Unsolved
This post is more than 5 years old
3 Posts
0
730
November 11th, 2015 04:00
Separating PGs by traffic type
Hi,
I have a customer, which has exclusive access to 8 FE-ports on our 3-engine 10K box. The customer basically has 2 types of servers: databases (OLTP type) and ESX server farms (running various applications - fairly random I/O patterns). These normally are rather different I/O types.
My question is, if the best solution is to use all the 8 ports in a single group for all servers and LUNs, or would it be better to create 2 x 4 port PGs, and use each group to the servers with a certain type of traffic?
Thanks.
No Events found!
StoragePupil
3 Posts
0
November 12th, 2015 03:00
Hello Andy,
It is advisable to use a certain set of Ports for specific OS Hosts i.e. You can use 4 Ports for ESX and remaining 4 for Unix Servers.
andy_et
3 Posts
0
November 12th, 2015 05:00
Hi,
OK, but I still cannot see the reason behind. Why and how does this help serving I/O requests from various traffic patterns? Why and how would that effect VMAX performance, as recognised on the hosts? "Inside" the box it is the same, regardless if the requests are arriving on 2 x 4 or 1 x 8 pathes, so the difference must be somewhere at the front-end handling of these requests.
So far the best explanation that I could find was that it is a tradition, a rule of thumb...
umichklewis
4 Apprentice
•
1.2K Posts
0
November 18th, 2015 03:00
The reasoning to segregate traffic based on access patterns. Your ESX hosts will have a wider array of random I/O, while the OLTP servers might have more sequential reads and writes.
On the host side, you'll see this as a lower response time. Let us know if that helps!
Karl
andy_et
3 Posts
0
November 18th, 2015 23:00
At the end it turned out, that our customer using these ports have two clients, and they want to divide traffic on a much less technical basis: to separate by client. So the whole discussion above became pointless.
Thanks for your help, anyway.