1

We have a computer running Windows Storage Server 2008 R2 with 6 SSD in RAID 0.

This storage computer has one PCI-E with 4 Ethernet ports and we connected it to a gigabit switch to other computers vi iSCSI.

The problem is that we are not able to get high read/write speeds.

Using HD tune directly in the storage computer we get around 500 MB/s, but using the iSCSI link (in another computer) we only get close to 200 MB/s.

We did set MPIO with multipath, JUMBO frames and disabled CheckSum IPV4.

EDIT

I don't care about data loss. I just need speed because this is a cache computer.

EDIT

Both server and client have 4 GB NICs (1GB each adapter) and multipath and MPIO is correctly configured AFAIK.

EDIT

One thing I cant understand: we have a Dell Equallogic storage and it gets close to 200MB/s using the same switch/configuration. How is it possible? The equallogic was supposed to be a lot more slower than a 6 SSD Disk raid 0 storage.

Also, I have read that a lot of storages out there use 4 1GB NICs and they can easy get close to 500 GB/s. Included one from DELL which has only SSDs, as you guys can see here

EDIT

Also I am thinking about not using Windows Storage Edition and give OpenFiler a try. Should I consider this?

Rafael Colucci
  • 183
  • 1
  • 10
  • 1
    "6 SSD in RAID 0"?!?! **Shudder**. You're just asking for data loss there. – EEAA Jul 19 '11 at 14:46
  • I dont care about data loss. The server is just a cache computer. I want speed and thats it. – Rafael Colucci Jul 19 '11 at 14:50
  • Fair enough. How long can you go without this system, though? It's going to be down while you deal with the failed drive and re-build the array. – EEAA Jul 19 '11 at 14:51
  • I can go without this system forever. I have a cluster (there are another machines ready to go in case this system fails). I am sorry, but that is not the point, but thanks for trying to help. – Rafael Colucci Jul 19 '11 at 14:52
  • Data loss with SSD ? They always fail on write, but have unlimited read, I think that the likelyness of dataloss is pretty low – Lucas Kauffman Jul 19 '11 at 15:53
  • 2
    Anyway guys. I dont care about data loss as i already said. – Rafael Colucci Jul 19 '11 at 16:23
  • You don't mention what your actual switch , or switch stack is. This might be helpful as not all switches are created equal. – Tim Meers Jul 19 '11 at 16:54
  • It is a Dell switch. Pretty good switch. – Rafael Colucci Jul 19 '11 at 17:06
  • @Lucas, you apparently haven't witnessed (in my case dozens of) failed SSDs; when they go bad, they completely die, gone, no read, no write, no drive recognized at all. But as Rafael has already pointed out, fault tollerance has been baked into the design elsewhere. – Chris S Jul 19 '11 at 17:50
  • The other computer in question has more than 2 GbE links? 200MB sounds like iSCSI is using 2 of the links. Double check the sessions in the iSCSI Initiator, MPIO Policy, etc? – Chris S Jul 19 '11 at 17:54
  • See my edit please. – Rafael Colucci Jul 19 '11 at 17:56
  • @Rafael, try running the test with [SQLio](http://www.microsoft.com/download/en/details.aspx?displaylang=en&id=20163) instead of HD Tune. I'm concerned HD Tune may be introducing extraneous factors into the reported numbers. – Chris S Jul 19 '11 at 18:08
  • @Chris I dont think it is a HD Tune issue. Its pretty noticeable that the storage is slower than it should be. But I will give SQLio a try. – Rafael Colucci Jul 19 '11 at 18:29

1 Answers1

1

OK, problem solved.

It turned out to be NIC issues. We changed the NICS and updated them to latest drivers and now I am getting close to 500 MB/s. We tested the speed using Sql Server, and now is great.

Thanks you all.

Rafael Colucci
  • 183
  • 1
  • 10