Accuracy of the geolite_city_bq_b2 dataset - geolocation

Question

I believe there are inaccuracies in the BigQuery fh-bigquery.geocode.geolite_city_bq_b2 dataset, and am curious if others have noticed this too.
Background: I have the BigQuery code from Ramtin M. Seraj running, and his/my logic appear to be sound. However there are IP addresses known to represent certain places, e.g. Tokyo # 150.249.199.17, but which are indicated by Ramtin's query to be in Rochester NY-USA or Ottawa ON-CA. If the query logic is sound then the only conclusion is that the underlying Geolite dataset is not.
To verify, look at results of this query:
SELECT *
FROM `fh-bigquery.geocode.geolite_city_bq_b2b`
WHERE classB = 38649
Note from these results that startIp = 150.245.0.0 and endIp = 150.249.255.255, therefore address 150.249.199.17 is within this IP range.
Now compare with the results from https://ipinfo.io/150.249.199.17, and also with the results from the following BigQuery. Notice that all computed values, such as IPV4_TO_INT64() of the IP address, fall within the ranges returned by the query above.
SELECT '150.249.199.17' as ipAddress
, NET.IPV4_TO_INT64(NET.IP_FROM_STRING('150.249.199.17')) AS clientIpNum_int
, TRUNC(NET.IPV4_TO_INT64(NET.IP_FROM_STRING('150.249.199.17'))/(256*256)) AS classB
, CAST(TRUNC(NET.IPV4_TO_INT64(NET.IP_FROM_STRING('150.249.199.17'))/(256*256)) as INT64) as client_classB_int
p.s. I would upvote the first answer, or add a comment, but I don't have enough Reputons yet!

Answer 1

2019, much improved answer:
https://medium.com/#hoffa/geolocation-with-bigquery-de-identify-76-million-ip-addresses-in-20-seconds-e9e652480bd2
#standardSQL
# replace with your source of IP addresses
# here I'm using the same Wikipedia set from the previous article
WITH source_of_ip_addresses AS (
SELECT REGEXP_REPLACE(contributor_ip, 'xxx', '0') ip, COUNT(*) c
FROM `publicdata.samples.wikipedia`
WHERE contributor_ip IS NOT null
GROUP BY 1
)
SELECT country_name, SUM(c) c
FROM (
SELECT ip, country_name, c
FROM (
SELECT *, NET.SAFE_IP_FROM_STRING(ip) & NET.IP_NET_MASK(4, mask) network_bin
FROM source_of_ip_addresses, UNNEST(GENERATE_ARRAY(9,32)) mask
WHERE BYTE_LENGTH(NET.SAFE_IP_FROM_STRING(ip)) = 4
)
JOIN `fh-bigquery.geocode.201806_geolite2_city_ipv4_locs`
USING (network_bin, mask)
)
GROUP BY 1
ORDER BY 2 DESC
I'm about to publish a much improved version of Geolite in BigQuery. Stay tuned to https://twitter.com/felipehoffa and https://medium.com/#hoffa. And I'll update this answer then too.
With that said, to answer the accuracy part which titles this question, Maxmind says:
GeoLite2 databases are free IP geolocation databases comparable to, but less accurate than, MaxMind’s GeoIP2 databases
https://dev.maxmind.com/geoip/geoip2/geolite2/