IO Delays

E

Erwin123

Member
May 14, 2008
207
1
16
What value's are acceptable here in general?
Also, what raid mode would be best in a 4 disk configuration?
Thanks!
 
Erwin123 said:
What value's are acceptable here in general?
Also, what raid mode would be best in a 4 disk configuration?
Thanks!
Click to expand...

"acceptable" is not really a sharp definition - normal is no delay the most times, only if you make installation of new VM´s, backups or copy huge files io delays should occur.

You get best performance with RAID10.

Performance benchmark:

Code: 
pveperf
here are my values (Raid10, 4 x 500 GB sata, adaptec 3805 with cache enabled and batteries backup module).
Code: 
proxmoxve:~# pveperf
CPU BOGOMIPS:      17027.42
REGEX/SECOND:      765295
HD SIZE:           94.49 GB (/dev/pve/root)
BUFFERED READS:    180.32 MB/sec
AVERAGE SEEK TIME: 9.97 ms
FSYNCS/SECOND:     1209.17
DNS EXT:           36.50 ms
DNS INT:           1.02 ms (maurer-it.com)
proxmoxve:~#
if you use 15krm enterprices SAS drives you will get much better performance, but here I just got cheap 7200 sata drives.
 
io delays ::

do u have any idea why i could have around of 2 % of delays,

2 promox servers
dell 8gb ram 2 x quad core 2 x raid 1 perc 6/e 146 15k sas drivers 2 port NETEXTREME II giga bit

one server is master the other slave, no of them is heavy loaded, simply VES, and now is at night, tomorrow when is going to be activity (work day) perhaps i would get more IO delays.... the delays happens on the master NODE

The switch that we use is a giga bit too...

So i wonder if u could help me to find the cause, the bean counters are ok.

THANKS!!
 
Whats the output of:

# pveperf

But i guess the Disk performance is ok with that hardware.

You should first find out which VM is causing the delays (Stop one VM, then see if the delay is still there, ...)

Then analyze the programs running inside the VM.

- Dietmar
 
raq102:~# pveperf
CPU BOGOMIPS: 31923.55
REGEX/SECOND: 593247
HD SIZE: 33.47 GB (/dev/pve/root)
BUFFERED READS: 85.14 MB/sec
AVERAGE SEEK TIME: 7.83 ms
FSYNCS/SECOND: 1517.28
DNS EXT: 25.78 ms
DNS INT: 2.78 ms (raqdedicados.com)

at this time there is no consume but the cp shows some IO delay, the disk we use are sas 15k 147GB with a perc 6/e controller raid 1

raq103:~# pveperf
CPU BOGOMIPS: 15963.24
REGEX/SECOND: 591645
HD SIZE: 57.09 GB (/dev/pve/root)
BUFFERED READS: 61.37 MB/sec
AVERAGE SEEK TIME: 11.16 ms
FSYNCS/SECOND: 2938.22
DNS EXT: 22.80 ms
DNS INT: 2.50 ms (raqdedicados.com)

This server has 2 sata2 disc running with same controller, but is a small server online 1 x quad core 2.0 with 4gb ram.. but it does not show delay more frecuent than the sas did
 
Last edited:
raqdedicados said:
This server has 2 sata2 disc running with same controller, but is a small server online 1 x quad core 2.0 with 4gb ram.. but it does not show delay more frecuent than the sas did
Click to expand...

I guess I dont really understand what you want to say - what the conclusion? Both servers looks fast, although buffered reads could be faster.

- Dietmar
 
dietmar

don't know why but iodelays goes to 8 % sometimes, the vm's inside has not high resources, i'm going to re install from the scratch. thanks for all!!
 
tom said:
"acceptable" is not really a sharp definition - normal is no delay the most times, only if you make installation of new VM´s, backups or copy huge files io delays should occur.

You get best performance with RAID10.
Click to expand...

And how is "IO delays" value calculated?

Is it the same as "wa" (iowait) shown by top?
 
Last edited:
BUMP


Could we please know how "IO Delays" is calculated.

Under heavy load (and low performance RAID array) I am seeing IO Delays of around %30

I'm not worried... just wondering what this means (how is it calculated).

(Many thanks)
 
davewantsmoore said:
BUMP


Could we please know how "IO Delays" is calculated.

