Data Domain: PCR post-comp for a MTree is larger than DD file system active tier size
Summary: This KB describe a symptom that DD PCR reports post-comp for a MTree that is larger than DD file system active tier size.
Symptoms
DD Physical capacity measurement/reporting (PCM/PCR) for a MTree report post-comp size is larger than the DD file system active tier total used size.
The latest PCR report shows that the MTree located at /data/col1/avamar-123456789 has a post-comp size of 253112.8 GiB. However, the total post-comp size used by the DDFS /data partition (active tier) is only 241913.2 GiB.
sysadmin@dd123# compression physical-capacity-measurement sample show history mtrees /data/col1/avamar-123456789
MTree: /data/col1/avamar-123456789
Measurement Time Logical Used Physical Used Global-Comp Local-Comp Total-Comp
(Pre-Comp) (Post-Comp) Factor Factor Factor
(GiB) (GiB) (Reduction %)
------------------- ------------ ------------- ----------- ---------- ----------------
……
2021/02/27 10:00:03 62238106. 253112.8 91.31x 2.69x 245.89x (99.59%)
------------------- ------------ ------------- ----------- ---------- ----------------
DDFS space:
sysadmin@dd123# filesys show space
GENERATED: 2021-02-27 10:45:24 PDT
Active Tier:
Resource Size GiB Used GiB Avail GiB Use% Cleanable GiB*
---------------- -------- ----------- --------- ---- --------------
/data: pre-comp - 113255640.0 - - -
/data: post-comp 342785.4 241913.2 100872.2 71% 8030.7
/ddvar 47.1 16.5 28.2 37% -
/ddvar/core 984.2 0.1 934.1 0% -
---------------- -------- ----------- --------- ---- --------------
* Estimated based on last cleaning of 2021/02/25 09:47:35.
Cause
The "Physical Used (Post-Comp)" for the given MTree measured by PCR is Metadata + Physical L0 bytes. For the Metadata part, PCR reports logical LP segment size. In some workloads, such as virtual sythetic (VS) workload, LP sharing exists. As a result, the logical LP segment size can be larger than the DDFS /data partition physical size. This issue can be seen when the MTree data has a high deduplication ratio as it has high LP metadata.
From ddfs.inf0 logs, PCR job statistics
ddfs.inf0:02/16 18:21:56.176 (tid 0x7f9928c04c20): pcr_run_job: Job (1,6) has results: logical 66827658273324768, compression 2.693984, metadata 218970311166156, num lps 16042377403, num l0s 7793825601792, l0 sum 66827658273324768, bits set 15696840, num sampled 7577743715, estimated unique num l0s with sampling 16109240, estimated unique num l0s 16495861760, bf size 38654704, files seen 38738195, paths not found 0, average l0 size 8624.134370, sampled pre lc sum 138928250366, sample post lc sum 51569813096, sampled uniq l0 count 8389415, Physical L0 bytes = (estimated unique num l0s * average l0 size ) / compression = 16495861760 * 8624.134370 / 2.693984 = 52,806,665,450,960.36 byte LP metadata = 218,970,311,166,156 byte PCR “Physical Used (Post-Comp)” = 52,806,665,450,960.36 + 218,970,311,166,156 = ~253.1 TB.
Resolution
There is no resolution or fix for this symptom. This is how the PCR was designed and works as designed.