10 Gbit Ethernet, the super I/O pipe for virtual servers? (WMWorld 2008)
by Johan De Gelas on March 1, 2008 12:00 AM EST- Posted in
- Virtualization
10 Gbit Ethernet just got even more attractive. You might remember from our last storage article that we have high hopes for iSCSI as a high performance but a very cost effective shared storage solution for SME's. Our hopes were based on getting 10 Gbit (10 GBase-T) on UTP Cat 6 (or even CAT5e) but unfortunately the only switch that I could find (thanks Renée!) that supports 10 Gbit this way was the SMC TigerSwitch 10G. With pricing at $1000 per port, not really a budget friendly offering.
Still, 10 Gbit Ethernet is an incredibly interesting solution for a virtualized server or an iSCSI storage array that is serving data to a lot of (virtualized or not) servers.
So maybe it is best to give optic cabling another look. Some of the 10 Gbit Ethernet NICs are getting quite cheap these days, but an enthousiastic Ravi Chalaka, Vice President of Neterion told us that it might be wise to invest a bit more in NICs with IOV (I/O virtualization) support. According to Neterion, the newest Neterion X3100 Series is the first adapter to support the new industry-standard, SR-IOV 1.0 (Single-Root /O Virtualization.) SR-IOV is a PCI-SIG workgroup extension to PCIe. One of the features of such a NIC is that is has multiple channels that can accept multiple requests of virtualized servers, which significantly reduces the latency and overhead of multiple servers sharing the same network I/O. Even more important is that the Neterion X3100 is natively supported in VMWare ESX 3.5.
We will test the Neterion X3100 in the coming months. It seems like a very promising product as Neterion claims :
- 7 times more bandwidth
- 50% less latency
- 40 % less TCP overhead
Than a comparable 1 Gbit solution. So while many of us are probably quite pleased with the bandwidth of 2 GBit (2x 1 Gbit MPIO), especially 50% lower latency sounds great for iSCSI. Fibre Channel, which is moving towards 8 GBit, might just have lost another advantage...
Facebook
Twitter
RSS
14 Comments
View All Comments
JohanAnandtech - Monday, March 3, 2008 - link
Latency of storage devices almost always determines the reponse time and subjective user "feeling". iSCSI latency is not much higher than FC in our experience, but if you can shave up quite a few ms, it is a big win towards your endusersI have still to study IOV, but it basically boils down to the fact that there is no switching involved, that most of the TCP processing is in parallel. A kind of "superscalar" NIC, not a "multicore NIC"
Olaf van der Spek - Monday, March 3, 2008 - link
I thought latency due to networking was already sub-ms?
I don't see how you could shave 'quite a few' ms from that.
JohanAnandtech - Monday, March 3, 2008 - link
You are right that a basic ping on a hardly used network line is a lot less than 1 ms.However, sending back a basic block of 64 KB for example, requires quite a few of ethernet packages (43 if you do not use Jumbo frames) before you got your first block back.
http://it.anandtech.com/IT/showdoc.aspx?i=3147&...">http://it.anandtech.com/IT/showdoc.aspx?i=3147&...
Look at the Random latency, up to 25 ms. And that is not only harddisk latency, See the DAS numbers (6 ms).
And that is measured from one server direct
Phyllis Hershberger - Tuesday, May 15, 2018 - link
Thank you for your post! if i can find some support on 10Gtek?