iSCSI performance on TS-469pro

[vmnomad]

Hi everyone.

I am quite new with QNAP devices and seeking for advice on performance.

We have TS-469 Pro model with 4 x 7.2K 3TB disks configured in RAID-10.

I have configured it as iSCSI target and presented to our ESXi 5.1 servers with proper multipath configuration. When i move my VM to the QNAP I can see that the average throughput is about 40MB/s.
I have checked network stats on ESXi server and I can see the iSCSI traffic is evenly balanced between 2 x 1Gb interfaces. I have also tried different MTUs - 1500 and 7418, same result. So, it is definitely not a networking issue.

is this kind of performance I should expect from QNAP or I can check something else on the QNAP?

Thanks

[sirozha]

I'm running into a somewhat similar issue. I have an ESXi host with 8 VMs on it: Windows 7, Ubuntu 12.04 LTS, and six Cisco Unified Communications servers (2 on Red Hat Enterprise 5 32-bit and 4 on Red Hat Enterprise 6 64-bit). As I'm installing more VMs, the existing VMs are becoming not usable (while a new VM is being installed). The datastore that all these VMs are installed on is on a QNAP TS-469L, which his similar to TS-469 Pro. The highest Write rate I see in ESXi to the QNAP is 13520 KBps, which is a little over 110 Mbps. The ESXi host is connected directly with a CAT6 cable to NIC2 on the QNAP.

NIC1 on the QNAP is connected to an Ethernet switch (Cisco 3560G). AFP is bound to NIC1 and SMB, NFS, and iSCSI are bound to NIC2. I'm not using SMB in my network, but I have a Windows 7 VM that uses SMB to mount network shares on the QNAP via QNAP's NIC2. I access QNAP network shares via AFP from all of my Macs (via QNAP's NIC1).

When a new VM is being installed, neither HTTP nor SSH sessions work well with existing VMs. The sessions barely crawl. At the same time, trying to browse QNAP shares via AFP is a painful experience. It takes minutes to open a folder. I can't attribute this to anything other than low IOPS on the QNAP. I have three WD RED 2TB drives in RAID 5 on the QNAP.

I would love to know what the maximum IOPS are that QNAP 569 (L or Pro) supports. I can find the IOPS required for the Cisco UC servers, and I believe that the QNAP TS-569 (L or Pro) are just not up to snuff with even 8 VMs being hosted via iSCSI.

I'm not running this in production - this is a home VoIP lab, which I use for work but only as a lab environment.

If anyone is interested, the ESXi host is a Late 2012 Mac Mini (6,2) with 2.6 GHz 4-core i7 CPU and 16 GB or RAM. I figure I can probably fit about 14 VMs on the Mac Mini, which is what I need for my lab. However, I'm afraid that I've almost reached the limit that TS-469L can support based on the IOPS.

[sirozha]

There's now an option to increase IOPS that the QNAP datastore can yield by a factor of 5 or 6. SSD cache is the answer to the problem; however, this feature is not supported in TS-X69 but is supported in TS-X70.

[sirozha]

Where are the mods when they are needed?

[storageman]

Even you want better random IOPs out of these base models without the extra cost of SSDs, try the WD 1TB 10k Velociraptors.
Personally I think you asking alot running VMs on the x69 range.

[sirozha]

Even you want better random IOPs out of these base models without the extra cost of SSDs, try the WD 1TB 10k Velociraptors.
Personally I think you asking alot running VMs on the x69 range.

There are IOPS calculators out there that show what a RAID5 array can yield in terms of IOPS. 10k drives would improve the performance somewhat, but not nearly enough to justify the price. SSD RAID would be the best way to approach this, but to save money, I think SSD cache would work as well. There's a blog that shows an improvement by the factor of 5 to 6 in terms of IOPS on the TS-x70 platform when SSD cache is enabled.

I will have to upgrade to TS-670 at this point to enable SSD cache.