Awesome Conferences

WAN Accellerators

What is your opinion of WAN Accelerators? Please post to the comments section.

I haven't used or configured a WAN Accelerator but from the reading I've done so far it seems that the older the protocol is, the better a W.A. will help. Older protocols were designed when we were ignorant of the networking realities we have today. More modern protocols tend to do their own compression, caching, don't do stupid things that fail over high latency links, and so on. In particular: Am I right but if you mostly have home-grown protocols, you can tune them better than a W.A. can accelerate them?

Please post to the comments section.

Posted by Tom Limoncelli

No TrackBacks

TrackBack URL: https://everythingsysadmin.com/cgi-bin/mt-tb.cgi/1591

4 Comments | Leave a comment

Tom,

(disclaimer: I'm working in the Riverbed TAC)

There are several aspects of WAN optimization:

- Transport layer optimization: TCP has lots of features to improve on network related issues (packet loss, LFN, window sizes, forward error correction) which are not implemented on all TCP stacks. Being able to do these things without upgrading all computers behind a WAN optimizer is nice.

- Protocol optimization, where chatty protocols (MAPI, CIFS, Lotus Notes, FCIP, NFS etc) are dealt with by the WAN optimizers so that various read-ahead and write-behind operations can be done before the client asks for it. SMBv2 is better than SMBv1, but still...

- Data reduction, where traffic which has already been seen (seen when uploaded via SMTP, then seen again when downloaded via IMAP) won't be send as-is but is send as a reference. At the receiving WAN optimizer it then will explode again into the original data. Even if the application already compresses it, disabling compression and let the WAN optimizers do the magic will give better reduction.

- Encryption: Some protocols have their payload encrypted (MAPI, CIFS, SSL encrypted stuff like HTTP and IMAP/TLS and SMTP/TLS). When you are able to decrypt the payload, you can do the mentioned data reduction on it. Send it secure to the other WAN optimizer, which then explodes it again and re-encrypts it again to the receiver. Some call it breaking encryption, some call it MiTM, some call it re-encryption: Check the security policy of your company to decide if you want to use it or not.

Hope this helps understand what WAN optimization does.

Edwin

In my experience a lot of older (and not so old) TCP/IP stacks - I'm looking at you HP ILO right now - do not tend to cope well with Jumbo frames - which is the first trick these accelerators try

We have a lot of protocols that bypass the accelerators after stuff ups. Make sure you test a lot, and are on hand to disable protocols or clients when you go to production

John

In my experience Wan Optimizer products come in two flavors, systems designed for Data Center to Data Center optimization (typically for data replication optimization), and systems designed for User to Data Center optimization (which optimize for many users accessing the same data).

Riverbed falls into the User/DC solution space, and I've been working with them for five years at my company. My experience with them has been very positive, and I wouldn't consider building a global corporate WAN without them (or a similar product) at this point.

On average the reports show greater then 3X improvement in network speeds, but that's only really a part of the picture. I tend to refer to Riverbed appliances as effectively changing the speed of light, reducing the impact of latency on TCP connections and in some cases making TCP connections flow at otherwise impossible speeds.

John,

Jumbo frames are strictly speaking Ethernet layer. Unless you tell your WAN optimisers that the size of an Ethernet frame is more than 1500 bytes they won't change that.

If you mean TCP windows scaling, that is negotiated during the setup of the TCP session.


With regarding to "lots of protocols that are bypassed after stuff ups", you should talk with their TAC to make sure this gets fixed or at least documented.

Edwin

Leave a comment

Credits