Re: [lvs-users] https slow using LVS-NAT

To: " users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: [lvs-users] https slow using LVS-NAT
Cc: Graeme Fowler <graeme@xxxxxxxxxxx>
From: Dirk Bonenkamp - Bean IT <dirk@xxxxxxxxxx>
Date: Thu, 14 Oct 2010 14:22:43 +0200

Telefoon 023-5422299
Fax 023-5422728

Op 14-10-2010 12:33, Graeme Fowler schreef:
> On Thu, 2010-10-14 at 10:32 +0200, Dirk Bonenkamp - Bean IT wrote:
>> I'm having some trouble with https over LVS-NAT. In short: it works, but
>> it is slow. Much slower than http.
> <snip>
>> Any ideas on what I'm missing here..?
> Possibly.
> Are all of your machines in this testbed connected to the same switch?
> Does the director have two separate NICs with the ranges on, or one?
> In an idealised environment where the three machines are connected
> together as simply as possible it's quite likely that you'll "stack"
> small latency figures which quickly add up to what might look like a
> significant delay. However, it might not be that significant - if all of
> your machines are connected via 100 megabit or gigabit ethernet, that's
> unlikely to reflect the situation with the actual clients who will be at
> the far end of a long piece of wet string.
> Obviously there is an overhead with connecting via the director,
> especially if it's only got a single NIC - but you might need to explain
> your testbed a little more clearly first.
Thank you for your reply.

My setup:

The directors VIP is on eth0, eth0 is connected to a 100 megabit switch.
The directors gatewayIP is on bond0, which is a mode 4 bonded interface
(eth1 & eth3). bond0 is connected to a stacked gigabit switch (each
interface to a different switch).

The realserver is connected to the same gigabit switches with a mode 4
bonded interface (eth0 & eth1).

Because the testing through the VIP goes over a 100 megabit switch, I've
also connected a 100 megabit switch to the gigabyte switch. The
'internal' test machine is connected to this switch, so both test a
through one 100 megabit switch.

I just redid the tests with the same machine (I previously tested with 2
different test machines). I've also included figures for a single
machine, online, with the same apache setup (we are on a 2 megabit

        Direct on RS       Director        Online
http    0,025 s            0,06 s          0,2 s
https   0,7 s              4,7 s           2,8 s

Looks to me like https through the LVS director is too slow, and I
probably misconfigured something... Hope this gives the gurus some more

Thanks, Kind regards,


Please read the documentation before posting - it's available at: mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
or go to

<Prev in Thread] Current Thread [Next in Thread>