[j-nsp] SRX advice

[Julien Goodwin]

On 04/02/11 16:12, Ryan Goldberg wrote:
> watchguard a) is the outbound nat box for about 70 small offices (we are a small ISP too, these are fiber-connected customers).  it also handles some amount of inbound nat for those customer's various servers, which may be in the customers office, or a virtual host in our racks.  and maybe a half dozen ssl-vpn road-warrior types.  There's also a dozen or so lan-to-lan ipsec tunnels on it.  sustained 2-20 inbound.  light outbound.

That's an odd setup.

> watchguard b) is for internet facing windows boxes.  lotsa inbound nat.  sustained 2-20Mbit outbound
> watchguard c) is for our office, 55ish users.  some inbound nat too.  0-50Mbit inbound, widely varying
> watchguard d) is for one particular hosting customer where stability is paramount.  The other firewalls get touched a lot (and as of late, have been puking when they feel like it).  2-15Mbit of sustained web traffic, with the odd spike or lull.

These three are fairly trivial.

> a 2821) terminates a bunch of lan-to-lan ipsec tunnels (VTI style) to 1841s all over the place.  box is completely VRFed, no global table, all the tunnels land in the INTERNET vrf and pop out in customer vlans, each their own vrf.  10-30Mbit
> 
> So - goal is to collapse all this onto a single pair of boxes running in an HA config.  Watchguard a, b, and c are problematic, and are becoming more problematic.  watchguard d is pretty quiet, but we are contractually obligated to remove all SPOF from that clients setup.  the 2821 is very quiet, no troubles.  

> My main question revolves around number of virtual routers.  We can't afford a big enough box to stuff everything (as in, every customer network) in its own vrf/routing-instance.  I will admit that I've become hooked on using vrfs in cisco land on ISRs (a lot of double-ISP configs, random dirty hacks).  But for our future firewall setup, I don't know what a bunch of routing-instances really buys us, if anything (aside from the psychological aspect).  All we really need is for all the private networks behind this thing to get natted to their corresponding public ip(s), and if something behind the firewall needs to talk to something else behind the firewall, it should go out and back in (getting source nattted, then dest natted).  If the J-boxes can do that without separate routing-instances, then we're good.

You only need instances (on SRX) for two things:
* Overlapping IP's need forwarding instances
* Multiple protocol instances (eg, seperate OSPF on either side) take
non-forwarding instances

> My other question involves HA stability.  I've seen instances with other kit where introducing "HA" actually reduced availability.  SRX boxes like running in HA, or are they fussy?

I don't run SRX's in HA, the only issues I've had in production are
firmware related (note that you can't do an online firmware upgrade of a
HA cluster, there's still a hit).

I would suggest an SRX650 over the SRX240, would be far more confident
in that handling your traffic load (on the services side, forwarding
should be trivial for even an SRX100).

-- 
Julien Goodwin
Studio442
"Blue Sky Solutioneering"

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 262 bytes
Desc: OpenPGP digital signature
URL: <https://puck.nether.net/pipermail/juniper-nsp/attachments/20110204/8bbf68f2/attachment.pgp>