Latency is mostly due to two things: speed of light limitations and processing overhead of the hardware in the handling of packets. ASICs rule here.
There is, however, another problem when considering LAN/SAN switching: congestion due to oversubscription in the hardware. Mostly, this is because the backplane for the unit can't keep up with the amount of data being fed to it by the ports. If you've got a switch with a 6gb/s backplane, and it has 48x100mb/s full duplex ports, you have a theoretical maximum of 48*100*2mb/s (9.6gb/s) being handled by the backplane, which is more than 50% above what the backplane can handle. At peak load, you'll see problems. I'm simplifying things a bit, as Ethernet in a switched environment won't really achieve the theoretical maximum, but Ethernet in a switched environment isn't the pure CSMA/CD thing it was way back when, either. Kurt On Wed, May 28, 2014 at 12:08 PM, Matthew W. Ross <[email protected]> wrote: > How do you reduce latency? Is it at the network level, or the server level? > Or both? > > I'm no high-end networking guru. But I am starting to dabble into the dark > arts that are crazy-expensive networking switches for SAN communications and > TOR switches. Until recently, I have found that the good-olde HP Procurve > 2510 switch line (and its' replacement, the 2530) has been good enough for > most of our needs... and now I'm seeing that change. > > For example, we were experiencing minor network issues with our mail server > (it was complaining that you had lost it's connection on occasion when you > used the web interface). For a while, I thought the problem was with our > VMware servers, but on a hunch, I decried to bypass the 2510G switch they > were aggregated through directly to our 5308xl. Guess what, the problem > all-but disappeared. Obviously, the 2510 can't handle the traffic. > > On a similar note, We used to be running HP 2910s as our iSCSI traffic > connection switches between our 3 VMware hosts and our EqualLogic, but we > were again having latency issues. As recommended by our vendor, we upgraded > to Dell PowerConnect 6224s. On paper, these two switches are quite similar, > but the Dell switches handle the latency where the HPs were not. > > So, What does one look for? Is there a resource available that test, or even > benchmarks these switches? Or do I have to go by the voodoo method of asking > my vendor "which switch is best" for my situation? (In our vendor's case, > they were pushing a very expensive Brocade switch that we were not prepared > to pay for.) > > > --Matt Ross > Ephrata School District > > > Ben Scott <[email protected]> , 5/28/2014 10:57 AM: > > On Wed, May 28, 2014 at 1:36 PM, Matthew W. Ross > <[email protected]> wrote: >> Last, It's amazing how little bandwidth is actually used by the end user. >> Except for saving some big files, 10Mbit would likely _still_ be enough >> for >> the average user. > > Latency is the real killer. > > Not that management (or the telco sales guys) understand this. > > -- Ben > >