Under heavy load (and low performance RAID array) I am seeing IO Delays of around %30

I'm not worried... just wondering what this means (how is it calculated).

(Many thanks)
Click to expand...

it it helps: "IO Delays" is equivalent to the "wa" displayed via top (on the console). In basic words: if this goes up, the CPU is waiting for the rest of your hardware. means you should think of improving your IO performance - find the bottleneck - most times the raid controller.
 
I have that I/O Delay goes up until 50%-60% and then some VMs become "blocked" . I see that operation into a VMs is a cause of this problem.
This machine is a Win2k3 Sp2 with IDE cqow2 and realtek net...and I disabled some services into it..
Have you some tips?
How can I read hw info of controller?
Virtio can hel me?
Thanks
 
BiagioParuolo said:
I have that I/O Delay goes up until 50%-60% and then some VMs become "blocked" . I see that operation into a VMs is a cause of this problem.
This machine is a Win2k3 Sp2 with IDE cqow2 and realtek net...and I disabled some services into it..
Have you some tips?
How can I read hw info of controller?
Virtio can hel me?
Thanks
Click to expand...

I run most of our production machines with IDE (raw) and e1000. some with virtio (net and block).
 
tom said:
I run most of our production machines with IDE (raw) and e1000. some with virtio (net and block).
Click to expand...

and just to mention, if you do not use image files (qcow2 and raw files on local disks) you will get better performance => use LVM (if available on a iSCSI or FC SAN or on a additional fast local raid volume)
 
Hi,We experienced meaningful several hour delivery delays yesterday afternoon and are currently experiencing delays as well. Could we get an update as to how long mail is currently being current delayed and when you expect things to be running smoothly again?Thanks,Frank
 
cearkeli73 said:
Hi,We experienced meaningful several hour delivery delays yesterday afternoon and are currently experiencing delays as well. Could we get an update as to how long mail is currently being current delayed and when you expect things to be running smoothly again?Thanks,Frank
Click to expand...

What exactly are you talking about?
 
I'm experiencing very poor performance on a new dedicated enterprise grade server (8 cores/8GB/500GBx2 RAID1). No VM's are running, and the server is pretty much idle per top. Any suggestions for troubleshooting? Thanks!

s1:/home/user# pveperf
CPU BOGOMIPS: 27207.43
REGEX/SECOND: 514344
HD SIZE: 94.49 GB (/dev/pve/root)
BUFFERED READS: 60.41 MB/sec
AVERAGE SEEK TIME: 9.44 ms
FSYNCS/SECOND: 145.63
DNS EXT: 68.36 ms
DNS INT: 45.57 ms (domain)
 
laleger said:
I'm experiencing very poor performance on a new dedicated enterprise grade server (8 cores/8GB/500GBx2 RAID1). No VM's are running, and the server is pretty much idle per top. Any suggestions for troubleshooting? Thanks!

s1:/home/user# pveperf
CPU BOGOMIPS: 27207.43
REGEX/SECOND: 514344
HD SIZE: 94.49 GB (/dev/pve/root)
BUFFERED READS: 60.41 MB/sec
AVERAGE SEEK TIME: 9.44 ms
FSYNCS/SECOND: 145.63
DNS EXT: 68.36 ms
DNS INT: 45.57 ms (domain)
Click to expand...

Do you have HW raid? cache enabled?
 
cyberbootje said:
Do you have HW raid? cache enabled?
Click to expand...

"lspci" reveals 3ware Inc 7xxx/8xxx-series PATA/SATA-RAID (rev 01)

As far as cache...

"blockdev --getra /dev/sda" initially revealed value of 256 but value been increased to 4096. No noticeable change.

I attempted to just move 800MB of files from one dir to another, the process took 5 minutes and used considerable CPU resources...

server:/var/lib/vz/template/iso# ls -lh; uptime ; mv *.iso /home/test ; uptime
total 827M
-rw-r--r-- 1 root root 668M Jun 20 10:00 file1.iso
-rw-r--r-- 1 root root 159M Jun 20 10:00 file2.iso
21:19:55 up 23:05, 2 users, load average: 0.27, 0.91, 1.02
21:24:35 up 23:10, 2 users, load average: 3.58, 2.61, 1.72
server:/var/lib/vz/template/iso#

Welcome any suggestions.
 
You must log in or register to reply here.
Top Bottom