- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
HA with load balance all retransmission
Hi,
We have a Fortigate HA with load balance all enabled, and we are monitoring the behavior, and I can see many retransmissions between slave and master, just when the slave unit process the packet, see this picture:
Anyone known if this is normal? when we have load balance all enabled?
Regars
Lucas
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello Lucas,
Yes, it is possible if the traffic is high. load-balance-all option will make the primary unit load balance all TCP sessions.
It really doesn't help to improve the throughput because of extra overhead required for load balancing. So load-balance-all is disabled by default.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi vjoshi,
Thanks.
The strange thing is that this environment is lab, has no charge. :(
Regards
Lucas
vjoshi wrote:Hello Lucas,
Yes, it is possible if the traffic is high. load-balance-all option will make the primary unit load balance all TCP sessions.
It really doesn't help to improve the throughput because of extra overhead required for load balancing. So load-balance-all is disabled by default.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello Lucas,
Sorry, what do you mean by 'no charge'?
If it is in lab and you are testing this, fine.
If this is just a test bed and you are planning for a production setup, and if you really want to use the HA for optimum load balancing, I would suggest virtual clustering with VDOMs.
This is possible in a-p mode where each device will be master for specific VDOMs and other device will serve as Slave unit.
Cheers,
lpiris wrote:Hi vjoshi,
Thanks.
The strange thing is that this environment is lab, has no charge. :(
Regards
Lucas
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Joshi,
Sorry, I wanted to talk without load.
Actually we have a HA in production, And we have this behavior.
So, We made lab with two 110C FG and we can see the same behavior.
Regards
Lucas
vjoshi wrote:Hello Lucas,
Sorry, what do you mean by 'no charge'?
If it is in lab and you are testing this, fine.
If this is just a test bed and you are planning for a production setup, and if you really want to use the HA for optimum load balancing, I would suggest virtual clustering with VDOMs.
This is possible in a-p mode where each device will be master for specific VDOMs and other device will serve as Slave unit.
Cheers,
lpiris wrote:Hi vjoshi,
Thanks.
The strange thing is that this environment is lab, has no charge. :(
Regards
Lucas
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello Lucas,
I got it. It is strange it happens without load.
Is it causing any specific issue to the application?
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Josh,
Yes! this customer have an internal application that are impacted.
We had to change ha mode to a-p, and it works fine.
I will monitoring anothers HA, To see if I can find the same behavior.
Cheers
Lucas
vjoshi wrote:Hello Lucas,
I got it. It is strange it happens without load.
Is it causing any specific issue to the application?
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello Lucas,
Do you see the same behavior in your lab setup as well?
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Josh,
yes.
vjoshi wrote:Hello Lucas,
Do you see the same behavior in your lab setup as well?
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello Lucas,
Weird, I expect it to happen, but without any traffic doesn't seem to be correct.
I would recommend not to use the load balance all, instead use the virtual cluster for effective load sharing.
