I had been interested in WAN acceleration products for years, but we had no pressing WAN performance problems so there really wasn't any way for me to justify an expenditure on a standalone WAN acceleration platform like Riverbed. However, last year our Cisco router equipment was getting near end-of-life and I decided to get a quote for new equipment with integrated WAAS. Since we were already purchasing new router gear it was easier to cost justify the WAAS equipment to extend the life of our existing links as well as allow for further datacenter consolidation of edge services. When I actually received approval, I was excited to finally implement WAAS. We've now been living with WAAS for over 6 months and the excitement has waned, not because the product doesn't have great potential, but because of software bugs and the annoying service experience when trying to get those bugs addressed.
The WAAS boxes are simple enough to integrate into the WAN environment. There are several options available, including policy router, WCCP, and inline. We decided to implement in the edge routers with WCCP. Once the devices were integrated we applied a generic default policy with no major changes from the default policy. Acceleration of services started right away, but not without issues. Here's a short list:
- Exchange Administrator would hang when run from sites that passed through WAAS boxes. This was caused by a bug in the EPM (End Point Mapper) accelerator. A case opened with Cisco led to the workaround suggestion of "disable the EPM accelerator". Well, this certainly worked, but also disabled all MAPI acceleration for Outlook clients, one of the major reasons we purchased WAN acceleration. After about 2 months of back and forth Cisco released a update (4.1.1c I think) which corrected this issue.
- HTTP Acceleration would hang for an entire site every once in a while. Once again I opened a case, provided traces showing the problem, and was pointed to a known but. Once again I was unable to get a fixed piece of code and the workaround was to disable the HTTP Accelerator. I hate this "workaround" as I'm not sure why I would ever purchase a product just to be forced to disabled features. After many weeks of hammering them I did finally get a engineering special with some debug features that did correct the problem. I believe that 4.1.1d included the fixes for this.
- CIFS Acceleration would completely break CIFS access from Redhat Enterprise 4 & 5 clients to other Redhat systems running SAMBA. Once again the solution was to disable CIFS Acceleration (see a pattern here?). After much arguing, gathering traces, etc, etc, I finally managed to get Cisco to reproduce this bug almost 3 months after opening the case. The bug was identified and now a fix has been coded, but I've been told the fixed code won't be made available to me until November. At that point it will be almost 9 months from when I opened the case. So far they've refused to give me an engineering special with the fixed code.
This is basically just a short list of the major problems I've had with WAAS. While the fixes were slow in coming, I can say that currently we have no major issues except for the CIFS Acceleration bug. The WAAS boxes have been very stable for the last 6 months and they have allowed us to consolidate several of our servers into our datacenter while maintaining LAN like performance for the clients, including clients in Europe. Typical compression is about 3x with some things, like printing and email being as high as 8x. When working the accelerators provide a tremendous boost in perceived performance.
The latest 4.1.3 code introduced SSL acceleration, and we just recently implemented this for testing with our SSL enabled ERP application. Performance improvement is significant, especially during initial download of the java based client applications, saving significant WAN bandwidth when the client is updated. It's a little bit of a pain to configure, but does work as advertised.
We're hoping that Cisco eventually gets us fully working code, it's now September and they are promising code that fixes our CIFS problem by November so we're hoping that, over a year after implemention, we'll finally have the WAN acceleration solution that we had hoped for.
Sunday, February 28. 2010 at 15:12 (Link) (Reply)
Friday, March 5. 2010 at 21:52 (Link) (Reply)
Overall, we see around 2.5x the WAN link performance, and the system works quite well for Outlook, CIFS, printer, and even SSL acceleration. A friend with Riverbed seems to get significantly more CIFS acceleration for things like copying small files, etc, but WAAS isn't horrible, at least in our setup.