Throughput calculation

I'm not 100% on how the F5 figures are calculated but keep in mind, if this is say HTTP, you are sending 2G of requests, the responses will likely be somewhat bigger and certainly won't be the same.

 

 

I'd read that as 2G in from the 'client', 3.3Gb out to the client from the servers.

 

You're probably better off using Module Statistics > Traffic Summary Statistics to drill down and see stats for client and server side.

 

Thanks , where on internet can i read about what you mentioned " I'd read that as 2G in from the 'client', 3.3Gb out to the client from the servers. " , as i googled & cannot reach info

I'm not sure what you mean SSHSSH. I mean, I understand that to mean you are sending 2Gb of test requests and as the overall figure is 5.3Gb, you are probably seeing 3.3Gb response data. This is the nature of HTTP; a request will be small in comparison to a response in most cases.

 

 

As I said Module Statistics > Traffic Summary Statistics will give you the facts.

Thanks , i got it

 

is there something standard explaining approximate numbers about ratio between HTTP request & response

 

Also at my environment servers is accessed by Client through F5 , then server initiate connection to other DB server passing through F5 again to get client request , then response is sent back to Client

 

if i understood then if for ex data is 2 G , i will need throughput as : [ 3.3 ( 2G + Response ) ] * 2 >>>> * 2 because server initiate again through F5 ... am i right ?

 

Hi SSHSSH

 

 

I don't think there is any documents describing the ratio between HTTP requests and responses, but maybe it would be time well spent if you used some time on reading about HTTP. When the TCP session is created to a site like google.com, the browser will request the root of the site. This request will usually be a in the size of a few hundred bytes (346 bytes in this case of testing). When the HTML of the site has been returned to the client (in a 605 bytes large packet), the client will begin requesting all the elements of the site that was not delived in that HTML (pictures for example). These elements might not be able to be transferred in one TCP packet, but might require many packets = much bigger reponse than request.

 

 

I recommend downloading Wireshark and try it out on a very basic site without SSL (will hide all the fun stuff). This can be very informative and quite fun if you like that kind of stuff :)

 

 

The same goes pretty much for SQL. I don't really know much about SQL traffic patterns, but probably it's a bit more "equal" in terms of up/download. I base this on the fact that even though you might make a query that received a lot of information, you probably also does updates, inserts, deletions where the inital request is actually bigger than the response from the database.

 

Thanks David.

 

 

SSHSSH, can I repeat that you should drill down into the statistics and gain an understanding from that. I've struggled with this myself in the past but they stats are very good these days. If you wanted to test and record stats for your 2Gb test;

 

 

1) Clear all the network interface stats (we're use these but remember these will include TCP/IP protocol overheads)

 

2) Run your test with 2Gb of requests

 

3) Check the interface stats as necessary. In and Out for the 'outside' interface, In and Out for the web server interface and in and out (if possible) for the web to SQL servers. Remember that 'out' for the server interface will equal out for the client side interface.

 

 

Understand? I'm tempted to do a diagram but have no time! =]

Thanks David & Steve

 

Hi Steve , i will do . one more Question ... on performance Tab i can see Client side Throughput & Server side Throughput ... what is that in my case ?

 

in my Case client access servers on external Vlan VIP , Also note that servers behind F5 have both in & out legs connected to F5 & it uses IN Leg to rx request from Client & respond , while it uses Out Leg to get client needed data from Database servers ( passing through F5 as well )

 

I assume you're looking at the Traffic Summary > General page? Sorry I can't check as I've a few problems with my VE at the moment. That would be the wrong stats to look at as they are system wide.

 

 

Ideally you'd want to look at a number of statistics and compare them too to ensure everything matches up. So, if you had dedicated physical interfaces for client-side, server-side and server-sideDB you'd clear the counters, test and then look at: Statistics > Module Statistics > Network and get the figures for all the network interfaces (in and out). Then you would take a look at Statistics > Module Statistics > Local Traffic and gather the statistics for the Virtual Servers and Pool Members, in and out and they should look similar.

 

 

Of course, if you're tagging VLANs etc. you'll have to just rely on the LTM statistics and in this case, that might be better for you. The F5 is a proxy so you'll have to work each step out;

 

 

1) First VS, client-side is the actual PC/test device, server-side is what is returned by the real servers in that VSs Pool, server-side is the F5 VS

 

2) First VSs Pool, client-side is the F5, server-side is the real server(s)

 

3) Second VS, client-side is the real servers, server-side is the F5 VS

 

4) Second VSs Pool, client-side is the F5, server-side is the real server(s)

 

 

Just try not to double-count anything! If I were you, I'd draw a diagram to understand all the flows and what stats related to what host etc. Remember, server-side stats for VS1 will match client-side for Pool1 but not server-side as the same servers are talking to the DB/backend servers etc. etc.

Here's a diagram as promised. This is for a single VS, I'm still working on the double-bounce scenario.

 

 

One interesting thing I discovered; there don't seem to be any statistics for VLANs in the GUI. Haven't checked tmsh yet but I'm sure they are there.

 

 

Feedback good or bad is welcome. Sorry for the poor resolution, DC is shrinking my 262K file to 22K for some reason.

 

OK, I've put a full res version on SkyDrive here: https://skydrive.live.com/redir?resid=CDED4624AAE21FEC!136&authkey=!ABoge_TmP7ycGM8