Riverbed/WAAS for WAN optimization

jheckart · ‎05-01-2007

I am looking at moving forward on a Riverbed Steelhead solution. I have a hub and spoke network with all WAN links being full P2P T1's. I have sites ranging from 20-45 people. The main problem that I am trying to solve is server centralization at the hub site.

As of now, with sites that have a server centralized at the hub site, anytime someone try's opening a 3MB+ email, or transfers large files to a file server, there is a significant delay. I am looking for improvement on CIFS, SMB, and MAPI.

Does anyone have experience with either the Riverbed or WAAS? As I said, the Riverbed seems to be the fully baked solution, but I'm looking for thoughts on the WAAS as well.

Thanks.

dgahm · ‎05-01-2007

I did an evaluation of Riverbed Steelhead and Cisco WAAS, and tested performance with CIFS and SAN file copies. I also tested the remote print server capability of the Cisco solution, which Riverbed does not offer.

In general, the optimization performance was very comparable, but I would give Riverbed extra points for a nicer GUI, easier configuration, and more extensive performance reporting features. Cisco WAAS also requires that you dedicate one appliance to the role of central manager.

Both Riverbed and Cisco worked well with either WCCP or PBR diverting the traffic.

For us the advantages of using one network equipment vendor will probably outweigh the minor drawbacks with Cisco WAAS.

Please rate helpful posts.

Dave

mike-greene · ‎05-02-2007

Hi,

I've also been testing WAN Optimization for a while...FTP, CIFS and Citrix Printing is what I'm testing. Below is a short version of my testing.

In the environment I'm in the WAN only sees the data once and the product is built. Riverbeds technology when I tested it three months ago really did well on the second and third transfer because of the "learning" it's doing. The WAAS and Citrix solutions do some TCP flow optimization which "fool" the client/server communication into thinking they are on the local LAN kinda which make the first transfer more predictable as far as speed.

Also one issue with us was what the data looked like on the WAN after optimization because we have VPN?s that brought up by certain traffic types. With the WAAS solution the TCP headers are kept intact so you can see that it's an FTP session in Netflow or other WAN monitoring solutions. When I tested Riverbed they would put what I call a smiley face on the WAN that represented 10Mb and looking at Netflow you could not tell if that was FTP, HTTP etc. It kinda built tunnels between the Stealheads and we did not like that approach. They have probably changed this by know though.

I was also involved in the beta testing with the inline module for the WAE appliance and it performed just fine (I?m testing the 612). I'm about to start testing WCCP in our datacenter but don't expect to see a real difference from the numbers I got from the inline testing. The Riverbeds, F5?s, Citrix etc.. do seem to be easier to configure though. One concern with the current WAAS solution is the max TCP flows it can optimize is 6000. I have not tested this yet but Riverbed and other vendors probably will do more per box. A good thing is the WAAS solution now has an auto discovery for the CIFS acceleration so you don?t have to add servers one-by-one.

The technology is changing so fast that it's hard to have a three vendor bake off because when you get to the third vendor the first has released a new version with more features. . I hope though that after our solution is in place the baby sitting is at a minimum and we can forget there on the network

Good luck and I hope that helps.

jheckart · ‎05-02-2007

Mike,

I'm just at the start of this testing. I'm waiting on the steelhead's to come in now, which will be the first that I test.

The fact that you couldn't match certain traffic to be encrypted inside the VPN would be a drawback. That'll be something that I test with the Riverbed.

From the feedback that I am getting elsewhere, it sounds like the Riverbed is the clear choice. I've heard difficulties with the WAAS that cause the "cache/learning" to be wiped out after a reload, which is ridiculous. It seems like Cisco's technology in this area may be a step behind Riverbed's.

I have been watching my smallest office, which is around 20 people, and the max TCP flows that I'm seeing at a time are around 100-120. I'll have to calculate flows for all other offices, and see if the total max for all offices will be over that 6000 mark.

Thanks for the responses.

jheckart · ‎05-10-2007

Mike,

I've been held up on the Riverbed, but the WAE's are on the way.

Have you tested with MAPI at all? That's one service that I am trying to improve upon, but I haven't been able to find many other notes on MAPI.

Thanks again.

mike-greene · ‎05-10-2007

Jeff,

I have not tested MAPI only FTP, Citrix and CIFS mainly. We have a bunch of apps and it would take me months to test them all. I have not seen any issues with rebooting the WAE's yet and losing the cache. That sounds like the F5 solution...they kept there cache in RAM and not on disk. WCCP is turning out to be a pain though. I've had WAE lose communication with the router twice since testing for about thirty minutes each time. Cisco wants me to upgrade from 4.0.7 to 4.0.9 to see if that helps.

Test,test and more tests.

jheckart · ‎07-16-2007

removed