How much data is in Synapse and how much is it used?
How much data is there?
Synapse has a object called a “file handle”, a low-level object which references a cloud storage bucket and key and which has basic metadata like file name and size. Conceptually the data “footprint” of Synapse is the sum of the sizes of all the files which Synapse indexes as file handles. The Athena filesnapshots tables lists all the file handles in Synapse and should be useful for computing count and aggregate size statistics. However there are duplicate records to be addressed. First, "snapshots" are taken periodically so we want to eliminate this by taking just the latest snapshot of each record. There can also be multiple file IDs for a <bucket, key> pair, which we can address by taking just the latest ID. The following query in Athena asks how much file data is indexed in Synapse:
with latest as (
select bucket, key, max(snapshot_timestamp) as snapshot_timestamp
from filesnapshots
where snapshot_timestamp > current_timestamp - INTERVAL '14' DAY
group by bucket, key, id, content_size, status
),
max_id as (
select bucket, key, max(id) as id
from filesnapshots
where snapshot_timestamp > current_timestamp - INTERVAL '14' DAY
group by bucket, key
)
select sum(fs.content_size)/1099511627776 total_size_tebibytes
from filesnapshots fs join latest on fs.snapshot_timestamp=latest.snapshot_timestamp and fs.bucket=latest.bucket and fs.key=latest.key
join max_id on fs.id=max_id.id and fs.bucket=max_id.bucket and fs.key=max_id.key
where fs.snapshot_timestamp > current_timestamp - INTERVAL '14' DAY
The result is:
2793
(2.7 petabytes)
Note that this total includes files that have been ‘garbage collected’ and moved to glacier. We can drill down on this by looking at the ‘status’ column, the values of which are, AVAILABLE, ARCHIVED, UNLINKED. The values are described here. The files which are actually accessible through Synapse are AVAILABLE, while the others are in various stages of garbage collection:
with latest as (
select bucket, key, max(snapshot_timestamp) as snapshot_timestamp
from filesnapshots
where snapshot_timestamp > current_timestamp - INTERVAL '14' DAY
group by bucket, key, id, content_size, status
),
max_id as (
select bucket, key, max(id) as id
from filesnapshots
where snapshot_timestamp > current_timestamp - INTERVAL '14' DAY
group by bucket, key
)
select sum(fs.content_size)/1099511627776 size_tebibytes, status
from filesnapshots fs join latest on fs.snapshot_timestamp=latest.snapshot_timestamp and fs.bucket=latest.bucket and fs.key=latest.key
join max_id on fs.id=max_id.id and fs.bucket=max_id.bucket and fs.key=max_id.key
where fs.snapshot_timestamp > current_timestamp - INTERVAL '14' DAY
group by fs.status
This returns:
size_tebibytes status
1 2271 AVAILABLE
2 276 UNLINKED
3 236 ARCHIVED
Or 2.2 petabytes that are “available”.
Synapse is a federated data store. The 2+ PB of available data are not all in one bucket owned by Sage. Rather, users can provision and link their own AWS S3 or Google Bucket, then store objects there, indexed in Synapse. They can even create “external” files, for which Synapse simply stores the files' URLs. Let’s divide up the data by bucket, listing just the 25 largest buckets:
with latest as (
select bucket, key, max(snapshot_timestamp) as snapshot_timestamp
from filesnapshots
where snapshot_timestamp > current_timestamp - INTERVAL '14' DAY
group by bucket, key, id, content_size
),
max_id as (
select bucket, key, max(id) as id
from filesnapshots
where snapshot_timestamp > current_timestamp - INTERVAL '14' DAY
group by bucket, key
)
select fs.bucket, sum(fs.content_size)/1099511627776 size_tebibytes
from filesnapshots fs join latest on fs.snapshot_timestamp=latest.snapshot_timestamp and fs.bucket=latest.bucket and fs.key=latest.key
join max_id on fs.id=max_id.id and fs.bucket=max_id.bucket and fs.key=max_id.key
where
fs.snapshot_timestamp > current_timestamp - INTERVAL '14' DAY
and status='AVAILABLE'
group by fs.bucket
order by size_tebibytes desc
limit 25bucket size_tebibytes
1 proddata.sagebase.org 698
2 ad-knowledge-portal-main 324
3 ad-knowledge-portal-large 236
4 amp-mayo-sinai-synapseencryptedexternalbucket-1bmvn8rlwixv2 163
5 nda-bsmn-scratch 152
6 htan-dcc-htapp 85
7 exceptional-longevity 66
8 diverse-cohorts 52
9 mpnstwgs 45
10 nf-syn23664726-s3-bucket-n9uakf7bowwd 43
11 integrated-immunology 41
12 gates-ki-001-synapseencryptedexternalbucket-kl8vu7tqto1b 38
13 sc-237179673806-pp-hg7rd6vvvd7mk-s3bucket-eu11pjwpxxiv 37
14 htan-dcc-chop 36
15 htan-dcc-stanford 33
16 strides-ampad-project-tower-bucket 24
17 htan-dcc-hms 20
18 htan-dcc-vanderbilt 18
19 htan-dcc-duke 15
20 ntap-add5-project-tower-bucket 12
21 ctf-swnts-project-tower-bucket 11
22 htan-dcc-dfci 8
23 htan-dcc-washu 8
24 sc-237179673806-pp-pisxjdspvwrtg-s3bucket-deoh9m8xdfcc 8
25 nda-bsmn 7
The aggregate size of these 25 buckets is 2180 TiB, or 96% of the total.
