In the first thirteen years of my working life I have encountered a lot of different people in the field of networking. And for some reason they were all skilled, experienced and willing to learn. They understood their stuff, in case of issues a pointer to the right direction was enough to help them out.
In my experience at the Riverbed TAC I have encountered several new kind of networking people, although I wouldn't call them all "networking" people.
Did I miss a category? Most likely because I have repressed them, very deep...
Even after working for Riverbed Technology for two and a half years now, I still have to come up with a bullet-proof analogy of how WAN optimization works.
Consider a network from one side of this planet to the other side: A round trip time of 300 milliseconds for 20 million meters. Of that 300 milliseconds, you have two factors: The speed of light to get from A to B, and serialization delay which happens at every hop and is related to the bandwidth and the size of the packet. One is constant, the other one is variable.
To move data in a stream, the throughput is limited to the smallest bandwidth in the path. Although you can move data faster on other parts, it all has to go through this one.
As an example, say you have a stream of 300 Mb and a smallest bandwidth of 2 Mbps. Without any protocol overhead, this will take 20 minutes to go through there. With a fourty byte protocol overhead and an MTU size of 1500, this will take 20 minutes and 33 seconds.
Now with WAN optimization. It consists of three parts: Optimization on TCP level, which has been ignored for now. Latency optimization on application specific protocols, which has been ignored for now. And data optimization, where the data is either compressed or only referred to. If the same or similar data gets transfered via two WAN optimizers twice, the first time you would get a relative small reduction factor, depending on the compressability of the data, while the second time you would get a large reduction factor because the data patterns is alreayd known on both devices.
If that 300 Mb is split into segments of 1024 bytes, making it 300 000 segments, and each segment has a 64 byte label, you end up with only about 19 Mb worth of labels.
Transfering that 19 Mb through a 2 Mbps link will be take 80 seconds, about 15 times faster.
Now back to the topic: A good analogy for WAN optimization.
Is it faster than light? It feels like it, but the Round Trip Time of the WAN is still the same. And the speed the packets go via is still the same.
Is it a "wormhole"? Wormhole-based paths which are shorter than a non-wormhole-based paths. The path travelled travelled for optimized traffic still has the same distance.
Is it comparable with ships, where goods are stored in containers (labels) and then transported in large bulk carriers? If the speed limit of other ships was limited to the speed limit of the bulk carriers, then it would be a good start.
Is it a train analogy, where passengers are cramped into carriages and efficiently transported across the rail network? It could be, except that on the railroad network everything is put into train carriages and transported efficiently on it. Comparing it with the French TGV and the Japanese bullet trains does not work neither, because the speed of the packets is still the same while these trains are way fast.
So, the analogy needs to use the same speed limits on the transport mechanism, and needs to give the impression that the delivery gets faster without changing the distance.
The best thing I come up with is transport of goods via large trucks instead of via small delivery vans: Goods are shipped via small delivery vans to a distribution point, stored into a single large truck which then uses the same transport infrastructure as small fast vans would have used if they would have transported their payload. Instead of a long convoy of small vans, you get one truck towards the distribution point which there gets reloaded into numerous small vans. The only thing which does not make sense yet is that small delivery vans are often 2 x 4 x 1.5 meters and big trucks are 3 x rather long x rather high, which gives the impression that size still matters while this isn't the case on WAN optimized traffic...
That is the problem if you work with magic :-)
Just for the fun of it, I checked out some Riverbed appliances and hardware on eBay.
So far no luck, still haven't been able to find a full working Steelhead appliance on eBay. Time to go to bed!
The slides of my talk at the AusNOG04 conference about "WAN Optimization - Changes to more than just a box".
Posted on 2009-10-12 22:00:00
Guess what! After a "Social Media Introduction Course" last week, today I've been approved as an author for the Riverbed Blog Website.
I still don't know what the first article will be about, but I will give it my best try (Did you expect anything else here?)
It seems a long time ago since I've started at Riverbed TAC, but it's only four months (three months if you consider that the original training didn't finish until I was there for four weeks). So far so good, I completed the probation period of three months and this morning I sat for my RCSP exam.
