The expression level of the control gene should be in the same range of expression level as your target gene. The control gene should react neutral on treatment. It is better not to rely on only one control gene, but more, it is possible then to compare the control genes before and after treatment, and select the best one or two control genes.

It's definitely worth trying a few options and seeing which are optimal for your condtions. A number of companies sell "kits" containing primers to various putative 'stable' reference genes (I've had decent success with primerdesign's genorm primer set), which you run against a selection of your samples and then plug the values through programs like geNorm, normfinder or bestkeeper: these will then tell you the most suitable candidates. 

And as Martin says, use multiple reference genes. RNA is much less stable than protein, and subject to many more (fairly significant) potential perturbations over the timescale of sample collection >>analysis. Using multiple reference genes is always wise. If you have two or three reference genes and they all agree well, you can be much more confident that measured variations in your gene of interest are valid.

Thanks a lot for your answers  Martin and John. Concerning kits for reference genes suggested by John , they are expensive and unfortunately it will be difficult to afford them.

One more thing i'm worried about in trying more than one reference gene  is the amount of RNA sample i have . On average the samples are 30ul of  a concentration of about 15-20 ng/ul. Would  this amount be enough to try several reference genes plus testing  my target genes which are six -taking into consideration the double or triple repeats to be done for each sample??

Depends entirely on the genes you're looking for: abundant transcripts can be easily detected with even tiny amounts of starting material, while rare transcripts may be difficult to quantify with even large amounts.

600ng of RNA is not a lot, but it's certainly enough for cDNA synthesis.

Assuming a 1:1 RNA:cDNA conversion, I typically use 2-5ng.well-1for fairly abundant transcripts (though I could easily go lower, I don't like ending up with buckets of different tubes at different [cDNA]) and 25-30ng.well-1 for rare ones.

If you're worried about using up vital test samples just to establish reference genes, you might consider preparing some RNA from just generic untreated blood, which should be easier to obtain in bulk, and then using that as a first pass strategy to narrow down candidate genes. Three or so separate blood samples, prepped into RNA and maybe each made into two or three separate cDNA preps..should give you enough sample-to-sample variability to identify four or five good candidates, which you can then run on a small subset of your test samples to narrow it down to two.

There are qPCR machines that work in low volumes, maybe look around if someone has a nl scale instrument. Either the Quantstudio (Thermofischer), Fluidigm or Wafergen.

Thank you a lot John for your detailed explanation. Thanks Martin for your suggestion.