New setup for virtualization / cloud hosting

[lpt]

Hello,

I'm implementing a new HA san setup to host a lot of vms / cloud platform. Our actual hardware setup is:

SAN:

2x Xeon 5620
16GB DDR3 Reg.ECC
LSI9260-8i with BBU (Write Cache On) with FW from SW Set 4.6 (latest from LSI)
24x 1TB Seagate Constellation SAS2
2x Intel 10GBe NIC

In Hypervisors i have:

2x Intel Xeon Six Core x5660
48GB RAM
1x Intel 10GbE to connect to SAN's

Connection between SAN's and HV is made by a new 10GbE Dell 8024f. We have two 10GbE ports in each SAN and one in each hypervisor.

We will have two SAN's for failover. From what i understand, Starwind works in an active-active mode, and shares the same information on both SAN. Is this correct? In case of a failure of a SAN, other will hold all the VMs. When the problem is fixed on the impacted SAN and it come back again, is necessary any manual procedure? The sync is manual? How about the SAN order? Suppose SAN 1 were down, it will become SAN 1 again, or will become SAN 2?

Regarding RAID setup, i created a 22 drives big RAID 50 array, and 2 hotspare drives in each SAN. Was this a wise choice? Or RAID 10 will be better?

About Starwind installation what version of Windows you suggest? 2008 R2 Web Server is enough? Can you provide us a trial?

Our backup solution is R1Soft CDP. Even with HA, two sans, and R1soft installed on all VM's, i would also like to have a backup of each SAN. Is it possible to install CDP in Windows and backup the whole SAN? Is this a crazy cenario for a 16TB SAN? Any specific recomendations?

We tested with a competitor but results werent what we expect, so we hope Starwind a way to go.

Sorry all the questions, but i never used / saw starwind in action.

thanks a lot

[anton (staff)]

1) Setup is nice.

2) Your assumptions about the way StarWind works are 100% correct ones.

3) I would personally go with RAID10. Many-many reasons. Try searching this form as RAID5/6 Vs. RAID1/10 was topic was discussed here millions of times

4) Any server Windows version for production. Anything you want for evaluation purpose. Yes, we can provide trial (of course). Just apply for it.

5) You can in theory but StarWind has no custom VSS writers so I don't think snapshots are going to be consistent. I'd recommend using StarWind own replication / CDP with R1 stuff.

6) Could you please drop me a message in private what you did not like about competitors.

Thanks!

[lpt]

Hello,

Thank you for your answers.

Regarding the R1Soft backups, i saw you announce that. How does it works? i'm r1soft partner. Usually we install the agent on our servers and backup it to our cdp backup servers. How does it really work in a starwind san?

thanks

[anton (staff)]

No idea. I know the way CDP for StarWind works. And recommended to use it with or instead of VM CDP software agent.

Hello,

Thank you for your answers.

Regarding the R1Soft backups, i saw you announce that. How does it works? i'm r1soft partner. Usually we install the agent on our servers and backup it to our cdp backup servers. How does it really work in a starwind san?

thanks

[lpt]

Hello,

Starwind has been isntalled, and HA configured. However, i'm seeing incredible slow speeds/performance:

[root@newsw ~]# 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, 131.991 seconds, 8.1 MB/s

What can cause this, and where should i look? Should jumbo frames be activated in nics? Should write cache be enabled on LSI controller?

[anton (staff)]

1) Start with disk subsystem. YES, write cache should be enabled (what's use w/o it for HIGH AVAILABILITY config?). Local tests should show NO gaps and read more or less the same as write.

2) Network between nodes should give wire speed with TCP.

3) HA cross-link should be fast enough to handle doubled I/O traffic between nodes (10 GbE for 1 GbE as initiators or 2x teamed 1 GbE at least).

4) Configure HA nodes to use as much Write-Back cache as possible.

5) Dump DD as your test tool. Are you going to perform linear writes or reads in your real life? You're editing video or what?

In your test with such a small image (1GB) you should have ~120 MB/sec for overlapped write.

Hello,

Starwind has been isntalled, and HA configured. However, i'm seeing incredible slow speeds/performance:

[root@newsw ~]# 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, 131.991 seconds, 8.1 MB/s

What can cause this, and where should i look? Should jumbo frames be activated in nics? Should write cache be enabled on LSI controller?