Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

I guess the subject says it all: would switching to Windows HPC Server 2008 R2 have any effect on average network latency ping times? Even if it shaved a few milliseconds off the time, I'd be happy - but is that realistic?

share|improve this question
3  
Under what circumstances do you need to shave a few milliseconds for any meaningful result? –  Chris S May 25 '11 at 20:40
    
Sounds like a latency-sensitive application (e.g. finance). –  ewwhite May 25 '11 at 20:51
1  
@ChrisS High-Frequency trading is the one that springs immediately to mind. Those guys do some serious black magic to shave off tenths-of-milliseconds. –  voretaq7 May 25 '11 at 20:52
add comment

2 Answers

up vote 5 down vote accepted

The only thing that HPC would change in the networking stack is RDMA (Remote Direct Memory Access), which is used in computing clusters. Normal network operations are not changed.

If you're looking to shave latency, consider purchasing a NIC with a full TCP Offload Engine (TOE).

Edit

Also of note is this quote from the HPC FAQ:

"Both Windows HPC Server 2008 R2 Suite and Windows Server 2008 R2 HPC Edition are licensed solely for running clustered HPC (high performance computing) applications."

share|improve this answer
3  
+1 for the TCP Offload Engine -- in my list this substantially reduces delays from item #5 (the network card can handle a lot of work without having to bother the OS). Most server-class cards have a TOE in them, but if yours doesn't It can be a noticeable improvement in performance. –  voretaq7 May 25 '11 at 20:45
add comment

Changing your OS is unlikely to have a net effect on ping times -- Latency is affected by the following factors (not a comprehensive list, but in a rough kind of order)

  1. Distance to be traveled (Speed-of-Light limitations
  2. Equipment traversed (Routers/Switches along the way prioritizing packets)
  3. Line Quality (more apparent with TCP & retransmission, but if you see lots of packet loss...)
  4. Available bandwidth (similar to #3 - if packets start getting dropped)
  5. Your operating system (delays shoving stuff through the network stack)
  6. Your network card (assuming you're not saturating its buffers/engine bandwidth, otherwise it could be #5)
  7. General Network Conditions (traffic surges, etc.)
  8. Cats chewing on the ethernet cable
  9. Sunspots

There's very little you can do about 1 & 2. You can look after 3 within your own walls. Changing anything below about #4 is usually subject to diminishing returns ; there's nothing you can do about 7, 8 or 9.

share|improve this answer
3  
You forgot unicorns and rainbows. They're right between 8 and 9. –  Hyppy May 25 '11 at 20:44
1  
Rainbows. Those things wreak havoc with your data transmission times. –  Holocryptic May 25 '11 at 20:46
1  
That's why we use monomode fiber cables. None of the rainbow-prismatic effects you get with that multimode junk! –  voretaq7 May 25 '11 at 20:51
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.