With storage area networks (SAN) performance issues sometimes arrise. Knowing the maximum throughput speed of your network is key to remove it as a possible bottlekneck. Bellow I will outline the required calculations for determining the maximum throughput speed of 2 Gigabit Fibre Channel.

Calculation Rules

Network Speeds = Always in Bits

Storage/Disk Values = Always in Bytes

Files = Always in Bytes

1 nybble = 4 bits

1 word = 16 bits

8 bits = 1 Byte

1024 Bytes = 1 Kilobyte

1024 KB = 1 Megabyte

1024 MB = 1 Gigabyte

Calculating Maximum Throughput for 2 Gigabit Fibre Channel in MegaBytes

2 Gigabits = 2/8bits = 0.25 GigaBytes

0.25 GigaBytes x 2 (2 HBAs) = 0.5 GigaBytes

0.5 x 1024 = **512 MegaBytes/sec** throughput or **256 MB/sec per HBA**.

Depending on the monitoring software used performance numbers may not be in MB format. Bellow I’ve broken down the values for each common throughput value.

256 MegaBytes/sec =

262,144 KiloBytes/sec

268,435,456 Bytes/sec

2,147,483,648.00 Bits/sec

~~33,554,432 Bits/sec~~

Since FC speeds have tripled since I wrote this post an update is in order. Bellow I’ve outlined speeds for 4Gb and 8Gb FC.

4Gb = 512 MB/s

8Gb = 1024 MB/s

Now that we have determined the maximum throughput of each HBA we need to rule out any bottlenecks found in the server itself or in the SAN fabric. HBA throughput speed is directly impacted by the bus speed of each HBA as well as whether the switch ports utilized by either the storage array FA’s or the HBA itself are dedicated bandwidth or over subscribed ports.

…. brain dump in progress

### Like this:

Like Loading...

*Related*

Looks like there is an error in the last calculation. I believe it should be:

256 MegaBytes/sec =

262,144 KiloBytes/sec

268,435,456 Bytes/sec

2,147,483,648 bits/sec

Thanks for pointing this out… looks like I divided instead of multiplying by 8. Doh!

Hi there,

Very nicely explained. Thanks a lot. I have one query though, I hope the thread is still open.

I read somewhere that – Based on testing and customer experiences, any bandwidth numbers follow the “70% rule,” which dictates that any speed claimed by the hardware vendor be reduced to 70% of the top-rated speed. For instance, 1 Gigabit Ethernet (GbE) speed is theoretically 125 MB/sec, it would be – (125 * 70% = 87.5 MB/sec) practically. My question is, does it also apply to Fibre-channels. If so, will these figures change :

For 4Gb and 8Gb FC with 70% rule.

4Gb = 512 MB/s = 358 MB/s

8Gb = 1024 MB/s = 716 MB/s

Thanks in advance!

-Ashwin

Excellent question. To be quite honest I’m not sure whether there is a similar overhead ratio with Fibre Channel(FC) networks. In most high volume FC networks I’ve worked in the actual fabric isn’t the bottleneck. In some cases the physical server hardware is a constraint. An example would be using a PCI vs. PCI-Express slot which would give you a limit of 133MB/s vs. 256MB/s at the motherboard layer. It’s also important to keep in mind the port constraints of the fabric. Depending on the switch vendor and the size of your fabric different over subscription ratios may exist. An example would be if a group of say 6 ports share 12Gb/s of bandwidth. If each of the 6 ports pushes as much I/O as possible at the same time each port will max out at a theoretical 2Gb/s. If each host thinks it has a 4Gb/s FC link then you can quickly see where a potential bottleneck lies.

You can see how the specific environment components can make a huge difference when looking at the “maximum through put” of an FC link. And this isn’t even approaching application layer inefficiencies. One last piece to add. In order to really figure out whether you have a real bottleneck within your FC network setup a tool which polls FC port throughput details with a high frequency. Say once every 5 seconds. Collect the data for a day, week, or even a month and then import it into Excel and chart the frequency of bandwidth utilization. Use of histograms and pivot tables make this task easier. The frequency of spikes dictates the utilization of bandwidth. Based on the calculations I’ve already presented you can also gauge percent utilization etc.

Thanks a ton for explaining so well. I have noted down your suggestions, and whenever I get an opportunity to test it, I will certainly do that. Very informational blog, keep sharing knowledge! You are doing a great job.

2 Gigabit HBA delivers something closer to 212.5MB/s.

Please see the following link I stumbled upon:

http://www.unifiedcomputingblog.com/?p=108

And the list of various protocols and their throughputs (do search for Fibre):

http://en.wikipedia.org/wiki/List_of_device_bandwidths

Interesting post, and excellent reference.

Excellent post & learn from it. keep up the good work.

I am sofa-king confused…..

I always thought FC was expressed in GB per second and Ethernet in Gb per second….

Networking is always in Bits/Sec, where as disk is in Bytes.

It doesn’t look like these calculation include the 8b/10b encoding, I think you have to take 80% off the top to account for this. For 10 bits of transmitted data, only 8 bits are data.