I don't really like the style these exams are being taken in: You get 70 questions, some with one correct answer and some with multiple answers you need to pick. So after 30-45 minutes, you have enough of it and want only to know one thing: Do I have to come back next week or can I not worry about it for the next two years.
But the final result doesn't matter, you are given a list of topics and the percentage of how much of each topic you had correct. So you don't know which ones you had wrong. Or if the ones on which you doubted yourself on were correct or not. It is just a snapshot of how much you know now and if you get a lucky batch of questions you pass. These exams are not for learning from your mistakes, not meant to improve yourself.
Now the good news, I don't have to do it again for another two years!
Part two of the training (hands on, dirty details, procedures) was done in the Sunnyvale offices of Riverbed. Were the offices in San Francisco in the middle of the city (well, mostly), these ones were in the middle of Sunnyvale. Going there would be simple, grab the BART train, grab the Caltrain train and grab a taxi for the last two kilometers. I mentioned the issues with BART earlier, now its CALTrains turn.
Caltrain was GREAT! It started with the buying of the ticket where I didn't get 14 dollars back in quarters, I got 14 dollars back in dollar coins. I knew about the existence of silver dollar coins (collector items), but that was the first time I have seen normal dollar coins. Later in the week I found out that it costs the US government about 500 million dollars per year to replace all the worn-out dollar bills while it would only cost a fraction of that if they moved to dollar coins which would last about 30-40 years before they need to get replaced.
The second great thing was the bell on the front of the train, which gets sounded when it arrived at a station or a railway crossing. The third great thing were the railway crossings, something which I haven't seen for a long time but remember from an earlier life: One in eastern Eindhoven and one in Geldrop. Bing-bing-bing-bing they go when the train drives by. The fourth great thing was the presence of a train conductor which was there for questions and support. I am not sure if they are on the CountryLink trains in Australia, but they surely are not in the Sydney CityRail trains.
The design of the interior of the train is euhm... interesting. (Fifth great thing!). To get in the train you have to climb from the platform (which is at a certain height) up three steps into the train (which is thus higher). That is the ground level of chairs, in two rows with each two chairs on your both and left hand side. Then the upper level of the train, which is accessible via a step of stairs leading you to about one-and-a-half meter above the ground level. That distance is not high enough to make it possible for the people on the ground floor to stand up, so they have upstairs only one chair per row, both on the left and right hand side of the train, and a huge gaping hole in the middle of the upper level. Like I said, interesting design but not worlds smartest I think.
Two zones later, from Millbrae to Sunnyvale and I was at the place I would call home for the next week: A Best Western hotel. The week before I was in a hotel in inner San Francisco, now a Best Western hotel. Let me describe the differences:
|Queen-size bed||King-size bed|
|Shower and bath||Shower and spa|
|Internet (wired and wireless)||Internet (wired and wireless)|
|42 channels on TV||At least 99 channels on TV|
|Movie on Demand ($$$$)||Video library (Free!)|
|Breakfast ($$$$)||Breakfast (Free!)|
|Dinner / Bar||Pizza / Starbucks near the supermarket|
|Soda vending machine||Soda + Snack vending machine|
|10 minutes walk to work||10 minutes walk to work|
|US$ 279 per night||US$ 119 per night|
On average, Sunnyvale was better for the money, but San Francisco had the location. And less crap on TV :-)
Because the availability of the video library, I took the opportunity to watch some old movies: Timeline by Michael Crighton (for medieval castle lovers), Snatch (very funny movie) and Rat Race (always good for a laugh). There was a fourth one but I can't remember anything about it.
Just as I met up David Thiel and Anton Holleman in San Francisco, I met up with people here too: Jos Backus, former colleague of Origin and FreeBSD enthusiast, Xen Li and Marcel Moolenaar, both FreeBSD developers. Thanks for the hospitality guys!