proddata.sagebase.org is the primary/default Synapse bucket. The next two large buckets, ad-knowledge-portal-main and ad-knowledge-portal-large are in Sage’s AWS STRIDES account, where the costs are paid directly by NIH. All the buckets are recognizable by name as being related to a Sage project and residing in a Sage AWS account. We can look at storage from the S3 perspective as a rough validation of these figures: Our total S3 footprint should be larger than our Synapse footprint.
We can use S3 Data Lens to get a summary of our storage usage from the AWS perspective. As of June 22, 2023:
AWS Account(s) | S3 usage, TiB |
|---|---|
Sage Organization (~35 accounts) | 1953 |
*6185: Strides AD | 823.3 |
*5034: (Nextflow Tower) | 230.7 |
Total | 3007 |
Since 3007 PB > 2783 PB, the values seem to be correct. Remaining 200 TiB may be accounted for by data which Sage owns but which is unrelated to Synapse. Below we reconcile on a per-bucket basis. There are two numerical columns for bucket size reported by S3 in Sage’s “organization” and STRIDES, respectively, and one column for the amount indexed in Synapse.
Bucket | S3 bucket size (TiB) | STRIDES S3 bucket size (TiB) | Indexed in Synapse (“AVAILABLE”) (TiB) |
937.3 |
| 698 | |
ad-knowledge-portal-main |
| 333.1 | 324 |
ad-knowledge-portal-large |
| 236.1 | 236 |
amp-mayo-sinai-synapseencryptedexternalbucket-1bmvn8rlwixv2 | 312 |
| 163 |
nda-bsmn-scratch |
|
| 152 |
htan-dcc-htapp | 0 (bucket is gone) |
| 85 (also 62 TiB “UNLINKED”) |
exceptional-longevity |
| 66.2 | 66 |
diverse-cohorts |
| 54.9 | 52 |
mpnstwgs | 45 |
| 45 |
nf-syn23664726-s3-bucket-n9uakf7bowwd | 45.4 |
| 43 |
integrated-immunology | 41.8 |
| 41 |
gates-ki-001-synapseencryptedexternalbucket-kl8vu7tqto1b | 39.2 |
| 38 |
sc-237179673806-pp-hg7rd6vvvd7mk-s3bucket-eu11pjwpxxiv | 37.7 |
| 37 |
htan-dcc-chop | 35.2 |
| 36 |
htan-dcc-stanford | 26.9 |
| 33 |
strides-ampad-project-tower-bucket |
| 187.8 | 24 |
htan-dcc-hms | 16.1 |
| 20 |
htan-dcc-vanderbilt | 19.2 |
| 18 |
htan-dcc-duke | 15.8 |
| 15 |
ntap-add5-project-tower-bucket | 16.3 |
| 12 |
ctf-swnts-project-tower-bucket | 11.9 |
| 11 |
htan-dcc-dfci | 0 |
| 8 |
htan-dcc-washu | 0 |
| 8 |
sc-237179673806-pp-pisxjdspvwrtg-s3bucket-deoh9m8xdfcc | 6.8 |
| 8 |
nda-bsmn |
|
| 7 |
The five buckets highlighted in red, appear in Synapse but not in our list of top S3 buckets. They have been emptied or removed by their owners, leaving the index in Synapse ‘stale'. The the child page on stale buckets for more details.
