics.

Because Server/Desktop: the core-speed is normaly higher on desktop-cpus, if you need only a single thread (e.g. io), you are often (much) faster on a desktop box. Server hardware has better (more) caches / smp-performance and designed für 24/7 use.

Udo
 
Sorry,
Did not express myself properly.
109MB / s write to local disk

Since already many thanks.
 
You must log in or register to reply here.
Top Bottom