The time in Sunnyvale flew by and before I knew it was time to go back to Australia. This time I was lucky on the plane: The two people next to me didn't show up and I moved from an aisle seat to a window seat (I did offer it first to a mother with her child but she refused for some silly reason) and I slept for about ten hours on the plane. Business class comfort for an economy class price!
At home, Project Vegetable Patch needed moral support and some pruning, but it has survived my absence. Little Dirk had his hair cut, Hanorah has learned to walk and Naomi was very glad to have me back and take care of the kids for the rest of the weekend :-)
This course about the Riverbed Steelhead appliance has given me some very interesting details of the design of the Steelhead appliance and the capabilities. The speed improvements on network traffic, either due to initial compression, or later when it detects patterns it has sent out earlier or because it optimises protocols like CIFS, NFS or HTTP(S), their are very impressive. As they say, it makes your data on the servers on the data-center feel like it is sitting under your desk. Plus the integration in your network makes it close to zero-administration, all you need to start is two Steelhead appliances, two ethernet cables, two IP addresses and their default gateways and it's up and running. Talk to me if you want a demo :-)
What happened the rest of this week? The USA got a new president, I met up with David Thiel for dinner on Sunday, I met up with Anton Holleman for dinner on Wednesday and Saturday, I met a lot of collaegues at the San Francisco Riverbed offices, I walked in San Francisco through Chinatown, the Financial District, to the Tram museum, around the piers at the east coast because I couldn't walk over the Bay Bridge (Whoever decided to make a bridge without a pedestrian lane?).
Television is here euhm... less than interesting. Except for PBS there aren't much channels I did recognize or would recommend. Discovery Channel has "in-taxi" gameshows, there are seventy channels with people competing for something and then giving or getting feedback on how they were doing and half of the TV series are an insult to my brains. The presenters of the Weather channel seem to be the most enthousiastic ones in their profession. At least in Australia there is 40% of the channels worth watching, it's about 2.5% here. Not to mention the quality of the surfaces with the "same" colour on NTSC... (enough for now, it just makes me open the window and shout "I am sick of this and want everybody to know")
The best pub I've been in was Eddie Rickerbacker's on 133 2nd street. It has old motorbikes hanging on the ceilings, miniature trains in the vitrines and a miniature railroad around the wall. And a fat cat walking around... A very nice place to be.
So what is left for the rest of the week? Tomorrow I will go to Sunnyvale, first with the BART train service and then with the CalTrain service (Yes, I love trains!). Over the week I will have my share of cases at the Riverbed TAC in Sunnyvale.
In the next two weeks I will be on training for my new job at Riverbed in San Francisco and Sunnyvale. It will be two courses, Steelhead Appliance Deployment and Management and Steelhead Mobile Client Installation and Configuration.
A Steelhead Appliance is a device in your network, one centrally at your servers and one or more remotely on the user LANs, which provides WAN optimisation, WAN acceleration and Wide Area File Services. Or in plain English: It makes the data over your networks be transferred faster.
The best comparison for it is your webbrowser: When you initially a page with images, it will download them all from the webserver and store them in a local cache. If the next page you go to refers to these images, it will just use them from the cache instead of downloading them again. That is Wide Area File Services.
The next one, WAN optimisation, is the HTTP protocol. The old version setup a new TCP session for every HTTP request, the new version can continue on an earlier used TCP session.
And the last one, WAN acceleration, is compression of the static HTML pages by the webserver: Compressed data is often smaller, so it will be there faster.
The Steelhead appliances don't only work on the HTTP protocol, but also (including but not limited to) with CIFS (Windows File Sharing), MAPI (Exchange server), NFS. And for protocols it doesn't know about, it will still do compression.
That is the hardware version, which works on your WAN/LAN infrastructure. There is also the version called Steelhead Mobile for on notebooks computers, it will give you, at home over a VPN for example, the same performance boost as you have on your LAN at work.
Like I said, the first week is training, the next week is for getting my troubleshooting experience. From what I've read up so far it is pretty technical stuff I will get trained with, so it can't be all bad :-)
If you are in San Francisco between 2 November and 9 November or in Sunnyvale between 9 November and 14 November and want to meet up, drop me an email!