While relating content size to bucket name gives some insight, it does not explain the usage of the large, default Synapse bucket. We can investigate how this bucket is used, and get more insight into how other data usage, by relating content size to Synapse projects. We do this by joining the filesnapshots table, queried above, with the nodesnapshots table, which has project id and which relates to filesnapshots through the file_handle_id column. Lets first query the sum total size of data across projects:
The result is 2108. This is close to the result of the earlier query on filesnapshots alone. (TODO: Where did the other 163 TiB go?) Now we partition the result by Project and Bucket:
The top 50 account for 86% of the 2108 TiB:
Project | Bucket | Project size, TiB |
ad-knowledge-portal-main | 325 | |
249 | ||
amp-mayo-sinai-synapseencryptedexternalbucket-1bmvn8rlwixv2 | 154 | |
nda-bsmn-scratch | 150 | |
ad-knowledge-portal-large | 144 | |
htan-dcc-htapp | 85 | |
exceptional-longevity | 66 | |
mpnstwgs | 45 | |
nf-syn23664726-s3-bucket-n9uakf7bowwd | 43 | |
integrated-immunology | 39 | |
sc-237179673806-pp-hg7rd6vvvd7mk-s3bucket-eu11pjwpxxiv | 37 | |
htan-dcc-chop | 33 | |
27 | ||
diverse-cohorts | 27 | |
htan-dcc-stanford | 25 | |
strides-ampad-project-tower-bucket | 24 | |
diverse-cohorts | 23 | |
21 | ||
18 | ||
htan-dcc-vanderbilt | 17 | |
17 | ||
16 | ||
15 | ||
14 | ||
htan-dcc-hms | 14 | |
htan-dcc-duke | 14 | |
10 | ||
10 | ||
gates-ki-001-synapseencryptedexternalbucket-kl8vu7tqto1b | 9 | |
gates-ki-001-synapseencryptedexternalbucket-kl8vu7tqto1b | 9 | |
sc-237179673806-pp-pisxjdspvwrtg-s3bucket-deoh9m8xdfcc | 8 | |
ctf-swnts-project-tower-bucket | 8 | |
htan-dcc-washu | 8 | |
8 | ||
gates-ki-001-synapseencryptedexternalbucket-kl8vu7tqto1b | 7 | |
7 | ||
nda-bsmn | 7 | |
7 | ||
6 | ||
6 | ||
synodos.eu.frankfurt.penn | 6 | |
ntap-add5-project-tower-bucket | 6 | |
6 | ||
htan-dcc-dfci | 6 | |
ntap-cnf-cell-project-tower-bucket | 5 | |
5 | ||
5 | ||
htan-dcc-ohsu | 5 | |
htan-dcc-tnp-sardana | 5 | |
4 |
This query breaks down the data by is_public, is_controlled, is_restricted:
Result:
is_public | is_controlled | is_restricted | size_tebibytes |
FALSE | FALSE | FALSE | 827 |
TRUE | TRUE | TRUE | 615 |
TRUE | FALSE | FALSE | 196 |
TRUE | TRUE | FALSE | 155 |
FALSE | TRUE | FALSE | 124 |
FALSE | FALSE | TRUE | 73 |
TRUE | FALSE | TRUE | 63 |
FALSE | TRUE | TRUE | 53
|
TODO: What are the exact definitions of these terms?
Let’s repeat the project breakdown query but look only at “public” data:
The total drops to 948 TiB , dominated by AD Portal and PsychEncode, as before. If we restrict the query to the Synapse ‘default’ bucket , we find 279 TiB of data. To see how it breaks down by project:
The top 21 projects account for 83% of the data, with PsychEncode at the top of the list:
_col0 | bucket | project_size_tebibytes |
152 | ||
21 | ||
19 | ||
6 | ||
5 | ||
4 | ||
4 | ||
3 | ||
3 | ||
2 | ||
2 | ||
1 | ||
1 | ||
1 | ||
1 | ||
1 | ||
1 | ||
1 | ||
1 | ||
1 | ||
1 |
How much is the data used?
We can use the filedownloadrecords table to see how much each Synapse project is being accessed.
Download volume per month for the previous half year:
returns
Month | Download volume, GB |
6 | 2031037 |
7 | 4553643 |
At the time of this writing we have not yet accumulated a full month’s worth of data. Let’s compare the current month’s egress to AWS’s stat’s: For July 1-12, these are the egress charges from AWS Cost Explorer:
| Cost | Data out, GB |
|---|---|---|
Sage Organization | $2400 | 29,800 |
AD STRIDES | $6500 | 94,670 |
NF STRIDES | ~$0 | ~0 |
The numbers are far less than those reported by Synapse. This could be because the data did not leave the region or because the presigned URL received from Synapse (the creation of which triggered the recording of a download) was not actually used by the client.
Let’s break up the download by project as well as bucket:
15 projects accounted for 99% of the download activity. Only 6 of these projects use the Synapse default bucket:
Project | Bucket | Download volume, GB |
htan-dcc-stanford | 1804241 | |
ad-knowledge-portal-main | 846984 | |
diverse-cohorts | 530371 | |
mpnstwgs | 325521 | |
233216 | ||
htan-dcc-vanderbilt | 146573 | |
htan-dcc-duke | 132917 | |
126349 | ||
htan-dcc-ohsu | 73147 | |
ecmonsen-emorypipeline-lambda-055273631518-us-east-1 | 61412 | |
51600 | ||
sc-237179673806-pp-hg7rd6vvvd7mk-s3bucket-eu11pjwpxxiv | 50989 | |
50526 | ||
31063 | ||
26497 |
This analysis could further be refined to look just at down load of public data.