Get hands-on experience with 20+ free Google Cloud products and $300 in free credit for new customers.

Google Premium Network Tier doesn't use Cold-Potato routing as advertised?

To test the effect of Google's premium network tier, I created two VMs in the Netherlands, one with premium networking, one with regular tier. To test the performance, I picked a random IP in the USA and did two traceroutes, one from each of the two VMs.

Curiously, the public internet route is more efficient than that chosen by Google, which contradicts the claims made regarding premium tier. As expected, the standard tier VM uses the public internet already starting in Amsterdam. Based on Google's claim of performing cold-potato routing for premium tier, I would have expected the premium tier traceroute to only enter the public internet somewhere in the USA. To my surprise, the first traceroute hop of the premium tier is also in Amsterdam, and it in total needs more hops to get to the destination (12 instead of 7).

Question: Why is't Google transiting the traffic inside their WAN until closer to the destination, e.g. have the traffic exit onto the public internet in the USA (instead of already in Amsterdam)?

Standard:

 

 

traceroute to 66.96.149.26 (66.96.149.26), 30 hops max, 60 byte packets
 1  * * *
 2  be2175.ccr42.ams03.atlas.cogentco.com (154.54.61.41)  5.603 ms  5.525 ms  5.165 ms
 3  be12488.ccr42.lon13.atlas.cogentco.com (130.117.51.41)  74.130 ms be2249.ccr41.ams03.atlas.cogentco.com (154.54.36.213)  79.513 ms be12488.ccr42.lon13.atlas.cogentco.com (130.117.51.41)  81.690 ms
 4  be3488.ccr52.lhr01.atlas.cogentco.com (154.54.60.14)  85.445 ms  84.956 ms be2490.ccr42.jfk02.atlas.cogentco.com (154.54.42.85)  81.930 ms
 5  be3472.ccr32.bos01.atlas.cogentco.com (154.54.46.33)  81.726 ms be2317.ccr41.jfk02.atlas.cogentco.com (154.54.30.185)  81.607 ms  81.097 ms
 6  be3471.ccr31.bos01.atlas.cogentco.com (154.54.40.153)  81.553 ms endurance-international-group.demarc.cogentco.com (38.97.106.34)  80.119 ms  78.081 ms
 7  26.149.96.66.static.eigbox.net (66.96.149.26)  73.924 ms  74.752 ms 238.252.148.207.static.yourhostingaccount.com (207.148.252.238)  80.449 ms

 

 

Premium:

 

 

traceroute to 66.96.149.26 (66.96.149.26), 30 hops max, 60 byte packets
 1  * * *
 2  66.249.94.116 (66.249.94.116)  11.014 ms 66.249.94.118 (66.249.94.118)  5.091 ms  4.830 ms
 3  ae22-0-xcr1.att.cw.net (195.89.96.253)  4.934 ms * *
 4  ae0-pcr1.aet.cw.net (195.2.20.73)  5.238 ms * if-be-41-2.ecore1.av2-amsterdam.as6453.net (195.219.194.27)  17.585 ms
 5  80.231.85.133 (80.231.85.133)  5.049 ms  5.044 ms ae41-tcr1.adr.cw.net (195.2.20.86)  5.329 ms
 6  * * ae-2.r20.amstnl07.nl.bb.gin.ntt.net (129.250.7.64)  5.252 ms
 7  ae-15.r20.londen12.uk.bb.gin.ntt.net (129.250.5.1)  10.305 ms be3500.ccr42.ams03.atlas.cogentco.com (154.54.60.25)  8.432 ms ae-15.r20.londen12.uk.bb.gin.ntt.net (129.250.5.1)  10.293 ms
 8  ae-7.r20.nwrknj03.us.bb.gin.ntt.net (129.250.6.147)  76.732 ms be2249.ccr41.ams03.atlas.cogentco.com (154.54.36.213)  85.331 ms ae-7.r20.nwrknj03.us.bb.gin.ntt.net (129.250.6.147)  77.613 ms
 9  be2101.ccr32.bos01.atlas.cogentco.com (154.54.82.38)  79.239 ms be2490.ccr42.jfk02.atlas.cogentco.com (154.54.42.85)  85.124 ms  85.829 ms
10  ae-0.newfold-digital.bstnma07.us.bb.gin.ntt.net (129.250.193.78)  88.075 ms  87.907 ms be2491.ccr22.lpl01.atlas.cogentco.com (154.54.39.117)  89.348 ms
11  250.252.148.207.static.yourhostingaccount.com (207.148.252.250)  87.647 ms be3043.ccr22.ymq01.atlas.cogentco.com (154.54.44.166)  90.016 ms 238.252.148.207.static.yourhostingaccount.com (207.148.252.238)  79.191 ms
12  26.149.96.66.static.eigbox.net (66.96.149.26)  78.941 ms  88.098 ms  83.281 ms

 

 

 

1 2 439
2 REPLIES 2

Hi @Felix123 ,

The Premium Tier's path and route used its preferred PoP then Amsterdam's ISP causing additional hops, unlike in Standard Tier that directly used the network of Amsterdam's ISP.

You can refer to this documentation how routing path for Premium Tier works.

@Marvin_Lucero 

The image there claims that it should pop close to destination. Unless NY changed the name back, then this one is closer to the source.

 

So, imho, question still stands.