Urgent Problem i/o ;(

Hi,
First of all congratulate you on the forum :) and sorry for the English but thanks to google.
Well my problem is this .. I'm testing with open-and y xenserver but I did not get the performance you expect, because the I / O disk is very low and not what I expected.

He indicated the I / O disk of a VM with CentOS 5.8

[root@stor ~]# dd if=/dev/zero of=test bs=64k count=16k conv=fdatasync
16384+0 records in
16384+0 records out
1073741824 bytes (1.1 GB) copied, 118.238 seconds, 9.1 MB/s


I have it configured the storage as follows:

Xenserver 2 nic:

1 nic: ip public
2 nic: ip private example 192.168.xxx


Storage 2 nic:

Bond: ip private example 192.168.xx


xenserver is connected via the private IP to the storage via a dedicated 1Gbps switch

the features of my raid are:

Hardware RAID info
Attribute name Value
State: Optimal
Strip size: 64kB
Raid Level: RAID5
Number of disk in one array: 7
Size: 5720064MB
Default Write Policy: Write Back
Read Policy: Readahead
Access Policy: Read-Write
Cache Policy: Cached
Disk cache: Disk's Default


Grateful if you could help me.

Is this iSCSI target? or NAS Volume?
Have you tried without the bond?

ISCSI target.
Also turn off the bond.

Open a support case, so we can see your system logs

What model disk controller? Do you have Jumbo Frames enabled? Did you tweak the ISCSI target settings as described in various threads in this forum? Getting between 52 MB/s - 89 MB/s depending on traffic using your command line in a centos 5 vm using LSI 9260 disk controller and RAID 10 Sata Drives and Jumbo Frames,target settings tweaked and using MPIO (not bonded). Have about 32 vms active under Xenserver 6.0

Thanks for the answers, I contacted support and make the following changes:

maxRecvDataSegmentLen=262144
MaxBurstLength=16776192
Maxxmitdatasegment=262144
FirstBurstLength=65536
DataDigest=None
maxoutstandingr2t=8
InitialR2T=No
ImmediateData=Yes
headerDigest=None
Wthreads=8

and jumbo frames 9000 and my disk SEAGATE ST31000424SS 00069WK36GJR

Do you have any additional configuration xenserver?
What version do you have? payment or free?

What is the build of your Xenserver? is it the latest? as your xen might have an issue with your NIC drivers, did you checked for that!

Quote:

---

Originally Posted by javiercampos

Do you have any additional configuration xenserver?
What version do you have? payment or free?

---

payment or free is the same as far as base functinality of xenserver goes. Prefer MPIO to bonding because you can get close to 2 gig throughput, but in bonding your still only getting 1 gig max throughput to any one server.

switch are using some special?

using Dell 6248 with Gigabit ports. You need to make sure if using Jumbo Frames that its supported by your switch. Make sure your packets are not getting fragmented. Use something like this test from your xenservers to your DSS server ISCSI IP's: ping -M do -s 8972 -c 10 10.10.10.1 (where 10.10.10.1 is your dss ip addr) If the packets are fragmenting it will tell you, if they are ok then the repsone will be 10 normal looking ping responses.

Your Xenserver nics have to have MTU=9000 set and the same on the DSS side. On the Dell 6248 we have to set each port that we want to do Jumbo Frames to an MTU of 9016 which kicks it into Jumbo Frame mode. Check the docs for your switch to see if there is anything you have to do to enable JF's.

Hello,
I configured the server and storage with a crossover cable card directly to the storage server .. and gives the same result.

Is this a production server? or are you in testing phase? if testing you might try using raid 10 instead of raid 5. What disk controller did you say you were using and how much cache does it have. I am assuming your using Block I/O.

Running out of ideas as to what your issue might be :-)

Quote:

---

Originally Posted by webguyz

Is this a production server? or are you in testing phase? if testing you might try using raid 10 instead of raid 5. What disk controller did you say you were using and how much cache does it have. I am assuming your using Block I/O.

Running out of ideas as to what your issue might be :-)

---

Thanks for the reply.
Look at the raid card is:

LSI MegaSAS 9260

the issue that only allows me to do raid0, 1, 5 and 6 RAID10.
The other .. My storage has a capacity of 16 drives in raid1 2 are for the open-e system and I have raid5 array with 7 drives .. how rare such an arrangement of 7 discs I get the same performance as one with 3 disks.

and if I'm using Block I/O.

Quote:

---

Originally Posted by javiercampos

Hello,
I configured the server and storage with a crossover cable card directly to the storage server .. and gives the same result.

---

My guess is that you have a 100mb crossover cable and/or the nic in your host is only negotiating at 100mb. make sure you have a gigabit switch and not just a fast Ethernet switch. If you do not have a gigabit crossover check this link on how to make one. http://logout.sh/computers/net/gigabit/. When you said you are only getting the 9.1MB/sec rates that is the full throughput of 100mb Ethernet. expect 80-90MB/sec for gigabit.

I have ten 1tb RE4 WD's in a raid 10 (~4.5 tb usable). Using a sata DOM for open-e OS. Seems a waste to use 2 drive slots just for the OS. I am using the 9260 LSI controller as well. I do have CacheCade 2.0, but I don't think that makes any difference because I was getting decent times with that that same test string even before I added the Cachecade. Cachecade really helps in speeding up all the websites so they are snappy and respond quickly.

Please check the following PDF at : http://www.open-e.com/download/libra..._file/get/131/
that will show you steps how to install DSSv6 on USB/DOM or storage LUN, as you can install DSSv6 on your storage disks, by creating a LUN with size of 2GB, then install DSSv6 on that LUN, in this way you can get red of the DOM/USB flash memory and save more slots.

Also we recommend you to use the latest builds of DSSv6 as it contain the latest drivers for your hardware.